Learning unnormalized statistical models (e.g., energy-based models) is computationally challenging due to the complexity of handling the partition function. To eschew this complexity, noise-contrastive estimation (NCE) has been proposed by formulating the objective as the logistic loss between the real data and the artificial noise. However, previous research indicates that NCE may perform poorly in many tasks due to its flat loss landscape and slow convergence. In this paper, we study a direct approach for optimizing the negative log-likelihood of unnormalized models through the lens of compositional optimization. To tackle the partition function, a noise distribution is introduced such that the log partition function can be expressed as a compositional function whose inner function can be estimated using stochastic samples. Consequently, the objective can be optimized via stochastic compositional optimization algorithms. Despite being a simple method, we demonstrate it is more favorable than NCE by (1) establishing a fast convergence rate and quantifying its dependence on the noise distribution through the variance of stochastic estimators; (2) developing better results in Gaussian mean estimation by showing our method has a much favorable loss landscape and enjoys faster convergence; (3) demonstrating better performance on various applications, including density estimation, out-of-distribution detection, and real image generation.

This work addresses a key challenge in the effective adoption of Recurrent Neural Networks (RNNs) by reducing inference time and expanding the scope of a prediction. It introduces compressed learning, a novel approach that integrates Context-Free Grammar (CFG) and online tokenization into the training and inference of RNNs for streaming inputs. Through a hierarchical compression algorithm, it compresses an input sequence to a CFG and makes predictions based on the compressed sequence. Its algorithm design employs a set of techniques to overcome the issues from the myopic nature of online tokenization, the tension between inference accuracy and compression rate, and other complexities in sequence compression and prediction. Its effectiveness is theoretically analyzed and empirically validated on 16 real-world sequences, including program function calls, memory traces, and system logs. Empirical results demonstrate that compressed learning can successfully recognize and leverage repetitive patterns in input sequences, and effectively translate them into dramatic (1–1,762 \(\times\) ) inference speedups as well as much (1–7,830 \(\times\) ) expanded prediction scope, while keeping the inference accuracy satisfactory.

Cross-Domain Recommendation (CDR) leverages additional knowledge from auxiliary domains to address the long-standing data sparsity issue. However, existing methods typically acquire this knowledge by minimizing the average loss over all domains, overlooking the fact that different domains possess different user-preference distributions. As a result, the acquired knowledge may contain biased information from data-rich domains, leading to performance degradation in data-scarce domains. In this paper, we propose a novel CDR method, which takes domain distinctions into consideration to extract and adapt unbiased information. Specifically, our method consists of two key components: Unbiased Information Extraction (UIE) and Unbiased Information Adaptation (UIA). In the UIE, inspired by distributionally robust optimization, we optimize the worst-case performance across all domains to extract domain-invariant information, preventing the potential bias from auxiliary domains. In the UIA, we introduce a new user-item attention module, which employs domain-specific information from historically interacted items to attend the adaptation of domain-invariant information. To verify the effectiveness of our method, we conduct extensive experiments on three real-world datasets, each of which contains three extremely sparse domains. Experimental results demonstrate the considerable superiority of our proposed method compared to baselines.

We investigate constrained online convex optimization, in which decisions must belong to a fixed and typically complicated domain, and are required to approximately satisfy additional time-varying constraints over the long term. In this setting, the commonly used projection operations are often computationally expensive or even intractable. To avoid the time-consuming operation, several projection-free methods have been proposed with an O(T^¾ (log T)^½) regret bound and an O(T^⅞) cumulative constraint violation (CCV) bound for general convex losses. In this paper, we improve this result and further establish novel regret and CCV bounds when loss functions are strongly convex. The primary idea is to first construct a composite surrogate loss, involving the original loss and constraint functions, by utilizing the Lyapunov-based technique. Then, we propose a parameter-free variant of the classical projection-free method, namely online Frank-Wolfe (OFW), and run this new extension over the online-generated surrogate loss. Theoretically, for general convex losses, we achieve an O(T^¾) regret bound and an O(T^¾ log T) CCV bound, both of which are order-wise tighter than existing results. For strongly convex losses, we establish new guarantees of an O(T^⅔) regret bound and an O(T^⅚) CCV bound. Moreover, we also extend our methods to a more challenging setting with bandit feedback, obtaining similar theoretical findings. Empirically, experiments on real-world datasets have demonstrated the effectiveness of our methods.

In this paper, we investigate the acceleration of adaptive subgradient methods through frequent directions (FD), a widely-used matrix sketching technique. The state-of-the-art regret bound exhibits a _linear_ dependence on the dimensionality $d$, leading to unsatisfactory guarantees for high-dimensional problems. Additionally, it suffers from an $O(\tau^2 d)$ time complexity per round, which scales quadratically with the sketching size $\tau$. To overcome these issues, we first propose an algorithm named FTSL, achieving a tighter regret bound that is independent of the dimensionality. The key idea is to integrate FD with adaptive subgradient methods under _the primal-dual framework_ and add the cumulative discarded information of FD back. To reduce its time complexity, we further utilize fast FD to expedite FTSL, yielding a better complexity of $O(\tau d)$ while maintaining the same regret bound. Moreover, to mitigate the computational cost for optimization problems involving matrix variables (e.g., training neural networks), we adapt FD to Shampoo, a popular optimization algorithm that accounts for the structure of decision, and give a novel analysis under _the primal-dual framework_. Our proposed method obtains an improved dimension-free regret bound. Experimental results have verified the efficiency and effectiveness of our approaches.

In many real applications, the data attributes are incremental and the samples are stored with accumulated feature spaces gradually. Although there are several elegant approaches to tackling this problem, the theoretical analysis is still limited. There exist at least two challenges and fundamental questions. 1) How to derive the generalization bounds of these approaches? 2) Under what conditions do these approaches have a strong generalization guarantee? To solve these crucial but rarely studied problems, we provide a comprehensive theoretical analysis in this paper. We begin by summarizing and refining four strategies for addressing feature incremental data. Subsequently, we derive their generalization bounds, providing rigorous and quantitative insights. The theoretical findings highlight the key factors influencing the generalization abilities of different strategies. In tackling the above two fundamental problems, we also provide valuable guidance for exploring other learning challenges in dynamic environments. Finally, the comprehensive experimental and theoretical results mutually validate each other, underscoring the reliability of our conclusions.

Label shift is a prevalent phenomenon encountered in open environments, characterized by a notable discrepancy in the label distributions between the source (training) and target (test) domains, whereas the conditional distributions given the labels remain invariant. Existing label shift methods adopt a two-step strategy: initially computing the importance weight and subsequently utilizing it to calibrate the target outputs. However, this conventional strategy overlooks the intricate interplay between output adjustment and weight estimation. In this paper, we introduce a novel approach termed as One-step Label Shift Adaptation (OLSA). Our methodology jointly learns the predictive model and the corresponding weights through a bi-level optimization framework, with the objective of minimizing an upper bound on the target risk. To enhance the robustness of our proposed model, we incorporate a debiasing term into the upper-level classifier training and devise a regularization term for the lower-level weight estimation. Furthermore, we present theoretical analyses about the generalization bounds, offering guarantees for the model's performance. Extensive experimental results substantiate the efficacy of our proposal.

Continual learning aims to learn multiple tasks sequentially while preserving prior knowledge, but faces the challenge of catastrophic forgetting when adapting to new tasks. Recently, approaches leveraging pre-trained models have gained increasing popularity in mitigating this issue, due to the strong generalization ability of foundation models. To adjust pre-trained models for new tasks, existing methods usually employ low-rank adaptation, which restricts parameter updates to a fixed low-rank subspace. However, constraining the optimization space inherently compromises the model's learning capacity, resulting in inferior performance. To address this limitation, we propose Continuous Subspace Optimization for Continual Learning (CoSO) to fine-tune the model in a series of subspaces rather than a single one. These sequential subspaces are dynamically determined through the singular value decomposition of the gradients. CoSO updates the model by projecting gradients onto these subspaces, ensuring memory-efficient optimization. To mitigate forgetting, the optimization subspace of each task is constrained to be orthogonal to the historical task subspace. During task learning, CoSO maintains a task-specific component that captures the critical update directions for the current task. Upon completing a task, this component is used to update the historical task subspace, laying the groundwork for subsequent learning. Extensive experiments on multiple datasets demonstrate that CoSO significantly outperforms state-of-the-art methods, especially in challenging scenarios with long task sequences.

Self-play fine-tuning has demonstrated promising abilities in adapting large language models (LLMs) to downstream tasks with limited real-world data. The basic principle is to iteratively refine the model with real samples and synthetic ones generated from itself. However, the existing methods primarily focus on the relative gaps between the rewards for two types of data, neglecting their absolute values. Through theoretical analysis, we identify that the gap-based methods suffer from unstable evolution, due to the potentially degenerated objectives. To address this limitation, we introduce a novel self-play fine-tuning method, namely \underline{S}elf-\underline{P}l\underline{A}y via Noise \underline{C}ontrastive \underline{E}stimation (SPACE), which leverages noise contrastive estimation to capture the real-world data distribution. Specifically, SPACE treats synthetic samples as auxiliary components, and discriminates them from the real ones in a binary classification manner. As a result, SPACE independently optimizes the absolute reward values for each type of data, ensuring a consistently meaningful objective and thereby avoiding the instability issue. Theoretically, we show that the optimal solution of the objective in SPACE aligns with the underlying distribution of real-world data, and SPACE guarantees a provably stable convergence to the optimal distribution. Empirically, we show that SPACE significantly improves the performance of LLMs over various tasks, and outperforms supervised fine-tuning that employs much more real-world samples. Compared to gap-based self-play fine-tuning methods, SPACE exhibits remarkable superiority and stable evolution.

Recently, self-play fine-tuning (SPIN) has been proposed to adapt large language models to downstream applications with scarce expert-annotated data, by iteratively generating synthetic responses from the model itself. However, SPIN is designed to optimize the current reward advantages of annotated responses over synthetic responses at hand, which may gradually vanish during iterations, leading to \textit{unstable optimization}. Moreover, the utilization of reference policy induces a \textit{misalignment} issue between the reward formulation for training and the metric for generation. To address these limitations, we propose a novel \textbf{T}riplet-based \textbf{S}elf-\textbf{P}lay f\textbf{I}ne-tu\textbf{N}ing (TSPIN) method that integrates two key designs. First, beyond current advantages, TSPIN additionally incorporates historical advantages between iteratively generated responses and proto-synthetic responses produced by the initial policy. Even if the current advantages diminish, historical advantages remain effective, stabilizing the overall optimization. Second, TSPIN introduces the entropy constraint into the self-play framework, which is theoretically justified to support reference-free fine-tuning, eliminating the training-generation discrepancy. Empirical results on various tasks demonstrate not only the superior performance of TSPIN over SPIN, but also its stable evolution during iterations. Remarkably, compared to supervised fine-tuning, TSPIN achieves comparable or even better performance with only $25\\%$ samples, highlighting its effectiveness when faced with scarce annotated data.

This paper studies a multiobjective bandit problem under lexicographic ordering, wherein the learner aims to maximize $m$ objectives, each with different levels of importance. First, we introduce the local trade-off, $\lambda_*$, which depicts the trade-off between different objectives. For the case when an upper bound of $\lambda_*$ is known, i.e., $\lambda\geq\lambda_*$, we develop an algorithm that achieves a general regret bound of $\widetilde{O}(\Lambda^i(\lambda)T^{(d_z^i+1)/(d_z^i+2)})$ for the $i$-th objective, where $i\in\{1,2,\ldots,m\}$, $\Lambda^i(\lambda)=1+\lambda+\cdots+\lambda^{i-1}$, $d_z^i$ is the zooming dimension for the $i$-th objective, and $T$ is the time horizon. Next, we provide a matching lower bound for the lexicographic Lipschitz bandit problem, proving that our algorithm is optimal in terms of $\lambda_*$ and $T$. Finally, for the case where $m=2$, we remove the dependence on the knowledge about $\lambda_*$, albeit at the cost of increasing the regret bound to $\widetilde{O}(\Lambda^i(\lambda_*)T^{(3d_z^i+4)/(3d_z^i+6)})$, which remains optimal in terms of $\lambda_*$. Compared to existing work on lexicographic multi-armed bandits, our approach improves the current regret bound of $\widetilde{O}(T^{2/3})$ and extends the number of arms to infinity. Numerical experiments confirm the effectiveness of our algorithms.

Recently, several universal methods have been proposed for online convex optimization, and attain minimax rates for multiple types of convex functions simultaneously. However, they need to design and optimize one surrogate loss for each type of functions, making it difficult to exploit the structure of the problem and utilize existing algorithms. In this paper, we propose a simple strategy for universal online convex optimization, which avoids these limitations. The key idea is to construct a set of experts to process the original online functions, and deploy a meta-algorithm over the linearized losses to aggregate predictions from experts. Specifically, the meta-algorithm is required to yield a second-order bound with excess losses, so that it can leverage strong convexity and exponential concavity to control the meta-regret. In this way, our strategy inherits the theoretical guarantee of any expert designed for strongly convex functions and exponentially concave functions, up to a double logarithmic factor. As a result, we can plug in off-the-shelf online solvers as black-box experts to deliver problem-dependent regret bounds. For general convex functions, it maintains the minimax optimality and also achieves a small-loss bound. Furthermore, we extend our universal strategy to online composite optimization, where the loss function comprises a time-varying function and a fixed regularizer. To deal with the composite loss functions, we employ a meta-algorithm based on the optimistic online learning framework, which not only enjoys a second-order bound, but also can utilize estimations for upcoming loss functions. With suitable configurations, we show that the additional regularizer does not contribute to the meta-regret, thus ensuring the universality in the composite setting.

Non-stationary online learning has drawn much attention in recent years. In particular, dynamic regret and adaptive regret are proposed as two principled performance measures for online convex optimization in non-stationary environments. To optimize them, a two-layer online ensemble is usually deployed due to the inherent uncertainty of non-stationarity, in which multiple base-learners are maintained and a meta-algorithm is employed to track the best one on the fly. However, the two-layer structure raises concerns about computational complexity --- such methods typically maintain $O(\log T)$ base-learners simultaneously for a $T$-round online game and thus perform multiple projections onto the feasible domain per round, which becomes the computational bottleneck when the domain is complicated. In this paper, we present efficient methods for optimizing dynamic regret and adaptive regret that reduce the number of projections per round from $O(\log T)$ to $1$. The proposed algorithms require only one gradient query and one function evaluation at each round. Our technique hinges on the reduction mechanism developed in parameter-free online learning and requires non-trivial modifications for non-stationary online methods. Furthermore, we study an even stronger measure, namely "interval dynamic regret", and reduce the number of projections per round from $O(\log^2 T)$ to $1$ for minimizing it. Our reduction demonstrates broad generality and applies to two important applications: online stochastic control and online principal component analysis, resulting in methods that are both efficient and optimal. Finally, empirical studies verify our theoretical findings.

We investigate decentralized online convex optimization (D-OCO), in which a set of local learners are required to minimize a sequence of global loss functions using only local computations and communications. Previous studies have established $O(n^{5/4}\rho^{-1/2}\sqrt{T})$ and ${O}(n^{3/2}\rho^{-1}\log T)$ regret bounds for convex and strongly convex functions respectively, where $n$ is the number of local learners, $\rho<1$ is the spectral gap of the communication matrix, and $T$ is the time horizon. However, there exist large gaps from the existing lower bounds, i.e., $\Omega(n\sqrt{T})$ for convex functions and $\Omega(n)$ for strongly convex functions. To fill these gaps, in this paper, we first develop a novel D-OCO algorithm that can respectively reduce the regret bounds for convex and strongly convex functions to $\tilde{O}(n\rho^{-1/4}\sqrt{T})$ and $\tilde{O}(n\rho^{-1/2}\log T)$. The primary technique is to design an online accelerated gossip strategy that enjoys a faster average consensus among local learners. Furthermore, by carefully exploiting spectral properties of a specific network topology, we enhance the lower bounds for convex and strongly convex functions to $\Omega(n\rho^{-1/4}\sqrt{T})$ and $\Omega(n\rho^{-1/2}\log T)$, respectively. These results suggest that the regret of our algorithm is nearly optimal in terms of $T$, $n$, and $\rho$ for both convex and strongly convex functions. Finally, we propose a projection-free variant of our algorithm to efficiently handle practical applications with complex constraints. Our analysis reveals that the projection-free variant can achieve ${O}(nT^{3/4})$ and ${O}(nT^{2/3}(\log T)^{1/3})$ regret bounds for convex and strongly convex functions with nearly optimal $\tilde{O}(\rho^{-1/2}\sqrt{T})$ and $\tilde{O}(\rho^{-1/2}T^{1/3}(\log T)^{2/3})$ communication rounds, respectively.

This paper explores stochastic multi-level compositional optimization, where the objective function is a composition of multiple smooth functions. Traditional methods for solving this problem suffer from either sub-optimal sample complexities or require huge batch sizes. To address these limitations, we introduce the Stochastic Multi-level Variance Reduction (SMVR) method. In the expectation case, our SMVR method attains the optimal sample complexity of $\mathcal {O}(1/\epsilon ^{3})$O(1/ε3) to find an $\epsilon$ε-stationary point for non-convex objectives. When the function satisfies convexity or the Polyak-Łojasiewicz (PL) condition, we propose a stage-wise SMVR variant. This variant improves the sample complexity to $\mathcal {O}(1/\epsilon ^{2})$O(1/ε2) for convex functions and $\mathcal {O}(1/(\mu \epsilon ))$O(1/(με)) for functions meeting the $\mu$μ-PL condition or $\mu$μ-strong convexity. These complexities match the lower bounds not only in terms of $\epsilon$ε but also in terms of $\mu$μ (for PL or strongly convex functions), without relying on large batch sizes in each iteration. Furthermore, in the finite-sum case, we develop the SMVR-FS algorithm, which can achieve a complexity of $\mathcal {O}(\sqrt{n}/\epsilon ^{2})$O(n/ε2) for non-convex objectives, $\mathcal {O}(\sqrt{n}/\epsilon \log (1/\epsilon ))$O(n/εlog(1/ε)) for convex functions and $\mathcal {O}(\sqrt{n}/\mu \log (1/\epsilon ))$O(n/μlog(1/ε)) for objectives satisfying the $\mu$μ-PL condition, where $n$n denotes the number of functions in each level. To make use of adaptive learning rates, we propose the Adaptive SMVR method, which maintains the same complexities while demonstrating faster convergence in practice.

Most existing studies on continual learning (CL) consider the task-based setting, where task boundaries are known to learners during training. However, they may be impractical for real-world problems, where new tasks arrive with unnotified distribution shifts. In this article, we introduce a new boundary-unknown continual learning scenario called continuum incremental learning (CoIL), where the incremental unit may be a concatenation of several tasks or a subset of one task. To identify task boundaries, we design a continual out-of-distribution (OOD) detection method based on softmax probabilities, which can detect OOD samples for the latest learned task. Then, we incorporate it with continual learning approaches to solve the CoIL problem. Furthermore, we investigate the more challenging task-reappear setting and propose a method named continual learning with unknown task boundary (CLUTaB). CLUTaB first adopts in-distribution detection and OOD loss to determine whether a set of data is sampled from any learned distribution. Then, a two-step inference technique is designed to improve the continual learning performance. Experiments show that our methods work well with existing continual learning approaches and achieve good performance on CIFAR-100 and mini-ImageNet datasets.

Projection-free online learning has drawn increasing interest due to its efficiency in solving high-dimensional problems with complicated constraints. However, most existing projection-free online methods focus on minimizing the static regret, which unfortunately fails to capture the challenge of changing environments. In this paper, we investigate non-stationary projection-free online learning, and choose dynamic regret and adaptive regret to measure the performance. Specifically, we first provide a novel dynamic regret analysis for an existing projection-free method named BOGD_IP, and establish an O(T^¾ (1+P_T)) dynamic regret bound, where P_T denotes the path-length of the comparator sequence. Then, we improve the upper bound to O(T^¾ (1+P_T)^¼) by running multiple BOGD_IP algorithms with different step sizes in parallel, and tracking the best one on the fly. Our results are the first general-case dynamic regret bounds for projection-free online learning, and can recover the existing O(T^¾) static regret by setting P_T = 0. Furthermore, we propose a projection-free method to attain an O(?^¾) adaptive regret bound for any interval with length ?, which nearly matches the static regret over that interval. The essential idea is to maintain a set of BOGD_IP algorithms dynamically, and combine them by a meta algorithm. Moreover, we demonstrate that it is also equipped with an O(T^¾ (1+P_T)^¼) dynamic regret bound. Finally, empirical studies verify our theoretical findings.

We investigate decentralized online convex optimization (D-OCO), in which a set of local learners are required to minimize a sequence of global loss functions using only local computations and communications. Previous studies have established $O(n^{5/4}\rho^{-1/2}\sqrt{T})$ and ${O}(n^{3/2}\rho^{-1}\log T)$ regret bounds for convex and strongly convex functions respectively, where $n$ is the number of local learners, $\rho<1$ is the spectral gap of the communication matrix, and $T$ is the time horizon. However, there exist large gaps from the existing lower bounds, i.e., $\Omega(n\sqrt{T})$ for convex functions and $\Omega(n)$ for strongly convex functions. To fill these gaps, in this paper, we first develop novel D-OCO algorithms that can respectively reduce the regret bounds for convex and strongly convex functions to $\tilde{O}(n\rho^{-1/4}\sqrt{T})$ and $\tilde{O}(n\rho^{-1/2}\log T)$. The primary technique is to design an online accelerated gossip strategy that enjoys a faster average consensus among local learners. Furthermore, by carefully exploiting the spectral properties of a specific network topology, we enhance the lower bounds for convex and strongly convex functions to $\Omega(n\rho^{-1/4}\sqrt{T})$ and $\Omega(n\rho^{-1/2})$, respectively. These lower bounds suggest that our algorithms are nearly optimal in terms of $T$, $n$, and $\rho$.

AI-powered applications often involve multiple deep neural network (DNN)-based prediction tasks to support application-level functionalities. However, executing multi-DNNs can be challenging due to the high resource demands and computation costs that increase linearly with the number of DNNs. Multi-task learning (MTL) addresses this problem by designing a multi-task model that shares parameters across tasks based on a single backbone DNN. This paper explores an alternative approach called model fusion: rather than training a single multi-task model from scratch as MTL does, model fusion fuses multiple task-specific DNNs that are pre-trained separately and can have heterogeneous architectures into a single multi-task model. We materialize model fusion in a software framework called GMorph to accelerate multi-DNN inference while maintaining task accuracy. GMorph features three main technical contributions: graph mutations to fuse multi-DNNs into resource-efficient multi-task models, search-space sampling algorithms, and predictive filtering to reduce the high search costs. Our experiments show that GMorph can outperform MTL baselines and reduce the inference latency of multi-DNNs by 1.1-3× while meeting the target task accuracy.

While traditional distributionally robust optimization (DRO) aims to minimize the maximal risk over a set of distributions, Agarwal & Zhang (2022) recently proposed a variant that replaces risk with *excess risk*. Compared to DRO, the new formulation—minimax excess risk optimization (MERO) has the advantage of suppressing the effect of heterogeneous noise in different distributions. However, the choice of excess risk leads to a very challenging minimax optimization problem, and currently there exists only an inefficient algorithm for empirical MERO. In this paper, we develop efficient stochastic approximation approaches which directly target MERO. Specifically, we leverage techniques from stochastic convex optimization to estimate the minimal risk of every distribution, and solve MERO as a stochastic convex-concave optimization (SCCO) problem with biased gradients. The presence of bias makes existing theoretical guarantees of SCCO inapplicable, and fortunately, we demonstrate that the bias, caused by the estimation error of the minimal risk, is under-control. Thus, MERO can still be optimized with a nearly optimal convergence rate. Moreover, we investigate a practical scenario where the quantity of samples drawn from each distribution may differ, and propose a stochastic approach that delivers *distribution-dependent* convergence rates.

This paper investigates projection-free algorithms for stochastic constrained multi-level optimization. In this context, the objective function is a nested composition of several smooth functions, and the decision set is closed and convex. Existing projection-free algorithms for solving this problem suffer from two limitations: 1) they solely focus on the gradient mapping criterion and fail to match the optimal sample complexities in unconstrained settings; 2) their analysis is exclusively applicable to non-convex functions, without considering convex and strongly convex objectives. To address these issues, we introduce novel projection-free variance reduction algorithms and analyze their complexities under different criteria. For gradient mapping, our complexities improve existing results and match the optimal rates for unconstrained problems. For the widely-used Frank-Wolfe gap criterion, we provide theoretical guarantees that align with those for single-level problems. Additionally, by using a stage-wise adaptation, we further obtain complexities for convex and strongly convex functions. Finally, numerical experiments on different tasks demonstrate the effectiveness of our methods.

Recently, Cutkosky et al. introduce the online-to-non-convex framework, which utilizes online learning methods to solve non-smooth non-convex optimization problems, and achieves an $\mathcal{O}(\epsilon^{-3}\delta^{-1})$ gradient complexity for finding $(\delta,\epsilon)$-stationary points. However, their results rely on the bounded variance assumption of stochastic gradients and only hold in expectation. To address these limitations, we investigate the case that stochastic gradients obey heavy-tailed distributions with finite $\mathfrak{p}$-th moments for some $\mathfrak{p}\in(1,2]$, and propose a novel algorithm which is able to identify a $(\delta,\epsilon)$-stationary point with high probability, after consuming $\tilde{\mathcal{O}}(\epsilon^{-\frac{2\mathfrak{p}-1}{\mathfrak{p}-1}}\delta^{-1})$ stochastic gradients. The key idea is first incorporating the gradient clipping technique into the online-to-non-convex framework to produce a sequence of points, the averaged gradient norms of which is no greater than $\epsilon$. Then, we propose a validation method to select one $(\delta,\epsilon)$-stationary point among the candidates. When gradient distributions have bounded variance, i.e., $\mathfrak{p}=2$, our result turns into $\tilde{\mathcal{O}}(\epsilon^{-3}\delta^{-1})$, which improves the existing $\tilde{\mathcal{O}}(\epsilon^{-4}\delta^{-1})$ high-probability bound. When the objective is smooth, our algorithm can also find an $\epsilon$-stationary point with $\tilde{\mathcal{O}}(\epsilon^{-\frac{3\mathfrak{p}-2}{\mathfrak{p}-1}})$ gradient queries.

The temperature parameter plays a profound role during training and/or inference with large foundation models (LFMs) such as large language models (LLMs) and CLIP models. Particularly, it adjusts the logits in the softmax function in LLMs, which is crucial for next token generation, and it scales the similarities in the contrastive loss for training CLIP models. A significant question remains: “ Is it viable to learn a neural network to predict a personalized temperature of any input data for enhancing LFMs?" In this paper, we present a principled framework for learning a small yet generalizable temperature prediction network (TempNet) to improve LFMs. Our solution is composed of a novel learning framework with robust losses underpinned by constrained distributionally robust optimization (DRO), and a properly designed TempNet with theoretical inspiration. TempNet can be trained together with a large foundation model from scratch or learned separately given a pretrained foundation model. It is not only useful for predicting personalized temperature to promote the training of LFMs but also generalizable and transferable to new tasks. Our experiments on LLMs and CLIP models demonstrate that TempNet greatly improves the performance of existing solutions or models.

Online convex optimization (OCO) with arbitrary delays, in which gradients or other information of functions could be arbitrarily delayed, has received increasing attention recently. Different from previous studies that focus on stationary environments, this paper investigates the delayed OCO in non-stationary environments, and aims to minimize the dynamic regret with respect to any sequence of comparators. To this end, we first propose a simple algorithm, namely DOGD, which performs a gradient descent step for each delayed gradient according to their arrival order. Despite its simplicity, our novel analysis shows that the dynamic regret of DOGD can be automatically bounded by $O(\sqrt{\bar{d}T}(P_T+1))$ under mild assumptions, and $O(\sqrt{dT}(P_T+1))$ in the worst case, where $\bar{d}$ and $d$ denote the average and maximum delay respectively, $T$ is the time horizon, and $P_T$ is the path-length of comparators. Furthermore, we develop an improved algorithm, which reduces those dynamic regret bounds achieved by DOGD to $O(\sqrt{\bar{d}T(P_T+1)})$ and $O(\sqrt{dT(P_T+1)})$, respectively. The key idea is to run multiple DOGD with different learning rates, and utilize a meta-algorithm to track the best one based on their delayed performance. Finally, we demonstrate that our improved algorithm is optimal in the worst case by deriving a matching lower bound.

To deal with changing environments, adaptive regret has been proposed to minimize the regret over every interval. Previous studies have established a small-loss adaptive regret bound for general convex functions under the smoothness condition, offering the advantage of being much tighter than minimax rates for benign problems. However, it remains unclear whether similar bounds are attainable for other types of convex functions, such as exp-concave and strongly convex functions. In this paper, we first propose a novel algorithm that achieves a small-loss adaptive regret bound for exp-concave and smooth function. Subsequently, to address the limitation that existing algorithms can only handle one type of convex functions, we further design a universal algorithm capable of delivering small-loss adaptive regret bounds for general convex, exp-concave, and strongly convex functions simultaneously. That is challenging because the universal algorithm follows the meta-expert framework, and we need to ensure that upper bounds for both meta-regret and expert-regret are of small-loss types. Moreover, we provide a novel analysis demonstrating that our algorithms are also equipped with minimax adaptive regret bounds when functions are non-smooth.

In this paper, we investigate the empirical counterpart of Group Distributionally Robust Optimization (GDRO), which aims to minimize the maximal empirical risk across $m$ distinct groups. We formulate empirical GDRO as a *two-level* finite-sum convex-concave minimax optimization problem and develop an algorithm called ALEG to benefit from its special structure. ALEG is a double-looped stochastic primal-dual algorithm that incorporates variance reduction techniques into a modified mirror prox routine. To exploit the two-level finite-sum structure, we propose a simple group sampling strategy to construct the stochastic gradient with a smaller Lipschitz constant and then perform variance reduction for all groups. Theoretical analysis shows that ALEG achieves $\varepsilon$-accuracy within a computation complexity of $\mathcal{O}\left(\frac{m\sqrt{\bar{n}\ln{m}}}{\varepsilon}\right)$, where $\bar n$ is the average number of samples among $m$ groups. Notably, our approach outperforms the state-of-the-art method by a factor of $\sqrt{m}$. Based on ALEG, we further develop a two-stage optimization algorithm called ALEM to deal with the empirical Minimax Excess Risk Optimization (MERO) problem. The computation complexity of ALEM nearly matches that of ALEG, surpassing the rates of existing methods.

We study online composite optimization under the Stochastically Extended Adversarial (SEA) model. Specifically, each loss function consists of two parts: a fixed non-smooth and convex regularizer, and a time-varying function which can be chosen either stochastically, adversarially, or in a manner that interpolates between the two extremes. In this setting, we show that for smooth and convex time-varying functions, optimistic composite mirror descent (OptCMD) can obtain an $\mathcal{O}(\sqrt{\sigma_{1:T}^2} + \sqrt{\Sigma_{1:T}^2})$ regret bound, where $\sigma_{1:T}^2$ and $\Sigma_{1:T}^2$ denote the cumulative stochastic variance and the cumulative adversarial variation of time-varying functions, respectively. For smooth and strongly convex time-varying functions, we establish an $\mathcal{O}((\sigma_{\max}^2 + \Sigma_{\max}^2)\log(\sigma_{1:T}^2 + \Sigma_{1:T}^2))$ regret bound, where $\sigma_{\max}^2$ and $\Sigma_{\max}^2$ denote the maximal stochastic variance and the maximal adversarial variation, respectively. For smooth and exp-concave time-varying functions, we achieve an $\mathcal{O}(d \log (\sigma_{1:T}^2 + \Sigma_{1:T}^2))$ bound where $d$ denotes the dimensionality. Moreover, to deal with the unknown function type in practical problems, we propose a multi-level \textit{universal} algorithm that is able to achieve the desirable bounds for three types of time-varying functions simultaneously. It should be noticed that all our findings match existing bounds for the SEA model without the regularizer, which implies that there is \textit{no price} in regret bounds for the benefits gained from the regularizer.

Tool-augmented large language models (LLMs) leverage tools, often in the form of APIs, to improve their reasoning capabilities on complex tasks. This enables them to act as intelligent agents interacting with the real world. The recently introduced ToolLLaMA model by Qin et al. [2023] utilizes the depth-first search-based decision tree (DFSDT) mechanism for multi-step reasoning with $16000+$ real-world APIs, effectively enhancing the performance of tool-augmented LLMs compared to traditional chain reasoning mechanisms. However, their approach only employs successful paths from decision trees (also called inference trees) for supervised fine-tuning (SFT), missing out on the potential learning opportunities from failed paths. Inspired by this, we propose an inference trajectory optimization framework based on preference learning to address this limitation. We first introduce a novel method for constructing step-wise preference data from tree-like expert trajectories, which leverages the previously ignored failed explorations in the decision trees. In the subsequent training phase, we first fine-tune the LLM with successful tool-usage expert trajectories and then apply direct preference optimization (DPO) with the preference data to update the LLM's policy, resulting in our ToolPrefer-LLaMA (TP-LLaMA) model. This approach not only enhances the utilization of original expert data but also broadens the learning space of the model. Our experiments demonstrate that by obtaining insights from errors in inference trees, TP-LLaMA significantly outperforms the baselines across almost all test scenarios by a large margin and exhibits better generalization capabilities with unseen APIs. At the same time, TP-LLaMA has also demonstrated superior reasoning efficiency compared to the baselines, making it more suitable for complex tool-usage reasoning tasks.

This paper explores adaptive variance reduction methods for stochastic optimization based on the STORM technique. Existing adaptive extensions of STORM rely on strong assumptions like bounded gradients and bounded function values, or suffer an additional $\mathcal{O}(\log T)$ term in the convergence rate. To address these limitations, we introduce a novel adaptive STORM method that achieves an optimal convergence rate of $\mathcal{O}(T^{-1/3})$ for non-convex functions with our newly designed learning rate strategy. Compared with existing approaches, our method requires weaker assumptions and attains the optimal convergence rate without the additional $\mathcal{O}(\log T)$ term. We also extend the proposed technique to stochastic compositional optimization, obtaining the same optimal rate of $\mathcal{O}(T^{-1/3})$. Furthermore, we investigate the non-convex finite-sum problem and develop another innovative adaptive variance reduction method that achieves an optimal convergence rate of $\mathcal{O}(n^{1/4} T^{-1/2} )$, where $n$ represents the number of component functions. Numerical experiments across various tasks validate the effectiveness of our method.

Sign stochastic gradient descent (signSGD) is a communication-efficient method that transmits only the sign of stochastic gradients for parameter updating. Existing literature has demonstrated that signSGD can achieve a convergence rate of $\mathcal{O}(d^{1/2}T^{-1/4})$, where $d$ represents the dimension and $T$ is the iteration number. In this paper, we improve this convergence rate to $\mathcal{O}(d^{1/2}T^{-1/3})$ by introducing the Sign-based Stochastic Variance Reduction (SSVR) method, which employs variance reduction estimators to track gradients and leverages their signs to update. For finite-sum problems, our method can be further enhanced to achieve a convergence rate of $\mathcal{O}(m^{1/4}d^{1/2}T^{-1/2})$, where $m$ denotes the number of component functions. Furthermore, we investigate the heterogeneous majority vote in distributed settings and introduce two novel algorithms that attain improved convergence rates of $\mathcal{O}(d^{1/2}T^{-1/2} + dn^{-1/2})$ and $\mathcal{O}(d^{1/4}T^{-1/4})$ respectively, outperforming the previous results of $\mathcal{O}(dT^{-1/4} + dn^{-1/2})$ and $\mathcal{O}(d^{3/8}T^{-1/8})$, where $n$ represents the number of nodes. Numerical experiments across different tasks validate the effectiveness of our proposed methods.

Finetuning large language models (LLMs) in federated learning (FL) settings has become increasingly important as it allows resource-constrained devices to finetune a model using private data. However, finetuning LLMs using backpropagation requires excessive memory (especially from intermediate activations) for resource-constrained devices. While Forward-mode Auto-Differentiation (AD) can significantly reduce memory footprint from activations, we observe that directly applying it to LLM finetuning results in slow convergence and poor accuracy. In this paper, we introduce Spry, an FL algorithm that splits trainable weights of an LLM among participating clients, such that each client computes gradients using forward-mode AD that are closer estimations of the true gradients. Spry achieves a low memory footprint, high accuracy, and fast convergence. We formally prove that the global gradients in Spry are unbiased estimators of true global gradients for homogeneous data distributions across clients, while heterogeneity increases bias of the estimates. We also derive Spry's convergence rate, showing that the gradients decrease inversely proportional to the number of FL rounds, indicating the convergence up to the limits of heterogeneity. Empirically, Spry reduces the memory footprint during training by 1.4-7.1$\times$ in contrast to backpropagation, while reaching comparable accuracy, across a wide range of language tasks, models, and FL settings. Spry reduces the convergence time by 1.2-20.3$\times$ and achieves 5.2-13.5\% higher accuracy against state-of-the-art zero-order methods. When finetuning Llama2-7B with LoRA, compared to the peak memory consumption of 33.9GB of backpropagation, Spry only consumes 6.2GB of peak memory. For OPT13B, the reduction is from 76.5GB to 10.8GB. Spry makes feasible previously impossible FL deployments on commodity mobile and edge devices. Our source code is available for replication at https://github.com/Astuary/Spry.

We investigate bandit convex optimization (BCO) with delayed feedback, where only the loss value of the action is revealed under an arbitrary delay. Let $n,T,\bar{d}$ denote the dimensionality, time horizon, and average delay, respectively. Previous studies have achieved an $O(\sqrt{n}T^{3/4}+(n\bar{d})^{1/3}T^{2/3})$ regret bound for this problem, whose delay-independent part matches the regret of the classical non-delayed bandit gradient descent algorithm. However, there is a large gap between its delay-dependent part, i.e., $O((n\bar{d})^{1/3}T^{2/3})$, and an existing $\Omega(\sqrt{\bar{d}T})$ lower bound. In this paper, we illustrate that this gap can be filled in the worst case, where $\bar{d}$ is very close to the maximum delay $d$. Specifically, we first develop a novel algorithm, and prove that it enjoys a regret bound of $O(\sqrt{n}T^{3/4}+\sqrt{dT})$ in general. Compared with the previous result, our regret bound is better for $d=O((n\bar{d})^{2/3}T^{1/3})$, and the delay-dependent part is tight in the worst case. The primary idea is to decouple the joint effect of the delays and the bandit feedback on the regret by carefully incorporating the delayed bandit feedback with a blocking update mechanism. Furthermore, we show that the proposed algorithm can improve the regret bound to $O((nT)^{2/3}\log^{1/3}T+d\log T)$ for strongly convex functions. Finally, if the action sets are unconstrained, we demonstrate that it can be simply extended to achieve an $O(n\sqrt{T\log T}+d\log T)$ regret bound for strongly convex and smooth functions.

To address the uncertainty in function types, recent progress in online convex optimization (OCO) has spurred the development of universal algorithms that simultaneously attain minimax rates for multiple types of convex functions. However, for a $T$-round online problem, state-of-the-art methods typically conduct $O(\log T)$ projections onto the domain in each round, a process potentially time-consuming with complicated feasible sets. In this paper, inspired by the black-box reduction of Cutkosky and Orabona [2018], we employ a surrogate loss defined over simpler domains to develop universal OCO algorithms that only require $1$ projection. Embracing the framework of prediction with expert advice, we maintain a set of experts for each type of functions and aggregate their predictions via a meta-algorithm. The crux of our approach lies in a uniquely designed expert-loss for strongly convex functions, stemming from an innovative decomposition of the regret into the meta-regret and the expert-regret. Our analysis sheds new light on the surrogate loss, facilitating a rigorous examination of the discrepancy between the regret of the original loss and that of the surrogate loss, and carefully controlling meta-regret under the strong convexity condition. With only $1$ projection per round, we establish optimal regret bounds for general convex, exponentially concave, and strongly convex functions simultaneously. Furthermore, we enhance the expert-loss to exploit the smoothness property, and demonstrate that our algorithm can attain small-loss regret for multiple types of convex and smooth functions.

Watermarking images is critical for tracking image provenance and proving ownership. With the advent of generative models, such as stable diffusion, that can create fake but realistic images, watermarking has become particularly important to make human-created images reliably identifiable. Unfortunately, the very same stable diffusion technology can remove watermarks injected using existing methods. To address this problem, we present ZoDiac, which uses a pre-trained stable diffusion model to inject a watermark into the trainable latent space, resulting in watermarks that can be reliably detected in the latent vector even when attacked. We evaluate ZoDiac on three benchmarks, MS-COCO, DiffusionDB, and WikiArt, and find that ZoDiac is robust against state-of-the-art watermark attacks, with a watermark detection rate above 98% and a false positive rate below 6.4%, outperforming state-of-the-art watermarking methods. We hypothesize that the reciprocating denoising process in diffusion models may inherently enhance the robustness of the watermark when faced with strong attacks and validate the hypothesis. Our research demonstrates that stable diffusion is a promising approach to robust watermarking, able to withstand even stable-diffusion-based attack methods. ZoDiac is open-sourced and available at https://github.com/zhanglijun95/ZoDiac.

Cross-domain recommendation (CDR) aims to leverage the rich information from the source domain to enhance recommendation performance in the target domain. However, the data imbalance problem inherent across different domains compromises the effectiveness of CDR approaches, posing a significant challenge to CDR. Most current CDR methodologies focus on creating better user embeddings for the target domain, yet usually neglect the inconsistency in user activities due to data imbalance. As a result, the process of creating user embeddings tends to prioritize users with more frequent interactions and leave less active users underserved, leading these CDR methods to struggle in making accurate recommendations for those with fewer interactions. Such bias in creating embeddings reveals the fact that ''not all embeddings are created equal'' in CDR, which serves as the primary motivation of this study. Inspired by the recent development of contrastive learning, this paper proposes User-aware Contrastive Learning for Robust cross-domain recommendation (UCLR), enhancing the robustness of cross-domain recommendation. Specifically, our proposed method consists of two sub-modules: (i) pretrained global embedding, where the global user embeddings are pretrained across all the domains; (ii) contrastive dual-stream collaborative autoencoder, where more equal user embeddings are generated by optimizing contrastive loss with individualized temperatures. To further improve the performance of our method in each domain, we finetune the whole framework of UCLR based on Low-Rank Adaptation (LoRA). Theoretically, our method is equipped with a provable convergence guarantee during the contrastive learning stage. Furthermore, we also conduct comprehensive experiments on real-world datasets to validate the effectiveness of our proposed method.

We investigate online convex optimization in non-stationary environments and choose dynamic regret as the performance measure, defined as the difference between cumulative loss incurred by the online algorithm and that of any feasible comparator sequence. Let $T$ be the time horizon and $P_T$ be the path length that essentially reflects the non-stationarity of environments, the state-of-the-art dynamic regret is $\mathcal{O}(\sqrt{T(1+P_T)})$. Although this bound is proved to be minimax optimal for convex functions, in this paper, we demonstrate that it is possible to further enhance the guarantee for some easy problem instances, particularly when online functions are smooth. Specifically, we introduce novel online algorithms that can exploit smoothness and replace the dependence on $T$ in dynamic regret with problem-dependent quantities: the variation in gradients of loss functions, the cumulative loss of the comparator sequence, and the minimum of these two terms. These quantities are at most $\mathcal{O}(T)$ while could be much smaller in benign environments. Therefore, our results are adaptive to the intrinsic difficulty of the problem, since the bounds are tighter than existing results for easy problems and meanwhile safeguard the same rate in the worst case. Notably, our proposed algorithms can achieve favorable dynamic regret with only one gradient per iteration, sharing the same gradient query complexity as the static regret minimization methods. To accomplish this, we introduce the collaborative online ensemble framework. The proposed framework employs a two-layer online ensemble to handle non-stationarity, and uses optimistic online learning and further introduces crucial correction terms to enable effective collaboration within the meta-base two layers, thereby attaining adaptivity. We believe the framework can be useful for broader problems.

The stochastically extended adversarial (SEA) model, introduced by Sachs et al. (2022), serves as an interpolation between stochastic and adversarial online convex optimization. Under the smoothness condition on expected loss functions, it is shown that the expected static regret of optimistic follow-the-regularized-leader (FTRL) depends on the cumulative stochastic variance $\sigma_{1:T}^2$ and the cumulative adversarial variation $\Sigma_{1:T}^2$ for convex functions. Sachs et al. (2022) also provide a regret bound based on the maximal stochastic variance $\sigma_{\max}^2$ and the maximal adversarial variation $\Sigma_{\max}^2$ for strongly convex functions. Inspired by their work, we investigate the theoretical guarantees of optimistic online mirror descent (OMD) for the SEA model with smooth expected loss functions. For convex and smooth functions, we obtain the same $\mathcal{O}(\sqrt{\sigma_{1:T}^2}+\sqrt{\Sigma_{1:T}^2})$ regret bound, but with a relaxation of the convexity requirement from individual functions to expected functions. For strongly convex and smooth functions, we establish an $\mathcal{O}\left(\frac{1}{\lambda}\left(\sigma_{\max}^2+\Sigma_{\max}^2\right)\log \left(\left(\sigma_{1:T}^2 + \Sigma_{1:T}^2\right)/\left(\sigma_{\max}^2+\Sigma_{\max}^2\right)\right)\right)$ bound, better than their $\mathcal{O}((\sigma_{\max}^2$ $ + \Sigma_{\max}^2) \log T)$ result. For exp-concave and smooth functions, our approach yields a new $\mathcal{O}(d\log(\sigma_{1:T}^2+\Sigma_{1:T}^2))$ bound. Moreover, we introduce the first expected dynamic regret guarantee for the SEA model with convex and smooth expected functions, which is more favorable than static regret bounds in non-stationary environments. Furthermore, we expand our investigation to scenarios with non-smooth expected loss functions and propose novel algorithms built upon optimistic OMD with an implicit update, successfully attaining both static and dynamic regret guarantees.

Neural networks with branched architectures, namely, tree-structured models, have been employed to jointly tackle multiple vision tasks in the context of multitask learning (MTL). Such tree-structured networks typically start with a number of shared layers, after which different tasks branch out into their own sequence of layers. Hence, the major challenge is to determine where to branch out for each task given a backbone model to optimize for both task accuracy and computation efficiency. To address the challenge, this article proposes a recommendation system that, given a set of tasks and a convolutional neural network-based backbone model, automatically suggests tree-structured multitask architectures that could achieve a high task performance while meeting a user-specified computation budget without performing model training. Extensive evaluations on popular MTL benchmarks show that the recommended architectures could achieve competitive task accuracy and computation efficiency compared with state-of-the-art MTL methods. Our tree-structured multitask model recommender is open-sourced and available at https://github.com/zhanglijun95/TreeMTL.

We investigate the problem of distributed online convex optimization with complicated constraints, in which the projection operation could be the computational bottleneck. To avoid projections, distributed online projection-free methods have been proposed and attain an O(T^{3/4}) regret bound for general convex losses. However, they cannot utilize the smoothness condition, which has been exploited in the centralized setting to improve the regret. In this paper, we propose a new distributed online projection-free method with a tighter regret bound of O(T^{2/3}) for smooth and convex losses. Specifically, we first provide a distributed extension of Follow-the-Perturbed-Leader so that the smoothness can be utilized in the distributed setting. Then, we reduce the computational cost via sampling and blocking techniques. In this way, our method only needs to solve one linear optimization per round on average. Finally, we conduct experiments on benchmark datasets to verify the effectiveness of our proposed method.

To deal with non-stationary online problems with complex constraints, we investigate the dynamic regret of online Frank-Wolfe (OFW), which is an efficient projection-free algorithm for online convex optimization. It is well-known that in the setting of offline optimization, the smoothness of functions and the strong convexity of functions accompanying specific properties of constraint sets can be utilized to achieve fast convergence rates for the Frank-Wolfe (FW) algorithm. However, for OFW, previous studies only establish a dynamic regret bound of $O(\sqrt{T}(V_T+\sqrt{D_T}+1))$ by utilizing the convexity of problems, where $T$ is the number of rounds, $V_T$ is the function variation, and $D_T$ is the gradient variation. In this paper, we derive improved dynamic regret bounds for OFW by extending the fast convergence rates of FW from offline optimization to online optimization. The key technique for this extension is to set the step size of OFW with a line search rule. In this way, we first show that the dynamic regret bound of OFW can be improved to $O(\sqrt{T(V_T+1)})$ for smooth functions. Second, we achieve a better dynamic regret bound of $O(T^{1/3}(V_T+1)^{2/3})$ when functions are smooth and strongly convex, and the constraint set is strongly convex. Finally, for smooth and strongly convex functions with minimizers in the interior of the constraint set, we demonstrate that the dynamic regret of OFW reduces to $O(V_T+1)$, and can be further strengthened to $O(\min\{P_T^\ast,S_T^\ast,V_T\}+1)$ by performing a constant number of FW iterations per round, where $P_T^\ast$ and $S_T^\ast$ denote the path length and squared path length of minimizers, respectively.

Stochastically Extended Adversarial (SEA) model is introduced by Sachs et al. (2022) as an interpolation between stochastic and adversarial online convex optimization. Under the smoothness condition, they demonstrate that the expected regret of optimistic follow-the-regularized-leader (FTRL) depends on the cumulative stochastic variance $\sigma_{1:T}^2$ and the cumulative adversarial variation $\Sigma_{1:T}^2$ for convex functions. They also provide a slightly weaker bound based on the maximal stochastic variance $\sigma_{\max}^2$ and the maximal adversarial variation $\Sigma_{\max}^2$ for strongly convex functions. Inspired by their work, we investigate the theoretical guarantees of optimistic online mirror descent (OMD) for the SEA model. For convex and smooth functions, we obtain the same $\mathcal{O}(\sqrt{\sigma_{1:T}^2}+\sqrt{\Sigma_{1:T}^2})$ regret bound, without the convexity requirement of individual functions. For strongly convex and smooth functions, we establish an $\mathcal{O}(\min\{\log (\sigma_{1:T}^2+\Sigma_{1:T}^2), (\sigma_{\max}^2 + \Sigma_{\max}^2) \log T\})$ bound, better than their $\mathcal{O}((\sigma_{\max}^2 + \Sigma_{\max}^2) \log T)$ result. For exp-concave and smooth functions, we achieve a new $\mathcal{O}(d\log(\sigma_{1:T}^2+\Sigma_{1:T}^2))$ bound. Owing to the OMD framework, we further establish dynamic regret for convex and smooth functions, which is more favorable in non-stationary online scenarios.

In this paper, we consider non-convex multi-block bilevel optimization (MBBO) problems, which involve $m\gg 1$ lower level problems and have important applications in machine learning. Designing a stochastic gradient and controlling its variance is more intricate due to the hierarchical sampling of blocks and data and the unique challenge of estimating hyper-gradient. We aim to achieve three nice properties for our algorithm: (a) matching the state-of-the-art complexity of standard BO problems with a single block; (b) achieving parallel speedup by sampling $I$ blocks and sampling $B$ samples for each sampled block per-iteration; (c) avoiding the computation of the inverse of a high-dimensional Hessian matrix estimator. However, it is non-trivial to achieve all of these by observing that existing works only achieve one or two of these properties. To address the involved challenges for achieving (a, b, c), we propose two stochastic algorithms by using advanced blockwise variance-reduction techniques for tracking the Hessian matrices (for low-dimensional problems) or the Hessian-vector products (for high-dimensional problems), and prove an iteration complexity of $O(\frac{m\epsilon^{-3}\mathbb{I}(I \textless m)}{I\sqrt{I}}+\frac{m\epsilon^{-3}}{I\sqrt{B}})$ for finding an $\epsilon$-stationary point under appropriate conditions. We also conduct experiments to verify the effectiveness of the proposed algorithms comparing with existing MBBO algorithms.

Learning unnormalized statistical models (e.g., energy-based models) is computationally challenging due to the complexity of handling the partition function. To eschew this complexity, noise-contrastive estimation (NCE) has been proposed by formulating the objective as the logistic loss of the real data and the artificial noise. However, as found in previous works, NCE may perform poorly in many tasks due to its flat loss landscape and slow convergence. In this paper, we study a direct approach for optimizing the negative log-likelihood of unnormalized models from the perspective of compositional optimization. To tackle the partition function, a noise distribution is introduced such that the log partition function can be written as a compositional function whose inner function can be estimated with stochastic samples. Hence, the objective can be optimized by stochastic compositional optimization algorithms. Despite being a simple method, we demonstrate that it is more favorable than NCE by (1) establishing a fast convergence rate and quantifying its dependence on the noise distribution through the variance of stochastic estimators; (2) developing better results for one-dimensional Gaussian mean estimation by showing our objective has a much favorable loss landscape and hence our method enjoys faster convergence; (3) demonstrating better performance on multiple applications, including density estimation, out-of-distribution detection, and real image generation.

In this paper, we aim to optimize a contrastive loss with individualized temperatures in a principled manner. The common practice of using a global temperature parameter $\tau$ ignores the fact that “not all semantics are created equal", meaning that different anchor data may have different numbers of samples with similar semantics, especially when data exhibits long-tails. First, we propose a new robust contrastive loss inspired by distributionally robust optimization (DRO), providing us an intuition about the effect of $\tau$ and a mechanism for automatic temperature individualization. Then, we propose an efficient stochastic algorithm for optimizing the robust contrastive loss with a provable convergence guarantee without using large mini-batch sizes. Theoretical and experimental results show that our algorithm automatically learns a suitable $\tau$ for each sample. Specifically, samples with frequent semantics use large temperatures to keep local semantic structures, while samples with rare semantics use small temperatures to induce more separable features. Our method not only outperforms prior strong baselines (e.g., SimCLR, CLIP) on unimodal and bimodal tasks with larger improvements on imbalanced data but also is less sensitive to hyper-parameters. To our best knowledge, this is the first methodical approach to optimizing a contrastive loss with individualized temperatures. Our proposed algorithms are implemented in the LibAUC library at https://libauc.org.

This paper investigates the problem of generalized linear bandits with heavy-tailed rewards, whose $(1+\epsilon)$-th moment is bounded for some $\epsilon\in (0,1]$. Although there exist methods for generalized linear bandits, most of them focus on bounded or sub-Gaussian rewards and are not well-suited for many real-world scenarios, such as financial markets and web-advertising. To address this issue, we propose two novel algorithms based on truncation and mean of medians. These algorithms achieve an almost optimal regret bound of $\widetilde{O}(dT^{\frac{1}{1+\epsilon}})$, where $d$ is the dimension of contextual information and $T$ is the time horizon. Our truncation-based algorithm supports online learning, distinguishing it from existing truncation-based approaches. Additionally, our mean-of-medians-based algorithm requires only $O(\log T)$ rewards and one estimator per epoch, making it more practical. Moreover, our algorithms improve the regret bounds by a logarithmic factor compared to existing algorithms when $\epsilon=1$. Numerical experimental results confirm the merits of our algorithms.

This paper investigates group distributionally robust optimization (GDRO), with the purpose to learn a model that performs well over $m$ different distributions. First, we formulate GDRO as a stochastic convex-concave saddle-point problem, and demonstrate that stochastic mirror descent (SMD), using $m$ samples in each iteration, achieves an $O(m (\log m)/\epsilon^2)$ sample complexity for finding an $\epsilon$-optimal solution, which matches the $\Omega(m/\epsilon^2)$ lower bound up to a logarithmic factor. Then, we make use of techniques from online learning to reduce the number of samples required in each round from $m$ to $1$, keeping the same sample complexity. Specifically, we cast GDRO as a two-players game where one player simply performs SMD and the other executes an online algorithm for non-oblivious multi-armed bandits. Next, we consider a more practical scenario where the number of samples that can be drawn from each distribution is different, and propose a novel formulation of weighted GDRO, which allows us to derive distribution-dependent convergence rates. Denote by $n_i$ the sample budget for the $i$-th distribution, and assume $n_1 \geq n_2 \geq \cdots \geq n_m$. In the first approach, we incorporate non-uniform sampling into SMD such that the sample budget is satisfied in expectation, and prove that the excess risk of the $i$-th distribution decreases at an $O(\sqrt{n_1 \log m}/n_i)$ rate. In the second approach, we use mini-batches to meet the budget exactly and also reduce the variance in stochastic gradients, and then leverage stochastic mirror-prox algorithm, which can exploit small variances, to optimize a carefully designed weighted GDRO problem. Under appropriate conditions, it attains an $O((\log m)/\sqrt{n_i})$ convergence rate, which almost matches the optimal $O(\sqrt{1/n_i})$ rate of only learning from the $i$-th distribution with $n_i$ samples.

Federated learning (FL) suffers from data heterogeneity, where the diverse data distributions across clients make it challenging to train a single global model effectively. Existing personalization approaches aim to address the data heterogeneity issue by creating a personalized model for each client from the global model that fits their local data distribution. However, these personalized models may achieve lower accuracy than the global model in some clients, resulting in limited performance improvement compared to that without personalization. To overcome this limitation, we propose a per-instance personalization FL algorithm Flow. Flow creates dynamic personalized models that are adaptive not only to each client’s data distributions but also to each client’s data instances. The personalized model allows each instance to dynamically determine whether it prefers the local parameters or its global counterpart to make correct predictions, thereby improving clients’ accuracy. We provide theoretical analysis on the convergence of Flow and empirically demonstrate the superiority of Flow in improving clients’ accuracy compared to state-of-the-art personalization approaches on both vision and language-based tasks.

We consider stochastic graphical bandits, where after pulling an arm, the decision maker observes rewards of not only the chosen arm but also its neighbors in a feedback graph. Most of existing work assumes that the rewards are drawn from bounded or at least sub-Gaussian distributions, which however may be violated in many practical scenarios such as social advertising and financial markets. To settle this issue, we investigate stochastic graphical bandits with heavy-tailed rewards, where the distributions have finite moments of order $1+\epsilon$, for some $\epsilon\in(0, 1]$. Firstly, we develop one UCB-type algorithm, whose expected regret is upper bounded by a sum of gap-based quantities over the clique covering of the feedback graph. The key idea is to estimate the reward means of the selected arm’s neighbors by more refined robust estimators, and to construct a graph-based upper confidence bound for selecting candidates. Secondly, we design another elimination-based strategy and improve the regret bound to a gap-based sum with size controlled by the independence number of the feedback graph. For benign graphs, the independence number could be smaller than the size of the clique covering, resulting in tighter regret bounds. Finally, we conduct experiments on synthetic data to demonstrate the effectiveness of our methods.

In this paper, we study stochastic optimization of areas under precision-recall curves (AUPRC), which is widely used for combating imbalanced classification tasks. Although a few methods have been proposed for maximizing AUPRC, stochastic optimization of AUPRC with convergence guarantee remains an undeveloped territory. A state-of-the-art complexity is $O(1/\epsilon^5)$ for finding an $\epsilon$-stationary solution. In this paper, we further improve the stochastic optimization of AURPC by (i) developing novel stochastic momentum methods with a better iteration complexity of $O(1/\epsilon^4)$ for finding an $\epsilon$-stationary solution; and (ii) designing a novel family of stochastic adaptive methods with the same iteration complexity, which enjoy faster convergence in practice. To this end, we propose two innovative techniques that are critical for improving the convergence: (i) the biased estimators for tracking individual ranking scores are updated in a randomized coordinate-wise manner; and (ii) a momentum update is used on top of the stochastic gradient estimator for tracking the gradient of the objective. The novel analysis of Adam-style updates is also one main contribution. Extensive experiments on various data sets demonstrate the effectiveness of the proposed algorithms. Of independent interest, the proposed stochastic momentum and adaptive algorithms are also applicable to a class of two-level stochastic dependent compositional optimization problems.

Hard parameter sharing in multi-domain learning (MDL) al-lows domains to share some of the model parameters to reduce storage cost while improving prediction accuracy. One com-mon sharing practice is to share the bottom layers of a deep neural network among domains while using separate top layers for each domain. In this work, we revisit this common practice via an empirical study on image classification tasks from a diverse set of visual domains and make two surprising obser-vations. (1) Using separate bottom-layer parameters could achieve significantly better performance than the common practice and this phenomenon holds with different experimen-tal settings. (2) A multi-domain model with a small proportion of domain-specific parameters from bottom layers can achieve competitive performance with independent models trained on each domain separately. Our observations suggest that people adopt the new strategy of using separate bottom-layer parame-ters as a stronger baseline for model design in MDL.

In this paper, we investigate the problem of stochastic multi-level compositional optimization, where the objective function is a composition of multiple smooth but possibly non-convex functions. Existing methods for solving this problem either suffer from sub-optimal sample complexities or need a huge batch size. To address this limitation, we propose a Stochastic Multi-level Variance Reduction method (SMVR), which achieves the optimal sample complexity of $\mathcal{O}\left(1 / \epsilon^{3}\right)$ to find an $\epsilon$-stationary point for non-convex objectives. Furthermore, when the objective function satisfies the convexity or Polyak-{Ł}ojasiewicz (PL) condition, we propose a stage-wise variant of SMVR and improve the sample complexity to $\mathcal{O}\left(1 / \epsilon^{2}\right)$ for convex functions or $\mathcal{O}\left(1 /(\mu\epsilon)\right)$ for non-convex functions satisfying the $\mu$-PL condition. The latter result implies the same complexity for $\mu$-strongly convex functions. To make use of adaptive learning rates, we also develop Adaptive SMVR, which achieves the same optimal complexities but converges faster in practice. All our complexities match the lower bounds not only in terms of $\epsilon$ but also in terms of $\mu$ (for PL or strongly convex functions), without using a large batch size in each iteration.

NDCG, namely Normalized Discounted Cumulative Gain, is a widely used ranking metric in information retrieval and machine learning. However, efficient and provable stochastic methods for maximizing NDCG are still lacking, especially for deep models. In this paper, we propose a principled approach to optimize NDCG and its top-$K$ variant. First, we formulate a novel compositional optimization problem for optimizing the NDCG surrogate, and a novel bilevel compositional optimization problem for optimizing the top-$K$ NDCG surrogate. Then, we develop efficient stochastic algorithms with provable convergence guarantees for the non-convex objectives. Different from existing NDCG optimization methods, the per-iteration complexity of our algorithms scales with the mini-batch size instead of the number of total items. To improve the effectiveness for deep learning, we further propose practical strategies by using initial warm-up and stop gradient operator. Experimental results on multiple datasets demonstrate that our methods outperform prior ranking approaches in terms of NDCG. To the best of our knowledge, this is the first time that stochastic algorithms are proposed to optimize NDCG with a provable convergence guarantee. Our proposed methods are implemented in the LibAUC library at https://libauc.org.

In this paper, we study contrastive learning from an optimization perspective, aiming to analyze and address a fundamental issue of existing contrastive learning methods that either rely on a large batch size or a large dictionary of feature vectors. We consider a global objective for contrastive learning, which contrasts each positive pair with all negative pairs for an anchor point. From the optimization perspective, we explain why existing methods such as SimCLR require a large batch size in order to achieve a satisfactory result. In order to remove such requirement, we propose a memory-efficient Stochastic Optimization algorithm for solving the Global objective of Contrastive Learning of Representations, named SogCLR. We show that its optimization error is negligible under a reasonable condition after a sufficient number of iterations or is diminishing for a slightly different global contrastive objective. Empirically, we demonstrate that SogCLR with small batch size (e.g., 256) can achieve similar performance as SimCLR with large batch size (e.g., 8192) on self-supervised learning task on ImageNet-1K. We also attempt to show that the proposed optimization technique is generic and can be applied to solving other contrastive losses, e.g., two-way contrastive losses for bimodal contrastive learning. The proposed method is implemented in our open-sourced library LibAUC (www.libauc.org).

Recently, several universal methods have been proposed for online convex optimization, and attain minimax rates for multiple types of convex functions simultaneously. However, they need to design and optimize one surrogate loss for each type of functions, making it difficult to exploit the structure of the problem and utilize existing algorithms. In this paper, we propose a simple strategy for universal online convex optimization, which avoids these limitations. The key idea is to construct a set of experts to process the original online functions, and deploy a meta-algorithm over the linearized losses to aggregate predictions from experts. Specifically, the meta-algorithm is required to yield a second-order bound with excess losses, so that it can leverage strong convexity and exponential concavity to control the meta-regret. In this way, our strategy inherits the theoretical guarantee of any expert designed for strongly convex functions and exponentially concave functions, up to a double logarithmic factor. As a result, we can plug in off-the-shelf online solvers as black-box experts to deliver problem-dependent regret bounds. For general convex functions, it maintains the minimax optimality and also achieves a small-loss bound.

Non-stationary online learning has drawn much attention in recent years. In particular, \emph{dynamic regret} and \emph{adaptive regret} are proposed as two principled performance measures for online convex optimization in non-stationary environments. To optimize them, a two-layer online ensemble is usually deployed due to the inherent uncertainty of the non-stationarity, in which a group of base-learners are maintained and a meta-algorithm is employed to track the best one on the fly. However, the two-layer structure raises the concern about the computational complexity--those methods typically maintain $O(\log T)$ base-learners simultaneously for a $T$-round online game and thus perform multiple projections onto the feasible domain per round, which becomes the computational bottleneck when the domain is complicated. In this paper, we present efficient methods for optimizing dynamic regret and adaptive regret, which reduce the number of projections per round from $O(\log T)$ to $1$. Moreover, our obtained algorithms require only one gradient query and one function evaluation at each round. Our technique hinges on the reduction mechanism developed in parameter-free online learning and requires non-trivial twists on non-stationary online methods. Empirical studies verify our theoretical findings.

Variance reduction techniques such as SPIDER/SARAH/STORM have been extensively studied to improve the convergence rates of stochastic non-convex optimization, which usually maintain and update a sequence of estimators for a single function across iterations. What if we need to track multiple functional mappings across iterations but only with access to stochastic samples of $\mathcal{O}(1)$ functional mappings at each iteration? There is an important application in solving an emerging family of coupled compositional optimization problems in the form of $\sum_{i=1}^m f_i(g_i(\mathbf{w}))$, where $g_i$ is accessible through a stochastic oracle. The key issue is to track and estimate a sequence of $\mathbf g(\mathbf{w})=(g_1(\mathbf{w}), \ldots, g_m(\mathbf{w}))$ across iterations, where $\mathbf g(\mathbf{w})$ has $m$ blocks and it is only allowed to probe $\mathcal{O}(1)$ blocks to attain their stochastic values and Jacobians. To improve the complexity for solving these problems, we propose a novel stochastic method named Multi-block-Single-probe Variance Reduced (MSVR) estimator to track the sequence of $\mathbf g(\mathbf{w})$. It is inspired by STORM but introduces a customized error correction term to alleviate the noise not only in stochastic samples for the selected blocks but also in those blocks that are not sampled. With the help of the MSVR estimator, we develop several algorithms for solving the aforementioned compositional problems with improved complexities across a spectrum of settings with non-convex/convex/strongly convex/Polyak-Lojasiewicz (PL) objectives. Our results improve upon prior ones in several aspects, including the order of sample complexities and dependence on the strong convexity parameter. Empirical studies on multi-task deep AUC maximization demonstrate the better performance of using the new estimator.

The online Frank-Wolfe (OFW) method has gained much popularity for online convex optimization due to its projection-free property. Previous studies show that OFW can attain an $O(T^{3/4})$ regret bound for convex losses and an $O(T^{2/3})$ regret bound for strongly convex losses. However, they assume that each gradient queried by OFW is revealed immediately, which may not hold in practice and limits the application of OFW. To address this limitation, we propose a delayed variant of OFW, which allows gradients to be delayed by arbitrary rounds. The main idea is to perform an update similar to OFW after receiving any delayed gradient, and play the latest decision for each round. Despite its simplicity, we prove that our delayed variant of OFW is able to achieve an $O(T^{3/4}+dT^{1/4})$ regret bound for convex losses and an $O(T^{2/3}+d\log T)$ regret bound for strongly convex losses, where $d$ is the maximum delay. This is quite surprising since under a relatively large amount of delay (e.g., $d=O(\sqrt{T})$ for convex losses and $d=O(T^{2/3}/\log T)$ for strongly convex losses), the delayed variant of OFW enjoys the same regret bound as that of the original OFW.

In this paper, we investigate an online prediction strategy named as Discounted-Normal-Predictor [Kapralov and Panigrahy, 2010] for smoothed online convex optimization (SOCO), in which the learner needs to minimize not only the hitting cost but also the switching cost. In the setting of learning with expert advice, Daniely and Mansour [2019] demonstrate that Discounted-Normal-Predictor can be utilized to yield nearly optimal regret bounds over any interval, even in the presence of switching costs. Inspired by their results, we develop a simple algorithm for SOCO: Combining online gradient descent (OGD) with different step sizes sequentially by Discounted-Normal-Predictor. Despite its simplicity, we prove that it is able to minimize the adaptive regret with switching cost, i.e., attaining nearly optimal regret with switching cost on every interval. By exploiting the theoretical guarantee of OGD for dynamic regret, we further show that the proposed algorithm can minimize the dynamic regret with switching cost in every interval.

Multi-task learning (MTL) jointly learns a set of tasks by sharing parameters among tasks. It is a promising approach for reducing storage costs while improving task accuracy for many computer vision tasks. The effective adoption of MTL faces two main challenges. The first challenge is to determine what parameters to share across tasks to optimize for both memory efficiency and task accuracy. The second challenge is to automatically apply MTL algorithms to an arbitrary CNN backbone without requiring time-consuming manual re-implementation and significant domain expertise. This paper addresses the challenges by developing the first programming framework AutoMTL that automates efficient MTL model development for vision tasks. AutoMTL takes as inputs an arbitrary backbone convolutional neural network (CNN) and a set of tasks to learn, and automatically produces a multi-task model that achieves high accuracy and small memory footprint simultaneously. Experiments on three popular MTL benchmarks (CityScapes, NYUv2, Tiny-Taskonomy) demonstrate the effectiveness of AutoMTL over state-of-the-art approaches as well as the generalizability of AutoMTL across CNNs. AutoMTL is open-sourced and available at https://github.com/zhanglijun95/AutoMTL.

For years, machine learning has become the dominant approach to a variety of information retrieval tasks. The performance of machine learning algorithms heavily depends on their hyperparameters. It is hence critical to identity the optimal hyperparameter configuration when applying machine learning algorithms. Most of existing hyperparameter optimization methods assume a static relationship between hyperparameter configuration and algorithmic performance and are thus not suitable for many information retrieval applications with non-stationary environments such as e-commerce recommendation and online advertising. To address this limitation, we study online hyperparameter optimization, where the hyperparameter configuration is optimized on the fly. We formulate online hyperparameter optimization as a non-stationary continuum-armed bandits problem in which each arm corresponds to a hyperparameter configuration and the algorithmic performance is viewed as reward. For this problem, we develop principled methods with strong theoretical guarantees in terms of dynamic regret. The key idea is to adaptively discretize the continuous arm set and estimate the mean reward of each arm via weighted averaging. As a case application, we show how our methods can be applied to optimize the hyperparameter of vector-based candidate generation algorithm and empirically demonstrate the effectiveness and efficiency of our methods on public advertising dataset and online A/B testing. Furthermore, to the best of our knowledge, our methods are the first to achieve sub-linear dynamic regret bounds for continuum-armed bandits, which may be of independent interest.

To deal with complicated constraints via locally light computations in distributed online learning, a recent study has presented a projection-free algorithm called distributed online conditional gradient (D-OCG), and achieved an $O(T^{3/4})$ regret bound for convex losses, where $T$ is the number of total rounds. However, it requires $T$ communication rounds, and cannot utilize the strong convexity of losses. In this paper, we propose an improved variant of D-OCG, namely D-BOCG, which can attain the same $O(T^{3/4})$ regret bound with only $O(\sqrt{T})$ communication rounds for convex losses, and a better regret bound of $O(T^{2/3}(\log T)^{1/3})$ with fewer $O(T^{1/3}(\log T)^{2/3})$ communication rounds for strongly convex losses. The key idea is to adopt a delayed update mechanism that reduces the communication complexity, and redefine the surrogate loss function in D-OCG for exploiting the strong convexity. Furthermore, we provide lower bounds to demonstrate that the $O(\sqrt{T})$ communication rounds required by D-BOCG are optimal (in terms of $T$) for achieving the $O(T^{3/4})$ regret with convex losses, and the $O(T^{1/3}(\log T)^{2/3})$ communication rounds required by D-BOCG are near-optimal (in terms of $T$) for achieving the $O(T^{2/3}(\log T)^{1/3})$ regret with strongly convex losses up to polylogarithmic factors. Finally, to handle the more challenging bandit setting, in which only the loss value is available, we incorporate the classical one-point gradient estimator into D-BOCG, and obtain similar theoretical guarantees.

By employing time-varying proximal functions, adaptive subgradient methods (ADAGRAD) have improved the regret bound and been widely used in online learning and optimization. However, ADAGRAD with full matrix proximal functions (ADA-FULL) cannot handle large-scale problems due to the impractical$O(d^3)$time and$O(d^2)$space complexities, though it has better performance when gradients are correlated. In this paper, we propose two efficient variants of ADA-FULL via a matrix sketching technique called frequent directions (FD). The first variant named as ADA-FD directly utilizes FD to maintain and manipulate low-rank matrices, which reduces the space and time complexities to$O(\tau d)$and$O(\tau ^2d)$respectively, where$d$is the dimensionality and$\tau \ll d$is the sketching size. The second variant named as ADA-FFD further adopts a doubling trick to accelerate FD used in ADA-FD, which reduces the average time complexity to$O(\tau d)$while only doubles the space complexity of ADA-FD. Theoretical analysis reveals that the regret of ADA-FD and ADA-FFD is close to that of ADA-FULL as long as the outer product matrix of gradients is approximately low-rank. Experimental results demonstrate the efficiency and effectiveness of our algorithms.

Learning with feature evolution studies the scenario where the features of the data streams can evolve, i.e., old features vanish and new features emerge. Its goal is to keep the model always performing well even when the features happen to evolve. To tackle this problem, canonical methods assume that the old features will vanish simultaneously and the new features themselves will emerge simultaneously as well. They also assume that there is an overlapping period where old and new features both exist when the feature space starts to change. However, in reality, the feature evolution could be unpredictable, which means that the features can vanish or emerge arbitrarily, causing the overlapping period incomplete. In this article, we propose a novel paradigm: prediction with unpredictable feature evolution (PUFE) where the feature evolution is unpredictable. To address this problem, we fill the incomplete overlapping period and formulate it as a new matrix completion problem. We give a theoretical bound on the least number of observed entries to make the overlapping period intact. With this intact overlapping period, we leverage an ensemble method to take the advantage of both the old and new feature spaces without manually deciding which base models should be incorporated. Theoretical and experimental results validate that our method can always follow the best base models and, thus, realize the goal of learning with feature evolution.

We study a variant of stochastic bandits where the feedback model is specified by a graph. In this setting, after playing an arm, one can observe rewards of not only the played arm but also other arms that are adjacent to the played arm in the graph. Most of the existing work assumes the reward distributions are stationary over time, which, however, is often violated in common scenarios such as recommendation systems and online advertising. To address this limitation, we study stochastic bandits with graph feedback in non-stationary environments and propose algorithms with graph-dependent dynamic regret bounds. When the number of reward distribution changes L is known in advance, one of our algorithms achieves an Õ(√(αLT)) dynamic regret bound. We also develop an adaptive algorithm that can adapt to unknown L and attain an Õ(√(θLT)) dynamic regret. Here, α and θ are some graph-dependent quantities and T is the time horizon.

We study bandits with graph-structured feedback, where a learner repeatedly selects an arm and then observes rewards of the chosen arm as well as its neighbors in the feedback graph. Existing work on graphical bandits assumes either stochastic rewards or adversarial rewards, both of which are extremes and appear rarely in real-world scenarios. In this paper, we study graphical bandits with a reward model that interpolates between the two extremes, where the rewards are overall stochastically generated but a small fraction of them can be adversarially corrupted. For this problem, we propose an online algorithm that can utilize the stochastic pattern and also tolerate the adversarial corruptions. The main idea is to restrict exploration to carefully-designed independent sets of the feedback graph and perform exploitation by adopting a soft version of arm elimination. Theoretical analysis shows that our algorithm attains an $O(\alpha \ln{K} \ln{T} + \alpha C)$ regret, where $\alpha$ is the independence number of the feedback graph, $K$ is the number of arms, $T$ is the time horizon, and $C$ quantifies the total corruptions introduced by the adversary. The effectiveness of our algorithm is demonstrated by numerical experiments.

To efficiently solve high-dimensional problems with complicated constraints, projection-free online learning has received ever-increasing research interest. However, previous studies either focused on static regret that is not suitable for dynamic environments, or only established the dynamic regret bound under the smoothness of losses. In this paper, without the condition of the smoothness, we propose a novel projection-free online algorithm, and achieve an O(max{T^{2/3}V_T^{1/3},T^{1/2}}) dynamic regret bound for convex functions and an O(max{(TV_Tlog T)^{1/2},log T}) dynamic regret bound for strongly convex functions, where T is the time horizon and V_T denotes the variation of loss functions. Specifically, we first improve an existing projection-free algorithm called online conditional gradient (OCG) to enjoy small dynamic regret bounds with the prior knowledge of V_T. To work with unknowable V_T, we maintain multiple instances of the improved OCG that can handle different functional variations, and combine them with a meta-algorithm that can track the best one. Experimental results validate the efficiency and effectiveness of our algorithm.

Approximate matrix multiplication with limited space has received ever-increasing attention due to the emergence of large-scale applications. Recently, based on a popular matrix sketching algorithm---frequent directions, previous work has introduced co-occuring directions (COD) to reduce the approximation error for this problem. Although it enjoys the space complexity of O((m_x+m_y)l) for two input matrices X∈ℝ^{m_x ╳ n} and Y∈ℝ^{m_y ╳ n} where l is the sketch size, its time complexity is O(n(m_x+m_y+l)l), which is still very high for large input matrices. In this paper, we propose to reduce the time complexity by exploiting the sparsity of the input matrices. The key idea is to employ an approximate singular value decomposition (SVD) method which can utilize the sparsity, to reduce the number of QR decompositions required by COD. In this way, we develop sparse co-occuring directions, which reduces the time complexity to Õ((nnz(X)+nnz(Y))l+nl^2) in expectation while keeps the same space complexity as O((m_x+m_y)l), where nnz(X) denotes the number of non-zero entries in X and the Õ notation hides constant factors as well as polylogarithmic factors. Theoretical analysis reveals that the approximation error of our algorithm is almost the same as that of COD. Furthermore, we empirically verify the efficiency and effectiveness of our algorithm.

To efficiently solve online problems with complicated constraints, projection-free algorithms including online frank-wolfe (OFW) and its variants have received significant interest recently. However, in the general case, existing efficient projection-free algorithms only achieved the regret bound of O(T^{3/4}), which is worse than the regret of projection-based algorithms, where T is the number of decision rounds. In this paper, we study the special case of online learning over strongly convex sets, for which we first prove that OFW can enjoy a better regret bound of O(T^{2/3}) for general convex losses. The key idea is to refine the decaying step-size in the original OFW by a simple line search rule. Furthermore, for strongly convex losses, we propose a strongly convex variant of OFW by redefining the surrogate loss function in OFW. We show that it achieves a regret bound of O(T^{2/3}) over general convex sets and a better regret bound of O(T^{1/2}) over strongly convex sets.

The conventional solution to learning to rank problems ranks individual documents by prediction scores greedily. Recent emerged re-ranking models, which take as input initial lists, aim to capture document interdependencies and directly generate the optimal ordered lists. Typically, a re-ranking model is learned from a set of labeled data, which can achieve favorable performance on average. However, it can be suboptimal for individual queries because the available training data is usually highly imbalanced. This problem is challenging due to the absence of informative data for some queries and furthermore, the lack of a good data augmentation policy.

Recurrent Neural Networks (RNN) are widely used for various prediction tasks on sequences such as text, speed signals, program traces, and system logs. Due to RNNs’ inherently sequential behavior, one key challenge for the effective adoption of RNNs is to reduce the time spent on RNN inference and to increase the scope of a prediction. This work introduces CFG-guided compressed learning, an approach that creatively integrates Context-Free Grammar (CFG) and online tokenization into RNN learning and inference for streaming inputs. Through a hierarchical compression algorithm, it compresses an input sequence to a CFG and makes predictions based on the compressed sequence. Its algorithm design employs a set of techniques to overcome the issues from the myopic nature of online tokenization, the tension between inference accuracy and compression rate, and other complexities. Experiments on 16 real-world sequences of various types validate that the proposed compressed learning can successfully recognize and leverage repetitive patterns in input sequences, and effectively translate them into dramatic (1-1762×) inference speedups as well as much (1-7830×) expanded prediction scope, while keeping the inference accuracy satisfactory.

With the increasing amount of multimedia data, cross-modality hashing has made great progress as it achieves sub-linear search time and low memory space. However, due to the huge discrepancy between different modalities, most existing cross-modality hashing methods cannot learn unified hash codes and functions for modalities at the same time. The gap between separated hash codes and functions further leads to bad search performance. In this paper, to address the issues above, we propose a novel end-to-end Deep Unified Cross-Modality Hashing method named DUCMH, which is able to jointly learn unified hash codes and unified hash functions by alternate learning and data alignment. Specifically, to reduce the discrepancy between image and text modalities, DUCMH utilizes data alignment to learn an auxiliary image to text mapping under the supervision of image-text pairs. For text data, hash codes can be obtained by unified hash functions, while for image data, DUCMH first maps images to texts by the auxiliary mapping, and then uses the mapped texts to obtain hash codes. DUCMH utilizes alternate learning to update unified hash codes and functions. Extensive experiments on three representative image-text datasets demonstrate the superiority of our DUCMH over several state-of-the-art cross-modality hashing methods.

In many sequential decision making applications, the change of decision would bring an additional cost, such as the wear-and-tear cost associated with changing server status. To control the switching cost, we introduce the problem of online convex optimization with continuous switching constraint, where the goal is to achieve a small regret given a budget on the \emph{overall} switching cost. We first investigate the hardness of the problem, and provide a lower bound of order $\Omega(\sqrt{T})$ when the switching cost budget $S=\Omega(\sqrt{T})$, and $\Omega(\min\{\frac{T}{S},T\})$ when $S=O(\sqrt{T})$, where $T$ is the time horizon. The essential idea is to carefully design an adaptive adversary, who can adjust the loss function according to the cumulative switching cost of the player incurred so far based on the orthogonal technique. We then develop a simple gradient-based algorithm which enjoys the minimax optimal regret bound. Finally, we show that, for strongly convex functions, the regret bound can be improved to $O(\log T)$ for $S=\Omega(\log T)$, and $O(\min\{T/\exp(S)+S,T\})$ for $S=O(\log T)$.

In this paper, we revisit the problem of smoothed online learning, in which the online learner suffers both a hitting cost and a switching cost, and target two performance metrics: competitive ratio and dynamic regret with switching cost. To bound the competitive ratio, we assume the hitting cost is known to the learner in each round, and investigate the simple idea of balancing the two costs by an optimization problem. Surprisingly, we find that minimizing the hitting cost alone is $\max(1, \frac{2}{\alpha})$-competitive for $\alpha$-polyhedral functions and $1 + \frac{4}{\lambda}$-competitive for $\lambda$-quadratic growth functions, both of which improve state-of-the-art results significantly. Moreover, when the hitting cost is both convex and $\lambda$-quadratic growth, we reduce the competitive ratio to $1 + \frac{2}{\sqrt{\lambda}}$ by minimizing the weighted sum of the hitting cost and the switching cost. To bound the dynamic regret with switching cost, we follow the standard setting of online convex optimization, in which the hitting cost is convex but hidden from the learner before making predictions. We modify Ader, an existing algorithm designed for dynamic regret, slightly to take into account the switching cost when measuring the performance. The proposed algorithm, named as Smoothed Ader, attains an optimal $O(\sqrt{T(1+P_T)})$ bound for dynamic regret with switching cost, where $P_T$ is the path-length of the comparator sequence. Furthermore, if the hitting cost is accessible in the beginning of each round, we obtain a similar guarantee without the bounded gradient condition, and establish an $\Omega(\sqrt{T(1+P_T)})$ lower bound to confirm the optimality.

To deal with changing environments, a new performance measure—adaptive regret, defined as the maximum static regret over any interval, was proposed in online learning. Under the setting of online convex optimization, several algorithms have been successfully developed to minimize the adaptive regret. However, existing algorithms lack universality in the sense that they can only handle one type of convex functions and need apriori knowledge of parameters. By contrast, there exist universal algorithms, such as MetaGrad, that attain optimal static regret for multiple types of convex functions simultaneously. Along this line of research, this paper presents the first universal algorithm for minimizing the adaptive regret of convex functions. Specifically, we borrow the idea of maintaining multiple learning rates in MetaGrad to handle the uncertainty of functions, and utilize the technique of sleeping experts to capture changing environments. In this way, our algorithm automatically adapts to the property of functions (convex, exponentially concave, or strongly convex), as well as the nature of environments (stationary or changing). As a by product, it also allows the type of functions to switch between rounds.

Bandit Convex Optimization (BCO) is a fundamental framework for modeling sequential decision-making with partial information, where the only feedback available to the player is the one-point or two-point function values. In this paper, we investigate BCO in non-stationary environments and choose the dynamic regret as the performance measure, which is defined as the difference between the cumulative loss incurred by the algorithm and that of any feasible comparator sequence. Let $T$ be the time horizon and $P_T$ be the path-length of the comparator sequence that reflects the non-stationarity of environments. We propose a novel algorithm that achieves $O(T^{3/4}(1+P_T)^{1/2})$ and $O(T^{1/2}(1+P_T)^{1/2})$ dynamic regret respectively for the one-point and two-point feedback models. The latter result is optimal, matching the $\Omega(T^{1/2}(1+P_T)^{1/2})$ lower bound established in this paper. Notably, our algorithm is adaptive to the non-stationary environments since it does not require prior knowledge of the path-length $P_T$ ahead of time, which is generally unknown. We further extend the algorithm to an anytime version that does not require to know the time horizon $T$ in advance. Moreover, we study the adaptive regret, another widely used performance measure for online learning in non-stationary environments, and design an algorithm that provably enjoys the adaptive regret guarantees for BCO problems. Finally, we present empirical studies to validate the effectiveness of the proposed approach.

Learning with streaming data has attracted much attention during the past few years. Though most studies consider data stream with fixed features, in real practice the features may be evolvable. For example, features of data gathered by limited-lifespan sensors will change when these sensors are substituted by new ones. In this article, we propose a novel learning paradigm: Feature Evolvable Streaming Learning where old features would vanish and new features would occur. Rather than relying on only the current features, we attempt to recover the vanished features and exploit it to improve performance. Specifically, we learn a mapping from the overlapping period to recover old features and then we learn two models from the recovered features and the current features, respectively. To benefit from the recovered features, we develop two ensemble methods. In the first method, we combine the predictions from two models and theoretically show that with the assistance of old features, the performance on new features can be improved and we provide a tighter bound when the loss function is exponentially concave. In the second approach, we dynamically select the best single prediction and establish a better performance guarantee when the best model switches. Experiments on both synthetic and real data validate the effectiveness of our proposal.

This paper studies the problem of cHarging tAskScheduling for direcTional wireless chargEr networks (HASTE), i.e., given a set of rotatable directional wireless chargers on a 2D area and a series of offline (online) charging tasks, scheduling the orientations of all the chargers with time in a centralized offline (distributed online) fashion to maximize the overall charging utility for all the tasks. We prove that HASTE is NP-hard. Then, we prove that a relaxed version of HASTE falls within the realm of maximizing a submodular function subject to a partition matroid constraint, and propose a centralized offline algorithm that achieves$(1-\rho)(1-\frac{1}{e})$approximation ratio to address HASTE where$\rho$is the switching delay of chargers. Further, we propose a distributed online algorithm and prove it achieves$\frac{1}{2}(1-\rho)(1-\frac{1}{e})$competitive ratio. We conduct simulations and field experiments on a testbed consisting of eight off-the-shelf power transmitters and 8 rechargeable sensor nodes. The results show that our distributed online algorithm achieves 92.97 percent of the optimal charging utility, and outperforms the comparison algorithms by up to 15.28 percent in terms of charging utility.

We aim to design universal algorithms for online convex optimization, which can handle multiple common types of loss functions simultaneously. The previous state-of-the-art universal method has achieved the minimax optimality for general convex, exponentially concave and strongly convex loss functions. However, it remains an open problem whether smoothness can be exploited to further improve the theoretical guarantees. In this paper, we provide an affirmative answer by developing a novel algorithm, namely UFO, which achieves O(√L*), O(d log L*) and O(log L*) regret bounds for the three types of loss functions respectively under the assumption of smoothness, where L* is the cumulative loss of the best comparator in hindsight, and d is dimensionality. Thus, our regret bounds are much tighter when the comparator has a small loss, and ensure the minimax optimality in the worst case. In addition, it is worth pointing out that UFO is the first to achieve the O(log L*) regret bound for strongly convex and smooth functions, which is tighter than the existing small-loss bound by an O(d) factor.

Regret minimization is treated as the golden rule in the traditional study of online learning. However, regret minimization algorithms tend to converge to the static optimum, thus being suboptimal for changing environments. To address this limitation, new performance measures, including dynamic regret and adaptive regret have been proposed to guide the design of online algorithms. The former one aims to minimize the global regret with respect to a sequence of changing comparators, and the latter one attempts to minimize every local regret with respect to a fixed comparator. Existing algorithms for dynamic regret and adaptive regret are developed independently, and only target one performance measure. In this paper, we bridge this gap by proposing novel online algorithms that are able to minimize the dynamic regret and adaptive regret simultaneously. In fact, our theoretical guarantee is even stronger in the sense that one algorithm is able to minimize the dynamic regret over any interval.

Bandit Convex Optimization (BCO) is a fundamental framework for modeling sequential decision-making with partial information, where the only feedback available to the player is the one-point or two-point function values. In this paper, we investigate BCO in non-stationary environments and choose the dynamic regret as the performance measure, which is defined as the difference between the cumulative loss incurred by the algorithm and that of any feasible comparator sequence. Let $T$ be the time horizon and $P_T$ be the path-length of the comparator sequence that reflects the non-stationarity of environments. We propose a novel algorithm that achieves $O(T^{3/4}(1+P_T)^{1/2})$ and $O(T^{1/2}(1+P_T)^{1/2})$ dynamic regret respectively for the one-point and two-point feedback models. The latter result is optimal, matching the $\Omega(T^{1/2}(1+P_T)^{1/2})$ lower bound established in this paper. Notably, our algorithm is more adaptive to non-stationary environments since it does not require prior knowledge of the path-length $P_T$ ahead of time, which is generally unknown.

This paper investigates the problem of non-stationary linear bandits, where the unknown regression parameter is evolving over time. Previous studies have adopted sophisticated mechanisms, such as sliding window or weighted penalty to achieve near-optimal dynamic regret. In this paper, we demonstrate that a simple restarted strategy is sufficient to attain the same regret guarantee. Specifically, we design an UCB-type algorithm to balance exploitation and exploration, and restart it periodically to handle the drift of unknown parameters. Let $T$ be the time horizon, $d$ be the dimension, and $P_T$ be the path-length that measures the fluctuation of the evolving unknown parameter, our approach enjoys an $\tilde{O}(d^{2/3}(1+P_T)^{1/3}T^{2/3})$ dynamic regret, which is nearly optimal, matching the $\Omega(d^{2/3}(1+P_T)^{1/3}T^{2/3})$ minimax lower bound up to logarithmic factors. Empirical studies also validate the efficacy of our approach.

Traditional Learning to Rank (LTR) models in E-commerce are usually trained on logged data from a single domain. However, data may come from multiple domains, such as hundreds of countries in international E-commerce platforms. Learning a single ranking function obscures domain differences, while learning multiple functions for each domain may also be inferior due to ignoring the correlations between domains. It can be formulated as a multi-task learning problem where multiple tasks share the same feature and label space. To solve the above problem, which we name Multi-Scenario Learning to Rank, we propose the Hybrid of implicit and explicit Mixture-of-Experts (HMoE) approach. Our proposed solution takes advantage of Multi-task Mixture-of-Experts to implicitly identify distinctions and commonalities between tasks in the feature space, and improves the performance with a stacked model learning task relationships in the label space explicitly. Furthermore, to enhance the flexibility, we propose an end-to-end optimization method with a task-constrained back-propagation strategy. We empirically verify that the optimization method is more effective than two-stage optimization required by the stacked approach. Experiments on real-world industrial datasets demonstrate that HMoE significantly outperforms the popular multi-task learning methods. HMoE is in-use in the search system of AliExpress and achieved 1.92% revenue gain in the period of one-week online A/B testing. We also release a sampled version of our dataset to facilitate future research.

In this paper, we study a family of non-convex and possibly non-smooth inf-projection minimization problems, where the target objective function is equal to minimization of a joint function over another variable. This problem include difference of convex (DC) functions and a family of bi-convex functions as special cases. We develop stochastic algorithms and establish their first-order convergence for finding a (nearly) stationary solution of the target non-convex function under different conditions of the component functions. To the best of our knowledge, this is the first work that comprehensively studies stochastic optimization of non-convex inf-projection minimization problems with provable convergence guarantee. Our algorithms enable efficient stochastic optimization of a family of non-decomposable DC functions and a family of bi-convex functions. To demonstrate the power of the proposed algorithms we consider an important application in variance-based regularization. Experiments verify the effectiveness of our inf-projection based formulation and the proposed stochastic algorithm in comparison with previous stochastic algorithms based on the min-max formulation for achieving the same effect.

To deal with complicated constraints via locally light computations in distributed online learning, a recent study has presented a projection-free algorithm called distributed online conditional gradient (D-OCG), and achieved an $O(T^{3/4})$ regret bound, where $T$ is the number of prediction rounds. However, in each round, the local learners of D-OCG need to communicate with their neighbors to share the local gradients, which results in a high communication complexity of $O(T)$. In this paper, we first propose an improved variant of D-OCG, namely D-BOCG, which enjoys an $O(T^{3/4})$ regret bound with only $O(\sqrt{T})$ communication complexity. The key idea is to divide the total prediction rounds into $\sqrt{T}$ equally-sized blocks, and only update the local learners at the beginning of each block by performing iterative linear optimization steps. Furthermore, to handle the more challenging bandit setting, in which only the loss value is available, we incorporate the classical one-point gradient estimator into D-BOCG, and obtain similar theoretical guarantees.

In this paper, we study the problem of stochastic linear bandits with finite action sets. Most of existing work assume the payoffs are bounded or sub-Gaussian, which may be violated in some scenarios such as financial markets. To settle this issue, we analyze the linear bandits with heavy-tailed payoffs, where the payoffs admit finite 1+epsilon moments for some epsilon in (0,1]. Through median of means and dynamic truncation, we propose two novel algorithms which enjoy a sublinear regret bound of widetilde{O}(d^(1/2)T^(1/(1+epsilon))), where d is the dimension of contextual information and T is the time horizon. Meanwhile, we provide an Omega(d^(epsilon/(1+epsilon))T^(1/(1+epsilon))) lower bound, which implies our upper bound matches the lower bound up to polylogarithmic factors in the order of d and T when epsilon=1. Finally, we conduct numerical experiments to demonstrate the effectiveness of our algorithms and the empirical results strongly support our theoretical guarantees.

The usual goal of online learning is to minimize the regret, which measures the performance of online learner against a fixed comparator. However, it is not suitable for changing environments in which the best decision may change over time. To address this limitation, new performance measures, including dynamic regret and adaptive regret have been proposed to guide the design of online algorithms. In dynamic regret, the learner is compared with a sequence of comparators, and in adaptive regret, the learner is required to minimize the regret over every interval. In this paper, we will review the recent developments in this area, and highlight our contributions. Specifically, we have proposed novel algorithms to minimize the dynamic regret and adaptive regret, and investigated the relationship between them.

In the big data era, with the increasing amount of multi-media data, approximate nearest neighbor~(ANN) search has been an important but challenging problem. As a widely applied large-scale ANN search method, hashing has made great progress, and achieved sub-linear search time with low memory space. However, the advances in hashing are based on the availability of large and representative datasets, which often contain sensitive information. Typically, the privacy of this individually sensitive information is compromised. In this paper, we tackle this valuable yet challenging problem and formulate a task termed as private hashing, which takes into account both searching performance and privacy protection. Specifically, we propose a novel noise mechanism, i.e., Random Flipping, and two private hashing algorithms, i.e., PHashing and PITQ, with the refined analysis within the framework of differential privacy, since differential privacy is a well-established technique to measure the privacy leakage of an algorithm. Random Flipping targets binary scenarios and leverages the "Imperceptible Lying" idea to guarantee ε-differential privacy by flipping each datum of the binary matrix (noise addition). To preserve ε-differential privacy, PHashing perturbs and adds noise to the hash codes learned by non-private hashing algorithms using Random Flipping. However, the noise addition for privacy in PHashing will cause severe performance drops. To alleviate this problem, PITQ leverages the power of alternative learning to distribute the noise generated by Random Flipping into each iteration while preserving ε-differential privacy. Furthermore, to empirically evaluate our algorithms, we conduct comprehensive experiments on the image search task and demonstrate that proposed algorithms achieve equal performance compared with non-private hashing methods.

We investigate online convex optimization in non-stationary environments and choose the dynamic regret as the performance measure, defined as the difference between cumulative loss incurred by the online algorithm and that of any feasible comparator sequence. Let $T$ be the time horizon and $P_T$ be the path-length that essentially reflects the non-stationarity of environments, the state-of-the-art dynamic regret is $\mathcal{O}(\sqrt{T(1+P_T)})$. Although this bound is proved to be minimax optimal for convex functions, in this paper, we demonstrate that it is possible to further enhance the dynamic regret by exploiting the smoothness condition. Specifically, we propose novel online algorithms that are capable of leveraging smoothness and replace the dependence on $T$ in the dynamic regret by problem-dependent quantities: the variation in gradients of loss functions, the cumulative loss of the comparator sequence, and the minimum of the previous two terms. These quantities are at most $\mathcal{O}(T)$ while could be much smaller in benign environments. Therefore, our results are adaptive to the intrinsic difficulty of the problem, since the bounds are tighter than existing results for easy problems and meanwhile guarantee the same rate in the worst case.

In this paper, we tackle a valuable yet very challenging visual recognition task, where the instances are within a subordinate category, and the target domain undergoes a shift with the source domain. This task, termed as cross-domain fine-grained recognition, relates closely to many real-life scenarios, e.g., recognizing retail products in storage racks by models trained with images collected in controlled environments. To deal with this problem, we design a new algorithm and propose a corresponding fine-grained domain adaptation dataset. Firstly, we propose a novel end-to-end CNN architecture that integrates two specialized modules: an adversarial module for domain alignment and a self-attention module for fine-grained recognition. The adversarial module is used to handle domain shift by gradually aligning the different domains with domain-level and class-level alignments, and strive to help the classifier learn with domain-invariant features generated by nets. The self-attention module is designed to capture discriminative image regions which are crucial for fine-grained visual recognition. Secondly, we collect a large-scale fine-grained domain adaptation dataset of retail products, which contains 52,011 images of 263 classes from 3 domains. Thirdly, we validate the effectiveness of our method on three datasets, showing that the proposed method can yield significant improvements over baseline methods on fine-grained datasets. Besides, we also evaluate the effectiveness of the self-attention module by performing visualization, which can capture the discriminative image regions in both source and target domains.

In this paper, we propose a simple variant of the original SVRG, called variance reduced stochastic gradient descent (VR-SGD). Unlike the choices of snapshot and starting points in SVRG and its proximal variant, Prox-SVRG, the two vectors of VR-SGD are set to the average and last iterate of the previous epoch, respectively. The settings allow us to use much larger learning rates, and also make our convergence analysis more challenging. We also design two different update rules for smooth and nonsmooth objective functions, respectively, which means that VR-SGD can tackle non-smooth and/or non-strongly convex problems directly without any reduction techniques. Moreover, we analyze the convergence properties of VR-SGD for strongly convex problems, which show that VR-SGD attains linear convergence. Different from most algorithms that have no convergence guarantees for nonstrongly convex problems, we also provide the convergence guarantees of VR-SGD for this case, and empirically verify that VR-SGD with varying learning rates achieves similar performance to its momentum accelerated variant that has the optimal convergence rate O(1=T2). Finally, we apply VR-SGD to solve various machine learning problems, such as convex and non-convex empirical risk minimization, and leading eigenvalue computation. Experimental results show that VR-SGD converges significantly faster than SVRG and Prox-SVRG, and usually outperforms state-of-the-art accelerated methods, e.g., Katyusha.

Stochastic approximation (SA) is a classical approach for stochastic convex optimization. Previous studies have demonstrated that the convergence rate of SA can be improved by introducing either smoothness or strong convexity condition. In this paper, we make use of smoothness and strong convexity simultaneously to boost the convergence rate. Let $\lambda$ be the modulus of strong convexity, $\kappa$ be the condition number, $F_*$ be the minimal risk, and $\alpha>1$ be some small constant. First, we demonstrate that, in expectation, an $O(1/[\lambda T^\alpha] + \kappa F_*/T)$ risk bound is attainable when $T = \Omega(\kappa^\alpha)$. Thus, when $F_*$ is small, the convergence rate could be faster than $O(1/[\lambda T])$ and approaches $O(1/[\lambda T^\alpha])$ in the ideal case. Second, to further benefit from small risk, we show that, in expectation, an $O(1/2^{T/\kappa}+F_*)$ risk bound is achievable. Thus, the excess risk reduces exponentially until reaching $O(F_*)$, and if $F_*=0$, we obtain a global linear convergence. Finally, we emphasize that our proof is constructive and each risk bound is equipped with an efficient stochastic algorithm attaining that bound.

We study Lipschitz bandits, where a learner repeatedly plays one arm from an infinite arm set and then receives a stochastic reward whose expectation is a Lipschitz function of the chosen arm. Most of existing work assume the reward distributions are bounded or at least sub-Gaussian, and thus do not apply to heavy-tailed rewards arising in many real-world scenarios such as web advertising and financial markets. To address this limitation, in this paper we relax the assumption on rewards to allow arbitrary distributions that have finite $(1+\epsilon)$-th moments for some $\epsilon \in (0, 1]$, and propose algorithms that enjoy a sublinear regret of $\widetilde{O}(T^{(d_z\epsilon + 1)/(d_z \epsilon + \epsilon + 1)})$ where $T$ is the time horizon and $d_z$ is the zooming dimension. The key idea is to exploit the Lipschitz property of the expected reward function by adaptively discretizing the arm set, and employ upper confidence bound policies with robust mean estimators designed for heavy-tailed distributions. Furthermore, we provide a lower bound for Lipschitz bandits with heavy-tailed rewards, and show that our algorithms are optimal in terms of $T$. Finally, we conduct numerical experiments to demonstrate the effectiveness of our algorithms.

We investigate online convex optimization in changing environments, and choose the adaptive regret as the performance measure. The goal is to achieve a small regret over every interval so that the comparator is allowed to change over time. Different from previous works that only utilize the convexity condition, this paper further exploits smoothness to improve the adaptive regret. To this end, we develop novel adaptive algorithms for convex and smooth functions, and establish problem-dependent regret bounds over any interval. Our regret bounds are comparable to existing results in the worst case, and become much tighter when the comparator has a small loss.

Recent studies have highlighted that deep neural networks (DNNs) are vulnerable to adversarial examples. In this paper, we improve the robustness of DNNs by utilizing techniques of Distance Metric Learning. Specifically, we incorporate Triplet Loss, one of the most popular Distance Metric Learning methods, into the framework of adversarial training. Our proposed algorithm, Adversarial Training with Triplet Loss (AT2L), substitutes the adversarial example against the current model for the anchor of triplet loss to effectively smooth the classification boundary. Furthermore, we propose an ensemble version of AT2L, which aggregates different attack methods and model structures for better defense effects. Our empirical studies verify that the proposed approach can significantly improve the robustness of DNNs without sacrificing accuracy. Finally, we demonstrate that our specially designed triplet loss can also be used as a regularization term to enhance other defense methods.

In this paper, we study the multi-objective bandits (MOB) problem, where a learner repeatedly selects one arm to play and then receives a reward vector consisting of multiple objectives. MOB has found many real-world applications as varied as online recommendation and network routing. On the other hand, these applications typically contain contextual information that can guide the learning process which, however, is ignored by most of existing work. To utilize this information, we associate each arm with a context vector and assume the reward follows the generalized linear model (GLM). We adopt the notion of Pareto regret to evaluate the learner's performance and develop a novel algorithm for minimizing it. The essential idea is to apply a variant of the online Newton step to estimate model parameters, based on which we utilize the upper confidence bound (UCB) policy to construct an approximation of the Pareto front, and then uniformly at random choose one arm from the approximate Pareto front. Theoretical analysis shows that the proposed algorithm achieves an \tilde O(d\sqrt{T}) Pareto regret, where T is the time horizon and d is the dimension of contexts, which matches the optimal result for single objective contextual bandits problem. Numerical experiments demonstrate the effectiveness of our method.

How to restore back-lit images still remains a challenging task. State-of-the-art methods in this field are based on supervised learning and thus they are usually restricted to specific training data. In this paper, we propose a "zero-shot" scheme for back-lit image restoration, which exploits the power of deep learning, but does not rely on any prior image examples or prior training. Specifically, we train a small image-specific CNN, namely ExCNet (short for Exposure Correction Network) at test time, to estimate the "S-curve" that best fits the test back-lit image. Once the S-curve is estimated, the test image can be then restored straightforwardly. ExCNet can adapt itself to different settings per image. This makes our approach widely applicable to different shooting scenes and kinds of back-lighting conditions. Statistical studies performed on 1512 real back-lit images demonstrate that our approach can outperform the competitors by a large margin. To the best of our knowledge, our scheme is the first unsupervised CNN-based back-lit image restoration method. To make the results reproducible, the source code is available at https://cslinzhang.github.io/ExCNet/.

In this paper, we develop a relative error bound for nuclear norm regularized matrix completion, with the focus on the completion of full-rank matrices. Under the assumption that the top eigenspaces of the target matrix are incoherent, we derive a relative upper bound for recovering the best low-rank approximation of the unknown matrix. Although multiple works have been devoted to analyzing the recovery error of full-rank matrix completion, their error bounds are usually additive, making it impossible to obtain the perfect recovery case and more generally difficult to leverage the skewed distribution of eigenvalues. Our analysis is built upon the optimality condition of the regularized formulation and existing guarantees for low-rank matrix completion. To the best of our knowledge, this is the first relative bound that has been proved for the regularized formulation of matrix completion.

In this paper, we present a simple analysis of fast rates with high probability of empirical minimization for it stochastic composite optimization over a finite-dimensional bounded convex set with exponential concave loss functions and an arbitrary convex regularization. To the best of our knowledge, this result is the first of its kind. As a byproduct, we can directly obtain the fast rate with high probability for exponential concave empirical risk minimization with and without any convex regularization, which not only extends existing results of empirical risk minimization but also provides a unified framework for analyzing exponential concave empirical risk minimization with and without any convex regularization. Our proof is very simple only exploiting the covering number of a finite-dimensional bounded set and a concentration inequality of random vectors.

Deep neural network (DNN) as a popular machine learning model is found to be vulnerable to adversarial attack. This attack constructs adversarial examples by adding small perturbations to the raw input, while appearing unmodified to human eyes but will be misclassified by a well-trained classifier. In this paper, we focus on the black-box attack setting where attackers have almost no access to the underlying models. To conduct black-box attack, a popular approach aims to train a substitute model based on the information queried from the target DNN. The substitute model can then be attacked using existing white-box attack approaches, and the generated adversarial examples will be used to attack the target DNN. Despite its encouraging results, this approach suffers from poor query efficiency, i.e., attackers usually needs to query a huge amount of times to collect enough information for training an accurate substitute model. To this end, we first utilize state-of-the-art white-box attack methods to generate samples for querying, and then introduce an active learning strategy to significantly reduce the number of queries needed. Besides, we also propose a diversity criterion to avoid the sampling bias. Our extensive experimental results on MNIST and CIFAR-10 show that the proposed method can reduce more than 90% of queries while preserve attacking success rates and obtain an accurate substitute model which is more than 85% similar with the target oracle.

In the camera equipment manufacturing industry, the exposure calibration is one of the basic steps for manufacturers to consider before launching their products to the market. To this end, a method that can objectively and automatically assess the exposure levels of images taken by the camera is highly desired. However, few studies have been conducted in this area. In this paper, we attempt to solve this issue to some extent and our contributions are twofold. Firstly, in order to facilitate the study of image exposure assessment, an Image Exposure Database$(IE_{ps}D)$is established. In this database, there are 15, 582 images with various exposure levels, and for each image there is an associated subjective exposure score which could reflect its perceptual exposure level. Secondly, we propose a novel highly accurate DCNN-based model, namely$IE_{ps}M$(Image Exposure Metric), to predict the exposure level of a given image.

To cope with changing environments, recent developments in online learning have introduced the concepts of adaptive regret and dynamic regret independently. In this paper, we illustrate an intrinsic connection between these two concepts by showing that the dynamic regret can be expressed in terms of the adaptive regret and the functional variation. This observation implies that strongly adaptive algorithms can be directly leveraged to minimize the dynamic regret. As a result, we present a series of strongly adaptive algorithms that have small dynamic regrets for convex functions, exponentially concave functions, and strongly convex functions, respectively. To the best of our knowledge, this is the first time that exponential concavity is utilized to upper bound the dynamic regret. Moreover, all of those adaptive algorithms do not need any prior knowledge of the functional variation, which is a significant advantage over previous specialized methods for minimizing dynamic regret.

This paper studies the problem of cHarging tAsk Scheduling for direcTional wireless chargEr networks (HASTE), i.e., given a set of rotatable directional wireless chargers on a 2D area and a series of offline (online) charging tasks, scheduling the orientations of all the chargers with time in a centralized offline (distributed online) fashion to maximize the overall charging utility for all the tasks. We prove that HASTE is NP-hard. Then, we prove that a relaxed version of HASTE falls within the realm of maximizing a submodular function subject to a partition matroid constraint, and propose a centralized offline algorithm that achieves (1-ρ)(1-1/e) approximation ratio to address HASTE where ρ is the switching delay of chargers. Further, we propose a distributed online algorithm and prove it achieves 1/2(1-ρ)(1-1/e) competitive ratio. We conduct simulations, and field experiments on a testbed consisting of 8 off-the-shelf power transmitters and 8 rechargeable sensor nodes. The results show that our distributed online algorithm achieves 92.97% of the optimal charging utility, and outperforms the comparison algorithms by up to 26.19% in terms of charging utility.

By employing time-varying proximal functions, adaptive subgradient methods (ADAGRAD) have improved the regret bound and been widely used in online learning and optimization. However, ADAGRAD with full matrix proximal functions (ADA-FULL) cannot deal with large-scale problems due to the impractical time and space complexities, though it has better performance when gradients are correlated. In this paper, we propose ADA-FD, an efficient variant of ADA-FULL based on a deterministic matrix sketching technique called frequent directions. Following ADA-FULL, we incorporate our ADA-FD into both primal-dual subgradient method and composite mirror descent method to develop two efficient methods. By maintaining and manipulating low-rank matrices, at each iteration, the space complexity is reduced from $O(d^2)$ to $O(\tau d)$ and the time complexity is reduced from $O(d^3)$ to $O(\tau^2d)$, where $d$ is the dimensionality of the data and $\tau \ll d$ is the sketching size. Theoretical analysis reveals that the regret of our methods is close to that of ADA-FULL as long as the outer product matrix of gradients is approximately low-rank. Experimental results show that our ADA-FD is comparable to ADA-FULL and outperforms other state-of-the-art algorithms in online convex optimization as well as in training convolutional neural networks (CNN).

To cope with non-stationary environments, recent advances in online optimization have introduced the notion of adaptive regret, which measures the performance of an online learner against different comparators within different time intervals. Previous studies have proposed various algorithms to yield low adaptive regret under different scenarios. However, all of existing algorithms need to query the gradient of the loss function at least O(log t) times in every iteration t, which hinders their applications to broad domains, especially when the evaluation of gradients is expensive. To address this limitation, we propose a series of computationally efficient algorithms for minimizing the adaptive regret of general convex, strongly convex and exponentially concave functions respectively. The key idea is to replace each loss function with a carefully designed surrogate loss, which bounds the original loss function from below. We show that the proposed algorithms only query the gradient once per iteration, and attain the same theoretical guarantees as previous optimal algorithms. Empirical results demonstrate the efficiency and effectiveness of our methods.

Error bound conditions (EBC) are properties that characterize the growth of an objective function when a point is moved away from the optimal set. They have recently received increasing attention in the field of optimization for developing optimization algorithms with fast convergence. However, the studies of EBC in statistical learning are hitherto still limited. The main contributions of this paper are two-fold. First, we develop fast and intermediate rates of empirical risk minimization (ERM) under EBC for risk minimization with Lipschitz continuous, and smooth convex random functions. Second, we establish fast and intermediate rates of an efficient stochastic approximation (SA) algorithm for risk minimization with Lipschitz continuous random functions, which requires only one pass of $n$ samples and adapts to EBC. For both approaches, the convergence rates span a full spectrum between $\widetilde O(1/\sqrt{n})$ and $\widetilde O(1/n)$ depending on the power constant in EBC, and could be even faster than $O(1/n)$ in special cases for ERM. Moreover, these convergence rates are automatically adaptive without using any knowledge of EBC. Overall, this work not only strengthens the understanding of ERM for statistical learning but also brings new fast stochastic algorithms for solving a broad range of statistical learning problems.

In this paper, we study online convex optimization in dynamic environments, and aim to bound the dynamic regret with respect to any sequence of comparators. Existing work have shown that online gradient descent enjoys an $O(\sqrt{T}(1+P_T))$ dynamic regret, where $T$ is the number of iterations and $P_T$ is the path-length of the comparator sequence. However, this result is unsatisfactory, as there exists a large gap from the $\Omega(\sqrt{T(1+P_T)})$ lower bound established in our paper. To address this limitation, we develop a novel online method, namely adaptive learning for dynamic environment (Ader), which achieves an optimal $O(\sqrt{T(1+P_T)})$ dynamic regret. The basic idea is to maintain a set of experts, each attaining an optimal dynamic regret for a specific path-length, and combines them with an expert-tracking algorithm. Furthermore, we propose an improved Ader based on the surrogate loss, and in this way the number of gradient evaluations per round is reduced from $O(\log T)$ to $1$. Finally, we extend Ader to the setting that a sequence of dynamical models is available to characterize the comparators.

In this paper, we consider the problem of linear regression with heavy-tailed distributions. Different from previous studies that use the squared loss to measure the performance, we choose the absolute loss, which is capable of estimating the conditional median. To address the challenge that both the input and output could be heavy-tailed, we propose a truncated minimization problem, and demonstrate that it enjoys an $O(\sqrt{d/n})$ excess risk, where $d$ is the dimensionality and $n$ is the number of samples. Compared with traditional work on $\ell_1$-regression, the main advantage of our result is that we achieve a high-probability risk bound without exponential moment conditions on the input and output. Furthermore, if the input is bounded, we show that the classical empirical risk minimization is competent for $\ell_1$-regression even when the output is heavy-tailed.

In this paper, we address learning problems for high dimensional data. Previously, oblivious random projection based approaches that project high dimensional features onto a random subspace have been used in practice for tackling high-dimensionality challenge in machine learning. Recently, various non-oblivious randomized reduction methods have been developed and deployed for solving many numerical problems such as matrix product approximation, low-rank matrix approximation, etc. However, they are less explored for the machine learning tasks, e.g., classification. More seriously, the theoretical analysis of excess risk bounds for risk minimization, an important measure of generalization performance, has not been established for non-oblivious randomized reduction methods. It therefore remains an open problem what is the benefit of using them over previous oblivious random projection based approaches. To tackle these challenges, we propose an algorithmic framework for employing non-oblivious randomized reduction method for general empirical risk minimizing in machine learning tasks, where the original high-dimensional features are projected onto a random subspace that is derived from the data with a small matrix approximation error. We then derive the first excess risk bound for the proposed non-oblivious randomized reduction approach without requiring strong assumptions on the training data. The established excess risk bound exhibits that the proposed approach provides much better generalization performance and it also sheds more insights about different randomized reduction approaches. Finally, we conduct extensive experiments on both synthetic and real-world benchmark datasets, whose dimension scales to O(10^7), to demonstrate the efficacy of our proposed approach.

Although distance metric learning has been successfully applied to many real-world applications, learning a distance metric from large-scale and high-dimensional data remains a challenging problem. Due to the PSD constraint, the computational complexity of previous algorithms per iteration is at least O(d2) where d is the dimensionality of the data.In this paper, we develop an efficient stochastic algorithm for a class of distance metric learning problems with nuclear norm regularization, referred to as low-rank DML. By utilizing the low-rank structure of the intermediate solutions and stochastic gradients, the complexity of our algorithm has a linear dependence on the dimensionality d. The key idea is to maintain all the iterates in factorized representations and construct stochastic gradients that are low-rank. In this way, the projection onto the PSD cone can be implemented efficiently by incremental SVD. Experimental results on several data sets validate the effectiveness and efficiency of our method.

Although there exist plentiful theories of empirical risk minimization (ERM) for supervised learning, current theoretical understandings of ERM for a related problem—stochastic convex optimization (SCO), are limited. In this work, we strengthen the realm of ERM for SCO by exploiting smoothness and strong convexity conditions to improve the risk bounds. First, we establish an $\widetilde{O}(d/n + \sqrt{F}_*/n)$ risk bound when the random function is nonnegative, convex and smooth, and the expected function is Lipschitz continuous, where $d$ is the dimensionality of the problem, $n$ is the number of samples, and $F_*$ is the minimal risk. Thus, when $F_*$ is small we obtain an $\widetilde{O}(d/n)$ risk bound, which is analogous to the $\widetilde{O}(1/n)$ optimistic rate of ERM for supervised learning. Second, if the objective function is also $λ$-strongly convex, we prove an $\widetilde{O}(d/n + κF_*/n )$ risk bound where $κ$ is the condition number, and improve it to $O(1/[λn^2] + κF_*/n)$ when $n=\widetilde{Ω}(κd)$. As a result, we obtain an $O(κ/n^2)$ risk bound under the condition that $n$ is large and $F_*$ is small, which to the best of our knowledge, is the first $O(1/n^2)$-type of risk bound of ERM. Third, we stress that the above results are established in a unified framework, which allows us to derive new risk bounds under weaker conditions, e.g., without convexity of the random function. Finally, we demonstrate that to achieve an $O(1/[λn^2] + κF_*/n)$ risk bound for supervised learning, the $\widetilde{Ω}(κd)$ requirement on $n$ can be replaced with $Ω(κ^2)$, which is dimensionality-independent.

This paper focuses on convex constrained optimization problems, where the solution is subject to a convex inequality constraint. In particular, we aim at challenging problems for which both projection into the constrained domain and a linear optimization under the inequality constraint are time-consuming, which render both projected gradient methods and conditional gradient methods (a.k.a. the Frank-Wolfe algorithm) expensive. In this paper, we develop projection reduced optimization algorithms for both smooth and non-smooth optimization with improved convergence rates under a certain regularity condition of the constraint function. We first present a general theory of optimization with only one projection. Its application to smooth optimization with only one projection yields $O(1/\epsilon)$ iteration complexity, which improves over the $O(1/\epsilon^2)$ iteration complexity established before for non-smooth optimization and can be further reduced under strong convexity. Then we introduce a local error bound condition and develop faster algorithms for non-strongly convex optimization at the price of a logarithmic number of projections. In particular, we achieve an iteration complexity of $\widetilde O(1/\epsilon^{2(1-\theta)})$ for non-smooth optimization and $\widetilde O(1/\epsilon^{1-\theta})$ for smooth optimization, where $\theta\in(0,1]$ appearing the local error bound condition characterizes the functional local growth rate around the optimal solutions. Novel applications in solving the constrained $\ell_1$ minimization problem and a positive semi-definite constrained distance metric learning problem demonstrate that the proposed algorithms achieve significant speed-up compared with previous algorithms.

By using abundant unlabeled data, semi-supervised learning approaches have been found very useful in various tasks. Existing approaches, however, neglect the fact that the storage available for the learning process is different under different situations, and thus, the learning approaches should be flexible subject to the storage budget limit. In this paper, we focus on graph-based semi-supervised learning and propose two storage fit learning approaches which can adjust their behaviors to different storage budgets. Specifically, we utilize techniques of low-rank matrix approximation to find a low-rank approximator of the similarity matrix so as to reduce the space complexity. The first approach is based on stochastic optimization, which is an iterative approach that converges to the optimal low-rank approximator globally. The second approach is based on Nystrom method, which can find a good low-rank approximator efficiently and is suitable for real-time applications. Experiments on classification tasks show that the proposed methods can fit dynamically different storage budgets and obtain good performances in different scenarios.

Minimizing a convex function of matrices regularized by the nuclear norm arises in many applications such as collaborative filtering and multi-task learning. In this paper, we study the general setting where the convex function could be non-smooth. When the size of the data matrix, denoted by m x n, is very large, existing optimization methods are inefficient because in each iteration, they need to perform a singular value decomposition (SVD) which takes O(m^2 n) time. To reduce the computation cost, we exploit the dual characterization of the nuclear norm to introduce a convex-concave optimization problem and design a subgradient-based algorithm without performing SVD. In each iteration, the proposed algorithm only computes the largest singular vector, reducing the time complexity from O(m^2 n) to O(mn). To the best of our knowledge, this is the first SVD-free convex optimization approach for nuclear-norm regularized problems that does not rely on the smoothness assumption. Theoretical analysis shows that the proposed algorithm converges at an optimal O(1/\sqrt{T}) rate where T is the number of iterations. We also extend our algorithm to the stochastic case where only stochastic subgradients of the convex function are available and a special case that contains an additional non-smooth regularizer (e.g., L1 norm regularizer). We conduct experiments on robust low-rank matrix approximation and link prediction to demonstrate the efficiency of our algorithms.

Recently, deep learning has been successfully applied to the problem of hashing, yielding remarkable performance compared to traditional methods with hand-crafted features. However, most of existing deep hashing methods are designed for the supervised scenario and require a large number of labeled data. In this paper, we propose a novel semi-supervised hashing method for image retrieval, named Deep Hashing with a Bipartite Graph (DHBG), to simultaneously learn embeddings, features and hash codes. More specifically, we construct a bipartite graph to discover the underlying structure of data, based on which an embedding is generated for each instance. Then, we feed raw pixels as well as embeddings to a deep neural network, and concatenate the resulting features to determine the hash code. Compared to existing methods, DHBG is a universal framework that is able to utilize various types of graphs and losses. Furthermore, we propose an inductive variant of DHBG to support out-of-sample extensions. Experimental results on real datasets show that our DHBG outperforms state-of-the-art hashing methods.

Recently, there has been a growing research interest in the analysis of dynamic regret, which measures the performance of an online learner against a sequence of local minimizers. By exploiting the strong convexity, previous studies have shown that the dynamic regret can be upper bounded by the path-length of the comparator sequence. In this paper, we illustrate that the dynamic regret can be further improved by allowing the learner to query the gradient of the function multiple times, and meanwhile the strong convexity can be weakened to other non-degenerate conditions. Specifically, we introduce the squared path-length, which could be much smaller than the path-length, as a new regularity of the comparator sequence. When multiple gradients are accessible to the learner, we first demonstrate that the dynamic regret of strongly convex functions can be upper bounded by the minimum of the path-length and the squared path-length. We then extend our theoretical guarantee to functions that are semi-strongly convex or self-concordant. To the best of our knowledge, this is the first time that semi-strong convexity and self-concordance are utilized to tighten the dynamic regret.

Learning with streaming data has attracted much attention during the past few years.Though most studies consider data stream with fixed features, in real practice the features may be evolvable. For example, features of data gathered by limited lifespan sensors will change when these sensors are substituted by new ones. In this paper, we propose a novel learning paradigm: Feature Evolvable Streaming Learning where old features would vanish and new features would occur. Rather than relying on only the current features, we attempt to recover the vanished features and exploit it to improve performance. Specifically, we learn two models from the recovered features and the current features, respectively. To benefit from the recovered features, we develop two ensemble methods. In the first method, we combine the predictions from two models and theoretically show that with the assistance of old features, the performance on new features can be improved. In the second approach, we dynamically select the best single prediction and establish a better performance guarantee when the best model switches. Experiments on both synthetic and real data validate the effectiveness of our proposal.

Recommendation for e-commerce with a mix of durable and nondurable goods has characteristics that distinguish it from the well-studied media recommendation problem. The demand for items is a combined effect of form utility and time utility, i.e., a product must both be intrinsically appealing to a consumer and the time must be right for purchase. In particular for durable goods, time utility is a function of inter-purchase duration within product category because consumers are unlikely to purchase two items in the same category in close temporal succession. Moreover, purchase data, in contrast to rating data, is implicit with non-purchases not necessarily indicating dislike. Together, these issues give rise to the positive-unlabeled demand-aware recommendation problem that we pose via joint low-rank tensor completion and product category inter-purchase duration vector estimation. We further relax this problem and propose a highly scalable alternating minimization approach with which we can solve problems with millions of users and millions of items in a single thread. We also show superior prediction accuracies on multiple real-world datasets.

In this paper, we study Stochastic Composite Optimization (SCO) for sparse learning that aims to learn a sparse solution from a composite function. Most of the recent SCO algorithms have already reached the optimal expected convergence rate O(1/λT), but they often fail to deliver sparse solutions at the end either due to the limited sparsity regularization during stochastic optimization (SO) or due to the limitation in online-to-batch conversion. Even when the objective function is strongly convex, their high probability bounds can only attain O(√{log(1/δ)/T}) with δ is the failure probability, which is much worse than the expected convergence rate. To address these limitations, we propose a simple yet effective two-phase Stochastic Composite Optimization scheme by adding a novel powerful sparse online-to-batch conversion to the general Stochastic Optimization algorithms. We further develop three concrete algorithms, OptimalSL, LastSL and AverageSL, directly under our scheme to prove the effectiveness of the proposed scheme. Both the theoretical analysis and the experiment results show that our methods can really outperform the existing methods at the ability of sparse learning and at the meantime we can improve the high probability bound to approximately O(log(log(T)/δ)/λT).

Kernel Principal Component Analysis (PCA) is a popular extension of PCA which is able to find nonlinear patterns from data. However, the application of kernel PCA to large-scale problems remains a big challenge, due to its quadratic space complexity and cubic time complexity in the number of examples. To address this limitation, we utilize techniques from stochastic optimization to solve kernel PCA with linear space and time complexities per iteration. Specifically, we formulate it as a stochastic composite optimization problem, where a nuclear norm regularizer is introduced to promote low-rankness, and then develop a simple algorithm based on stochastic proximal gradient descent. During the optimization process, the proposed algorithm always maintains a low-rank factorization of iterates that can be conveniently held in memory. Compared to previous iterative approaches, a remarkable property of our algorithm is that it is equipped with an explicit rate of convergence. Theoretical analysis shows that the solution of our algorithm converges to the optimal one at an O(1/T) rate, where T is the number of iterations.

In this paper, we focus on improving the performance of the Nyström based kernel SVM. Although the Nyström approximation has been studied extensively and its application to kernel classification has been exhibited in several studies, there still exists a potentially large gap between the performance of classifier learned with the Nyström approximation and that learned with the original kernel. In this work, we make novel contributions to bridge the gap without increasing the training costs too much by proposing a refined Nyström based kernel classifier. We adopt a two-step approach that in the first step we learn a sufficiently good dual solution and in the second step we use the obtained dual solution to construct a new set of bases for the Nyström approximation to re-train a refined classifier. Our approach towards learning a good dual solution is based on a sparse-regularized dual formulation with the Nyström approximation, which can be solved with the same time complexity as solving the standard formulation. We justify our approach by establishing a theoretical guarantee on the error of the learned dual solution in the first step with respect to the optimal dual solution under appropriate conditions. The experimental results demonstrate that (i) the obtained dual solution by our approach in the first step is closer to the optimal solution and yields improved prediction performance; and (ii) the second step using the obtained dual solution to re-train the model further improves the performance.

In this paper, we present an accelerated numerical method based on random projection for sparse linear regression. Previous studies have shown that under appropriate conditions, gradient-based methods enjoy a geometric convergence rate when applied to this problem. However, the time complexity of evaluating the gradient is as large as $\mathcal{O}(nd)$, where $n$ is the number of data points and $d$ is the dimensionality, making those methods inefficient for large-scale and high-dimensional dataset. To address this limitation, we first utilize random projection to find a rank-$k$ approximator for the data matrix, and reduce the cost of gradient evaluation to $\mathcal{O}(nk+dk)$, a significant improvement when $k$ is much smaller than $d$ and $n$. Then, we solve the sparse linear regression problem via a proximal gradient method with a homotopy strategy to generate sparse intermediate solutions. Theoretical analysis shows that our method also achieves a global geometric convergence rate, and moreover the sparsity of all the intermediate solutions are well-bounded over the iterations. Finally, we conduct experiments to demonstrate the efficiency of the proposed method.

This work focuses on dynamic regret of online convex optimization that compares the performance of online learning to a clairvoyant who knows the sequence of loss functions in advance and hence selects the minimizer of the loss function at each step. By assuming that the clairvoyant moves slowly (i.e., the minimizers change slowly), we present several improved variation-based upper bounds of the dynamic regret under the true and noisy gradient feedback, which are \it optimal in light of the presented lower bounds. The key to our analysis is to explore a regularity metric that measures the temporal changes in the clairvoyant’s minimizers, to which we refer as path variation. Firstly, we present a general lower bound in terms of the path variation, and then show that under full information or gradient feedback we are able to achieve an optimal dynamic regret. Secondly, we present a lower bound with noisy gradient feedback and then show that we can achieve optimal dynamic regrets under a stochastic gradient feedback and two-point bandit feedback. Moreover, for a sequence of smooth loss functions that admit a small variation in the gradients, our dynamic regret under the two-point bandit feedback matches that is achieved with full information.

In this paper, we study a special bandit setting of online stochastic linear optimization, where only one-bit of information is revealed to the learner at each round. This problem has found many applications including online advertisement and online recommendation. We assume the binary feedback is a random variable generated from the logit model, and aim to minimize the regret defined by the unknown linear function. Although the existing method for generalized linear bandit can be applied to our problem, the high computational cost makes it impractical for real-world applications. To address this challenge, we develop an efficient online learning algorithm by exploiting particular structures of the observation model. Specifically, we adopt online Newton step to estimate the unknown parameter and derive a tight confidence region based on the exponential concavity of the logistic loss. Our analysis shows that the proposed algorithm achieves a regret bound of O(d\sqrtT), which matches the optimal result of stochastic linear bandits.

We consider the problem of image representation from the perspective of statistical design. Recent studies have shown that images are possibly sampled from a low dimensional manifold despite of the fact that the ambient space is usually very high dimensional. Learning low dimensional image representations is crucial for many image processing tasks such as recognition and retrieval. Most of the existing approaches for learning low dimensional representations, such as principal component analysis (PCA) and locality preserving projections (LPP), aim at discovering the geometrical or discriminant structures in the data. In this paper, we take a different perspective from statistical experimental design, and propose a novel dimensionality reduction algorithm called A-Optimal Projection (AOP). AOP is based on a linear regression model. Specifically, AOP finds the optimal basis functions so that the expected prediction error of the regression model can be minimized if the new representations are used for training the model. Experimental results suggest that the proposed approach provides a better representation and achieves higher accuracy in image retrieval.

Feature selection is a challenging problem for high dimensional data processing, which arises in many real applications such as data mining, information retrieval, and pattern recognition. In this paper, we study the problem of unsupervised feature selection. The problem is challenging due to the lack of label information to guide feature selection. We formulate the problem of unsupervised feature selection from the viewpoint of graph regularized data reconstruction. The underlying idea is that the selected features not only preserve the local structure of the original data space via graph regularization, but also approximately reconstruct each data point via linear combination. Therefore, the graph regularized data reconstruction error becomes a natural criterion for measuring the quality of the selected features. By minimizing the reconstruction error, we are able to select the features that best preserve both the similarity and discriminant information in the original data. We then develop an efficient gradient algorithm to solve the corresponding optimization problem. We evaluate the performance of our proposed algorithm on text clustering. The extensive experiments demonstrate the effectiveness of our proposed approach.

In online bandit learning, the learner aims to minimize a sequence of losses, while only observing the value of each loss at a single point. Although various algorithms and theories have been developed for online bandit learning, most of them are limited to convex losses. In this paper, we investigate the problem of online bandit learning with non-convex losses, and develop an efficient algorithm with formal theoretical guarantees. To be specific, we consider a class of losses which is a composition of a non-increasing scalar function and a linear function. This setting models a wide range of supervised learning applications such as online classification with a non-convex loss. Theoretical analysis shows that our algorithm achieves an O(poly(d)T2/3) regret bound when the variation of the loss function is small. To the best of our knowledge, this is the first work in online bandit learning that does not rely on convexity.

In this paper, we consider the problem of recovering the s largest elements of an arbitrary vector from noisy measurements. Inspired by previous work, we develop an homotopy algorithm which solves the \ell_1-regularized least square problem for a sequence of decreasing values of the regularization parameter. Compared to the previous method, our algorithm is more efficient in the sense it only updates the solution once for each intermediate problem, and more practical in the sense it has a simple stopping criterion by checking the sparsity of the intermediate solution. Theoretical analysis reveals that our method enjoys a linear convergence rate in reducing the recovery error. Furthermore, our guarantee for recovering the top s elements of the target vector is tighter than previous results, and that for recovering the target vector itself matches the state of the art in compressive sensing.

In this paper we derive \textithigh probability lower and upper bounds on the excess risk of stochastic optimization of exponentially concave loss functions. Exponentially concave loss functions encompass several fundamental problems in machine learning such as squared loss in linear regression, logistic loss in classification, and negative logarithm loss in portfolio management. We demonstrate an O(d \log T/T) upper bound on the excess risk of stochastic online Newton step algorithm, and an O(d/T) lower bound on the excess risk of any stochastic optimization method for \textitsquared loss, indicating that the obtained upper bound is optimal up to a logarithmic factor. The analysis of upper bound is based on recent advances in concentration inequalities for bounding self-normalized martingales, which is interesting by its own right, and the proof technique used to achieve the lower bound is a probabilistic method and relies on an information-theoretic minimax analysis.

In this paper, we consider the problem of column subset selection. We present a novel analysis of the spectral norm reconstruction for a simple randomized algorithm and establish a new bound that depends explicitly on the sampling probabilities. The sampling dependent error bound (i) allows us to better understand the tradeoff in the reconstruction error due to sampling probabilities, (ii) exhibits more insights than existing error bounds that exploit specific probability distributions, and (iii) implies better sampling distributions. In particular, we show that a sampling distribution with probabilities proportional to the square root of the statistical leverage scores is better than uniform sampling, and is better than leverage-based sampling when the statistical leverage scores are very nonuniform. And by solving a constrained optimization problem related to the error bound with an efficient bisection search we are able to achieve better performance than using either the leverage-based distribution or that proportional to the square root of the statistical leverage scores. Numerical simulations demonstrate the benefits of the new sampling distributions for low-rank matrix approximation and least square approximation compared to state-of-the art algorithms.

In this paper, we study randomized reduction methods, which reduce high-dimensional features into low-dimensional space by randomized methods (e.g., random projection, random hashing), for large-scale high-dimensional classification. Previous theoretical results on randomized reduction methods hinge on strong assumptions about the data, e.g., low rank of the data matrix or a large separable margin of classification, which hinder their in broad domains. To address these limitations, we propose dual-sparse regularized randomized reduction methods that introduce a sparse regularizer into the reduced dual problem. Under a mild condition that the original dual solution is a (nearly) sparse vector, we show that the resulting dual solution is close to the original dual solution and concentrates on its support set. In numerical experiments, we present an empirical study to support the analysis and we also present a novel application of the dual-sparse randomized reduction methods to reducing the communication cost of distributed learning from large-scale high-dimensional data.

Semi-supervised clustering leverages side information such as pairwise constraints to guide clustering procedures. Despite promising progress, existing semi-supervised clustering approaches overlook the condition of side information being generated sequentially, which is a natural setting arising in numerous real-world applications such as social network and e-commerce system analysis. Given emerged new constraints, classical semi-supervised clustering algorithms need to re-optimize their objectives over all data samples and constraints in availability, which prevents them from efficiently updating the obtained data partitions. To address this challenge, we propose an efficient dynamic semi-supervised clustering framework that casts the clustering problem into a search problem over a feasible convex set, i.e., a convex hull with its extreme points being an ensemble of m data partitions. According to the principle of ensemble clustering, the optimal partition lies in the convex hull, and can thus be uniquely represented by an m-dimensional probability simplex vector. As such, the dynamic semi-supervised clustering problem is simplified to the problem of updating a probability simplex vector subject to the newly received pairwise constraints. We then develop a computationally efficient updating procedure to update the probability simplex vector in O(m2) time, irrespective of the data size n. Our empirical studies on several real-world benchmark datasets show that the proposed algorithm outperforms the state-of-the-art semi-supervised clustering algorithms with visible performance gain and significantly reduced running time.

Real-world datasets often involve multiple views of data items, e.g., a Web page can be described by both its content and anchor texts of hyperlinks leading to it; photos in Flickr could be characterized by visual features, as well as user contributed tags. Different views provide information complementary to each other. Synthesizing multi-view features can lead to a comprehensive description of the data items, which could benefit many data analytic applications. Unfortunately, the simple idea of concatenating different feature vectors ignores statistical properties of each view and usually incurs the “curse of dimensionality” problem. We propose Multi-view Concept Learning (MCL), a novel nonnegative latent representation learning algorithm for capturing conceptual factors from multi-view data. MCL exploits both multi-view information and label information. The key idea is to learn a common latent space across different views which (1) captures the semantic relationships between data items through graph embedding regularization on labeled items, and (2) allows each latent factor to be associated with a subset of views via sparseness constraints. In this way, MCL could capture flexible conceptual patterns hidden in multi-view features. Experiments on a toy problem and three real-world datasets show that MCL performs well and outperforms baseline methods.

Expert finding for question answering is a challenging problem in community-based question answering (CQA) systems, arising in many real applications such as question routing and identification of best answers. In order to provide high-quality experts, many existing approaches learn the user model from their past question-answering activities in CQA systems. However, the past activities of users in most CQA systems are rather few, and thus the user model may not be well inferred in practice. In this paper, we consider the problem of expert finding from the viewpoint of missing value estimation. We then employ users' social networks for inferring user model, and thus improve the performance of expert finding in CQA systems. In addition, we develop a novel graph-regularized matrix completion algorithm for inferring the user model. We further develop two efficient iterative procedures, GRMC-EGM and GRMC-AGM, to solve the optimization problem. GRMC-EGM utilizes the Extended Gradient Method (EGM), while GRMC-AGM applies the Accelerated proximal Gradient search Method (AGM), for the optimization. We evaluate our methods on the well-known question answering system Quora, and the popular social network Twitter. Our empirical study shows the effectiveness of the proposed algorithms in comparison to the state-of-the-art expert finding algorithms.

In this paper, we focus on Stochastic Composite Optimization (SCO) for sparse learning that aims to learn a sparse solution. Although many SCO algorithms have been developed for sparse learning with an optimal convergence rate $O(1/T)$, they often fail to deliver sparse solutions at the end either because of the limited sparsity regularization during stochastic optimization or due to the limitation in online-to-batch conversion. To improve the sparsity of solutions obtained by SCO, we propose a simple but effective stochastic optimization scheme that adds a novel sparse online-to-batch conversion to the traditional SCO algorithms. The theoretical analysis shows that our scheme can find a solution with better sparse patterns without affecting the convergence rate. Experimental results on both synthetic and real-world data sets show that the proposed methods are more effective in recovering the sparse solution and have comparable convergence rate as the state-of-the-art SCO algorithms for sparse learning.

While the conventional compressive sensing assumes measurements of infinite precision, one-bit compressive sensing considers an extreme setting where each measurement is quantized to just a single bit. In this paper, we study the vector recovery problem from noisy one-bit measurements, and develop two novel algorithms with formal theoretical guarantees. First, we propose a passive algorithm, which is very efficient in the sense it only needs to solve a convex optimization problem that has a closed-form solution. Despite the apparent simplicity, our theoretical analysis reveals that the proposed algorithm can recover both the exactly sparse and the approximately sparse vectors. In particular, for a sparse vector with s nonzero elements, the sample complexity is O(s \log n/ε^2), where n is the dimensionality and εis the recovery error. This result improves significantly over the previously best known sample complexity in the noisy setting, which is O(s \log n/ε^4). Second, in the case that the noise model is known, we develop an adaptive algorithm based on the principle of active learning. The key idea is to solicit the sign information only when it cannot be inferred from the current estimator. Compared with the passive algorithm, the adaptive one has a lower sample complexity if a high-precision solution is desired.

Learning a statistical model for high-dimensional data is an important topic in machine learning. Although this problem has been well studied in the supervised setting, little is known about its unsupervised counterpart. In this work, we focus on the problem of clustering high-dimensional data with sparse centers. In particular, we address the following open question in unsupervised learning: “is it possible to reliably cluster high-dimensional data when the number of samples is smaller than the data dimensionality?" We develop an efficient clustering algorithm that is able to estimate sparse cluster centers with a single pass over the data. Our theoretical analysis shows that the proposed algorithm is able to accurately recover cluster centers with only O(s\log d) number of samples (data points), provided all the cluster centers are s-sparse vectors in a d dimensional space. Experimental results verify both the effectiveness and efficiency of the proposed clustering algorithm compared to the state-of-the-art algorithms on several benchmark datasets.

Random projection has been widely used in data classification. It maps high-dimensional data into a low-dimensional subspace in order to reduce the computational cost in solving the related optimization problem. While previous studies are focused on analyzing the classification performance in the low-dimensional space, in this paper, we consider the recovery problem, i.e., how to accurately recover the optimal solution to the original high-dimensional optimization problem based on the solution learned after random projection. We present a simple algorithm, termed dual random projection, which uses the dual solution of the low-dimensional optimization problem to recover the optimal solution to the original problem. Our theoretical analysis shows that with a high probability, the proposed algorithm is able to accurately recover the optimal solution to the original problem, provided that the data matrix is (approximately) low-rank and/or optimal solution is (approximately) sparse. We further show that the proposed algorithm can be applied iteratively to reducing the recovery error exponentially.

Random projection has been widely used in data classification. It maps high-dimensional data into a low-dimensional subspace in order to reduce the computational cost in solving the related optimization problem. While previous studies are focused on analyzing the classification performance of using random projection, in this work, we consider the recovery problem, i.e., how to accurately recover the optimal solution to the original optimization problem in the high-dimensional space based on the solution learned from the subspace spanned by random projections. We present a simple algorithm, termed Dual Random Projection, that uses the dual solution of the low-dimensional optimization problem to recover the optimal solution to the original problem. Our theoretical analysis shows that with a high probability, the proposed algorithm is able to accurately recover the optimal solution to the original problem, provided that the data matrix is of low rank or can be well approximated by a low rank matrix.

Many semi-supervised clustering algorithms have been proposed to improve the clustering accuracy by effectively exploring the available side information that is usually in the form of pairwise constraints. Despite the progress, there are two main shortcomings of the existing semi-supervised clustering algorithms. First, they have to deal with non-convex optimization problems, leading to clustering results that are sensitive to the initialization. Second, none of these algorithms is equipped with theoretical guarantee regarding the clustering performance. We address these limitations by developing a framework for semi-supervised clustering based on \it input pattern assisted matrix completion. The key idea is to cast clustering into a matrix completion problem, and solve it efficiently by exploiting the correlation between input patterns and cluster assignments. Our analysis shows that under appropriate conditions, only O(\log n) pairwise constraints are needed to accurately recover the true cluster partition. We verify the effectiveness of the proposed algorithm by comparing it to the state-of-the-art semi-supervised clustering algorithms on several benchmark datasets.

Traditional algorithms for stochastic optimization require projecting the solution at each iteration into a given domain to ensure its feasibility. When facing complex domains, such as the positive semidefinite cone, the projection operation can be expensive, leading to a high computational cost per iteration. In this paper, we present a novel algorithm that aims to reduce the number of projections for stochastic optimization. The proposed algorithm combines the strength of several recent developments in stochastic optimization, including mini-batches, extra-gradient, and epoch gradient descent, in order to effectively explore the smoothness and strong convexity. We show, both in expectation and with a high probability, that when the objective function is both smooth and strongly convex, the proposed algorithm achieves the optimal O(1/T) rate of convergence with only O(logT) projections. Our empirical study verifies the theoretical result.

In this work, we focus on Online Sparse Kernel Learning that aims to online learn a kernel classifier with a bounded number of support vectors. Although many online learning algorithms have been proposed to learn a sparse kernel classifier, most of them fail to bound the number of support vectors used by the final solution which is the average of the intermediate kernel classifiers generated by online algorithms. The key idea of the proposed algorithm is to measure the difficulty in correctly classifying a training example by the derivative of a smooth loss function, and give a more chance to a difficult example to be a support vector than an easy one via a sampling scheme. Our analysis shows that when the loss function is smooth, the proposed algorithm yields similar performance guarantee as the standard online learning algorithm but with a near optimal number of support vectors (up to a poly(lnT) factor). Our empirical study shows promising performance of the proposed algorithm compared to the state-of-the-art algorithms for online sparse kernel learning.

For smooth and strongly convex optimization, the optimal iteration complexity of the gradient-based algorithm is $O(\sqrt{\kappa}\log 1/\epsilon)$, where $\kappa$ is the conditional number. In the case that the optimization problem is ill-conditioned, we need to evaluate a larger number of full gradients, which could be computationally expensive. In this paper, we propose to reduce the number of full gradient required by allowing the algorithm to access the stochastic gradients of the objective function. To this end, we present a novel algorithm named Epoch Mixed Gradient Descent (EMGD) that is able to utilize two kinds of gradients. A distinctive step in EMGD is the mixed gradient descent, where we use an combination of the gradient and the stochastic gradient to update the intermediate solutions. By performing a fixed number of mixed gradient descents, we are able to improve the sub-optimality of the solution by a constant factor, and thus achieve a linear convergence rate. Theoretical analysis shows that EMGD is able to find an $\epsilon$-optimal solution by computing $O(\log 1/\epsilon)$ full gradients and $O(\kappa^2\log 1/\epsilon)$ stochastic gradients.

It is well known that the optimal convergence rate for stochastic optimization of smooth functions is $[O(1/\sqrt{T})]$, which is same as stochastic optimization of Lipschitz continuous convex functions. This is in contrast to optimizing smooth functions using full gradients, which yields a convergence rate of $[O(1/T^2)]$. In this work, we consider a new setup for optimizing smooth functions, termed as {\bf Mixed Optimization}, which allows to access both a stochastic oracle and a full gradient oracle. Our goal is to significantly improve the convergence rate of stochastic optimization of smooth functions by having an additional small number of accesses to the full gradient oracle. We show that, with an $[O(\ln T)]$ calls to the full gradient oracle and an $O(T)$ calls to the stochastic oracle, the proposed mixed optimization algorithm is able to achieve an optimization error of $[O(1/T)]$.

Document summarization is of great value to many real world applications, such as snippets generation for search results and news headlines generation. Traditionally, document summarization is implemented by extracting sentences that cover the main topics of a document with a minimum redundancy. In this paper, we take a different perspective from data reconstruction and propose a novel framework named Document Summarization based on Data Reconstruction (DSDR). Specifically, our approach generates a summary which consist of those sentences that can best reconstruct the original document. To model the relationship among sentences, we introduce two objective functions: (1) linear reconstruction, which approximates the document by linear combinations of the selected sentences; (2) nonnegative linear reconstruction, which allows only additive, not subtractive, linear combinations. In this framework, the reconstruction error becomes a natural criterion for measuring the quality of the summary. For each objective function, we develop an efficient algorithm to solve the corresponding optimization problem. Extensive experiments on summarization benchmark data sets DUC 2006 and DUC 2007 demonstrate the effectiveness of our proposed approach.

In this paper, we study the problem of large-scale Kernel Logistic Regression (KLR). A straightforward approach is to apply stochastic approximation to KLR. We refer to this approach as non-conservative online learning algorithm because it updates the kernel classifier after every received training example, leading to a dense classifier. To improve the sparsity of the KLR classifier, we propose two conservative online learning algorithms that update the classifier in a stochastic manner and generate sparse solutions. With appropriately designed updating strategies, our analysis shows that the two conservative algorithms enjoy similar theoretical guarantee as that of the non-conservative algorithm. Empirical studies on several benchmark data sets demonstrate that compared to batch-mode algorithms for KLR, the proposed conservative online learning algorithms are able to produce sparse KLR classifiers, and achieve similar classification accuracy but with significantly shorter training time. Furthermore, both the sparsity and classification accuracy of our methods are comparable to those of the online kernel SVM.

The goal of feature selection is to identify the most informative features for compact representation, whereas the goal of active learning is to select the most informative instances for prediction. Previous studies separately address these two problems, despite of the fact that selecting features and instances are dual operations over a data matrix. In this paper, we consider the novel problem of simultaneously selecting the most informative features and instances and develop a solution from the perspective of optimum experimental design. That is, by using the selected features as the new representation and the selected instances as training data, the variance of the parameter estimate of a learning function can be minimized. Specifically, we propose a novel approach, which is called Unified criterion for Feature and Instance selection (UFI), to simultaneously identify the most informative features and instances that minimize the trace of the parameter covariance matrix. A greedy algorithm is introduced to efficiently solve the optimization problem. Experimental results on two benchmark data sets demonstrate the effectiveness of our proposed method.

Different from traditional one-sided clustering techniques, coclustering makes use of the duality between samples and features to partition them simultaneously. Most of the existing co-clustering algorithms focus on modeling the relationship between samples and features, whereas the intersample and interfeature relationships are ignored. In this paper, we propose a novel coclustering algorithm named Locally Discriminative Coclustering (LDCC) to explore the relationship between samples and features as well as the intersample and interfeature relationships. Specifically, the sample-feature relationship is modeled by a bipartite graph between samples and features. And we apply local linear regression to discovering the intrinsic discriminative structures of both sample space and feature space. For each local patch in the sample and feature spaces, a local linear function is estimated to predict the labels of the points in this patch. The intersample and interfeature relationships are thus captured by minimizing the fitting errors of all the local linear functions. In this way, LDCC groups strongly associated samples and features together, while respecting the local structures of both sample and feature spaces. Our experimental results on several benchmark data sets have demonstrated the effectiveness of the proposed method.

We consider the active learning problem, which aims to select the most representative points. Out of many existing active learning techniques, optimum experimental design (OED) has received considerable attention recently. The typical OED criteria minimize the variance of the parameter estimates or predicted value. However, these methods see only global euclidean structure, while the local manifold structure is ignored. For example, I-optimal design selects those data points such that other data points can be best approximated by linear combinations of all the selected points. In this paper, we propose a novel active learning algorithm which takes into account the local structure of the data space. That is, each data point should be approximated by the linear combination of only its neighbors. Given the local reconstruction coefficients for every data point and the coordinates of the selected points, a transductive learning algorithm called Locally Linear Reconstruction (LLR) is proposed to reconstruct every other point. The most representative points are thus defined as those whose coordinates can be used to best reconstruct the whole data set. The sequential and convex optimization schemes are also introduced to solve the optimization problem. The experimental results have demonstrated the effectiveness of our proposed method.

Sparse coding has received an increasing amount of interest in recent years. It is an unsupervised learning algorithm, which finds a basis set capturing high-level semantics in the data and learns sparse coordinates in terms of the basis set. Originally applied to modeling the human visual cortex, sparse coding has been shown useful for many applications. However, most of the existing approaches to sparse coding fail to consider the geometrical structure of the data space. In many real applications, the data is more likely to reside on a low-dimensional submanifold embedded in the high-dimensional ambient space. It has been shown that the geometrical information of the data is important for discrimination. In this paper, we propose a graph based algorithm, called graph regularized sparse coding, to learn the sparse representations that explicitly take into account the local manifold structure of the data. By using graph Laplacian as a smooth operator, the obtained sparse representations vary smoothly along the geodesics of the data manifold. The extensive experimental results on image classification and clustering have demonstrated the effectiveness of our proposed algorithm.

In many real world applications, labeled data are usually expensive to get, while there may be a large amount of unlabeled data. To reduce the labeling cost, active learning attempts to discover the most informative data points for labeling. Recently, Optimal Experimental Design (OED) techniques have attracted an increasing amount of attention. OED is concerned with the design of experiments that minimizes variances of a parameterized model. Typical design criteria include D-, A-, and E-optimality. However, all these criteria are based on an ordinary linear regression model which aims to minimize the empirical error whereas the geometrical structure of the data space is not well respected. In this paper, we propose a novel optimal experimental design approach for active learning, called Laplacian G-Optimal Design (LapGOD), which considers both discriminating and geometrical structures. By using Laplacian Regularized Least Squares which incorporates manifold regularization into linear regression, our proposed algorithm selects those data points that minimizes the maximum variance of the predicted values on the data manifold. We also extend our algorithm to nonlinear case by using kernel trick. The experimental results on various image databases have shown that our proposed LapGOD active learning algorithm can significantly enhance the classification accuracy if the selected data points are used as training data.

In the form of topic discussions, users interact with each other to share knowledge and exchange information in online forums. Modeling the evolution of topic discussion reveals how information propagates on Internet and can thus help understand sociological phenomena and improve the performance of applications such as recommendation systems. In this paper, we argue that a user’s participation in topic discussions is motivated by either her friends or her own preferences. Inspired by the theory of information flow, we propose dynamic topic discussion models by mining influential relationships between users and individual preferences. Reply relations of users are exploited to construct the fundamental influential social network. The property of discussed topics and time lapse factor are also considered in our modeling. Furthermore, we propose a novel measure called ParticipationRank to rank users according to how important they are in the social network and to what extent they prefer to participate in the discussion of a certain topic. The experiments show our model can simulate the evolution of topic discussions well and predict the tendency of user’s participation accurately.

The phenomenon of propagation is universal in our daily life. For example, infectious diseases can be transmitted from one person to another, hot news is disseminated widely on the Internet, and classic passages written by the popular users can be shared by many other users in online social network sites. With the emerging of online social network sites, such as Facebook, YouTube, and Flickr, many literatures try to analyse patterns of information propagation and design effective virtual marketing strategies in these sites. However, few metrics have been designed to measure the characteristic of information propagation. In this paper, we propose a novel metric for measuring members’ contribution to information propagation in online social network sites. As a case, we analyse large-scale traces of members’ contribution to photo dissemination in Flickr and find that the distribution of members’ contribution follows a power law distribution, which reveals that most of the information is created and propagated by a few members.

Acoustic-based music recommender systems have received increasing interest in recent years. Due to the semantic gap between low level acoustic features and high level music concepts, many researchers have explored collaborative filtering techniques in music recommender systems. Traditional collaborative filtering music recommendation methods only focus on user rating information. However, there are various kinds of social media information, including different types of objects and relations among these objects, in music social communities such as Last.fm and Pandora. This information is valuable for music recommendation. However, there are two challenges to exploit this rich social media information: (a) There are many different types of objects and relations in music social communities, which makes it difficult to develop a unified framework taking into account all objects and relations. (b) In these communities, some relations are much more sophisticated than pairwise relation, and thus cannot be simply modeled by a graph. In this paper, we propose a novel music recommendation algorithm by using both multiple kinds of social media information and music acoustic-based content. Instead of graph, we use hypergraph to model the various objects and relations, and consider music recommendation as a ranking problem on this hypergraph. While an edge of an ordinary graph connects only two objects, a hyperedge represents a set of objects. In this way, hypergraph can be naturally used to model high-order relations. Experiments on a data set collected from the music social community Last.fm have demonstrated the effectiveness of our proposed algorithm.

Bag of features (BoF) representation has attracted an increasing amount of attention in large scale image processing systems. BoF representation treats images as loose collections of local invariant descriptors extracted from them. The visual codebook is generally constructed by using an unsupervised algorithm such as K-means to quantize the local descriptors into clusters. Images are then represented by the frequency histograms of the codewords contained in them. To build a compact and discriminative codebook, codeword selection has become an indispensable tool. However, most of the existing codeword selection algorithms are supervised and the human labeling may be very expensive. In this paper, we consider the problem of unsupervised codeword selection, and propose a novel algorithm called Discriminative Codeword Selection (DCS). Motivated from recent studies on discriminative clustering, the central idea of our proposed algorithm is to select those codewords so that the cluster structure of the image database can be best respected. Specifically, a multi-output linear function is fitted to model the relationship between the data matrix after codeword selection and the indicator matrix. The most discriminative codewords are thus defined as those leading to minimal fitting error. Experiments on image retrieval and clustering have demonstrated the effectiveness of the proposed method.

Online monitoring of temporally-sequenced news streams for interesting patterns and trends has gained popularity in the last decade. In this paper, we study a particular news stream monitoring task: timely detection of bursty events which have happened recently and discovery of their evolutionary patterns along the timeline. Here, a news stream is represented as feature streams of tens of thousands of features (i.e., keyword. Each news story consists of a set of keywords.). A bursty event therefore is composed of a group of bursty features, which show bursty rises in frequency as the related event emerges. In this paper, we give a formal definition to the above problem and present a solution with the following steps: (1) applying an online multi-resolution burst detection method to identify bursty features with different bursty durations within a recent time period; (2) clustering bursty features to form bursty events and associating each event with a power value which reflects its bursty level; (3) applying an information retrieval method based on cosine similarity to discover the event’s evolution (i.e., highly related bursty events in history) along the timeline. We extensively evaluate the proposed methods on the Reuters Corpus Volume 1. Experimental results show that our methods can detect bursty events in a timely way and effectively discover their evolution. The power values used in our model not only measure event’s bursty level or relative importance well at a certain time point but also show relative strengths of events along the same evolution.

Content Based Image Retrieval (CBIR) has become one of the most active research areas in computer science. Relevance feedback is often used in CBIR systems to bridge the semantic gap. Typically, users are asked to make relevance judgements on some query results, and the feedback information is then used to re-rank the images in the database. An effective relevance feedback algorithm must provide the users with the most informative images with respect to the ranking function. In this paper, we propose a novel active learning algorithm, called Convex Laplacian Regularized I-optimal Design (CLapRID), for relevance feedback image retrieval. Our algorithm is based on a regression model which minimizes the least square error on the labeled images and simultaneously preserves the intrinsic geometrical structure of the image space. It selects the most informative images which minimize the average predictive variance. The optimization problem of CLapRID can be cast as a semidefinite programming (SDP) problem, and solved via interior-point methods. Experimental results on COREL database have demonstrate the effectiveness of the proposed algorithm for relevance feedback image retrieval. Categories and Subject Descriptors H.3.3 [Information storage and retrieval]: Information search and retrieval—Relevance feedback; G.3 [Mathematics of Computing]: Probability and Statistics—Experimental design