The goal of the Inverse reinforcement learning (IRL) task is to identify the underlying reward function and the corresponding optimal policy from a set of expert demonstrations. While most IRL algorithms’ theoretical guarantees rely on a linear reward structure, we aim to extend the theoretical understanding of IRL to scenarios where the reward function is parameterized by neural networks. Meanwhile, conventional IRL algorithms usually adopt a nested structure, leading to computational inefficiency, especially in high-dimensional settings. To address this problem, we propose the first two-timescale single-loop IRL algorithm under neural network parameterized reward and provide a non-asymptotic convergence analysis under overparameterization. Although prior optimality results for linear rewards do not apply, we show that our algorithm can identify the globally optimal reward and policy under certain neural network structures. This is the first IRL algorithm with a non-asymptotic convergence guarantee that provably achieves global optimality in neural network settings.

We introduce a zero-shot merging framework for large language models (LLMs) that consolidates specialized domain experts into a single model without any further training.Our core contribution lies in leveraging relative task vectors-difference representations encoding each expert's unique traits with respect to a shared base model-to guide a principled and efficient merging process.By dissecting parameters into common dimensions (averaged across experts) and complementary dimensions (unique to each expert), we strike an optimal balance between generalization and specialization.We further devise a compression mechanism for the complementary parameters, retaining only principal components and scalar multipliers per expert, thereby minimizing overhead.A dynamic router then selects the most relevant domain at inference, ensuring that domain-specific precision is preserved.Experiments on code generation, mathematical reasoning, medical question answering, and instruction-following benchmarks confirm the versatility and effectiveness of our approach.Altogether, this framework enables truly adaptive and scalable LLMs that seamlessly integrate specialized knowledge for improved zero-shot performance.

Differential privacy (DP) offers a robust framework for safeguarding individual data privacy. To utilize DP in training modern machine learning models, differentially private optimizers have been widely used in recent years. A popular approach to privatize an optimizer is to clip the individual gradients and add sufficiently large noise to the clipped gradient. This approach led to the development of DP optimizers that have comparable performance with their non-private counterparts in fine-tuning tasks or in tasks with a small number of training parameters. However, a significant performance drop is observed when these optimizers are applied to large-scale training. This degradation stems from the substantial noise injection required to maintain DP, which disrupts the optimizer's dynamics. This paper introduces DiSK, a novel framework designed to significantly enhance the performance of DP optimizers. DiSK employs Kalman filtering, a technique drawn from control and signal processing, to effectively denoise privatized gradients and generate progressively refined gradient estimations. To ensure practicality for large-scale training, we simplify the Kalman filtering process, minimizing its memory and computational demands. We establish theoretical privacy-utility trade-off guarantees for DiSK, and demonstrate provable improvements over standard DP optimizers like DPSGD in terms of iteration complexity upper-bound. Extensive experiments across diverse tasks, including vision tasks such as CIFAR-100 and ImageNet-1k and language fine-tuning tasks such as GLUE, E2E, and DART, validate the effectiveness of DiSK. The results showcase its ability to significantly improve the performance of DP optimizers, surpassing state-of-the-art results under the same privacy constraints on several benchmarks.

Aligning to human preferences and/or intentions is an important requirement for contemporary foundation models. To ensure alignment, popular approaches such as reinforcement learning with human feedback (RLHF) break down the task into three stages: (i) a model is computed with supervised fine-tuning (SFT) based upon large demonstrations data, (ii) a reward model (RM) is estimated based upon human feedback data, and (iii) reinforcement learning (RL) is used to further refine the SFT model by optimizing the estimated reward model. Demonstrations and human feedback data reflect human user preferences in different ways. As a result, the reward model estimate obtained from only human feedback data is likely not as accurate as a reward model estimate obtained from both demonstration and human feedback data. A policy model that optimizes the reward model estimate obtained from both demonstration and human feedback data will likely exhibit better alignment performance. We introduce a tractable algorithm for finding the reward and policy models and provide a finite-time performance guarantee. Additionally, we demonstrate the efficiency of the proposed solution with extensive experiments including alignment problems in LLMs and robotic control problems in MuJoCo. We observe that the proposed solutions outperform the existing alignment algorithm by large margins, especially when the amounts of demonstration and preference data are unbalanced.

Large Language Models (LLMs) are increasingly deployed as chatbots, yet their ability to personalize responses to user preferences remains limited. We introduce PrefEval, a benchmark for evaluating LLMs' ability to infer, memorize and adhere to user preferences in long-context conversational setting. PrefEval comprises 3,000 manually curated user preference and query pairs spanning 20 topics. PrefEval contains user personalization or preference information in both explicit and implicit preference forms, and evaluates LLM performance using a generation and a classification task. With PrefEval, we have evaluated 10 open-sourced and proprietary LLMs in multi-session conversations with varying context lengths up to 100k tokens. We benchmark with various prompting, iterative feedback, and retrieval-augmented generation methods. Our benchmarking effort reveals that state-of-the-art LLMs face significant challenges in following users' preference during conversations. In particular, in zero-shot settings, preference following accuracy falls below 10\% at merely 10 turns (~3k tokens) across most evaluated models. Even with advanced prompting and retrieval methods, preference following still deteriorates in long-context conversations. Furthermore, we show that fine-tuning on PrefEval significantly improves performance. We believe PrefEval serves as a valuable resource for measuring, understanding, and enhancing LLMs' proactive preference following abilities, paving the way for personalized conversational agents.

Supervised fine-tuning is a standard method for adapting pre-trained large language models (LLMs) to downstream tasks. Quantization has been recently studied as a post-training technique for efficient LLM deployment. To obtain quantized fine-tuned LLMs, conventional pipelines would first fine-tune the pre-trained models, followed by post-training quantization. This often yields suboptimal performance as it fails to leverage the synergy between fine-tuning and quantization. To effectively realize low-bit quantization of weights, activations and KV caches in LLMs, we propose an algorithm named Rotated Straight-Through-Estimator (RoSTE), which combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy that identifies an effective rotation configuration to reduce activation outliers. We provide theoretical insights on RoSTE by analyzing its prediction error when applied to an overparameterized least square quantized training problem. Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration. Experiments on Pythia, Qwen and Llama models of different sizes demonstrate the effectiveness of RoSTE. Compared to existing post-SFT quantization baselines, our method consistently achieves superior performances across various tasks and different LLM architectures. Our code is available at https://github.com/OptimAI-Lab/RoSTE.

Large Language Models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks, yet generating reliable reasoning processes remains a significant challenge. We present a unified probabilistic framework that formalizes LLM reasoning through a novel graphical model incorporating latent thinking processes and evaluation signals. Our framework addresses two critical questions: (1) how to generate high-quality reasoning processes during inference automatically, and (2) how to integrate these processes into post-training. We propose the Bootstrapping Reinforced Thinking Process (BRiTE) algorithm and demonstrate its theoretical convergence at a rate of $1/T$, where $T$ is the number of iterations. The algorithm operates in two steps. First, it generates high-quality rationales by approximating the desired posterior distribution using a reinforcement learning approach with a novel reward shaping mechanism. Second, it fine-tunes the base LLM by maximizing the joint probability of rationale generation with respect to LLM parameters. Empirical evaluation on GSM8K and MATH benchmarks demonstrates that our approach consistently improves performance across different model sizes without requiring human-annotated thinking processes, outperforming standard chain-of-thought prompting while enhancing existing post-training methods.

The LLM unlearning technique has recently been introduced to comply with data regulations and address the safety and ethical concerns of LLMs by removing the undesired data-model influence. However, state-of-the-art unlearning methods face a critical vulnerability: they are susceptible to ``relearning'' the removed information from a small number of forget data points, known as relearning attacks. In this paper, we systematically investigate how to make unlearned models robust against such attacks. For the first time, we establish a connection between robust unlearning and sharpness-aware minimization (SAM) through a unified robust optimization framework, in an analogy to adversarial training designed to defend against adversarial attacks. Our analysis for SAM reveals that smoothness optimization plays a pivotal role in mitigating relearning attacks. Thus, we further explore diverse smoothing strategies to enhance unlearning robustness. Extensive experiments on benchmark datasets, including WMDP and MUSE, demonstrate that SAM and other smoothness optimization approaches consistently improve the resistance of LLM unlearning to relearning attacks. Notably, smoothness-enhanced unlearning also helps defend against (input-level) jailbreaking attacks, broadening our proposal's impact in robustifying LLM unlearning. Codes are available at https://github.com/OPTML-Group/Unlearn-Smooth.

In this work, we focus on the alignment problem of diffusion models with a continuous reward function, which represents specific objectives for downstream tasks, such as increasing darkness or improving the aesthetics of images. The central goal of the alignment problem is to adjust the distribution learned by diffusion models such that the generated samples maximize the target reward function. We propose a novel alignment approach, named Direct Noise Optimization (DNO), that optimizes the injected noise during the sampling process of diffusion models. By design, DNO operates at inference-time, and thus is tuning-free and prompt-agnostic, with the alignment occurring in an online fashion during generation. We rigorously study the theoretical properties of DNO and also propose variants to deal with non-differentiable reward functions. Furthermore, we identify that naive implementation of DNO occasionally suffers from the out-of-distribution reward hacking problem, where optimized samples have high rewards but are no longer in the support of the pretrained distribution. To remedy this issue, we leverage classical high-dimensional statistics theory to an effective probability regularization technique. We conduct extensive experiments on several important reward functions and demonstrate that the proposed DNO approach can achieve state-of-the-art reward scores within a reasonable time budget for generation.

Retrieval-Augmented Generation (RAG) has been empirically shown to enhance the performance of large language models (LLMs) in knowledge-intensive domains such as healthcare, finance, and legal contexts. Given a query, RAG retrieves relevant documents from a corpus and integrates them into the LLMs’ generation process. In this study, we investigate the adversarial robustness of RAG, focusing specifically on examining the retrieval system. First, across 225 different setup combinations of corpus, retriever, query, and targeted information, we show that retrieval systems are vulnerable to universal poisoning attacks in medical Q&A. In such attacks, adversaries generate poisoned documents containing a broad spectrum of targeted information, such as personally identifiable information. When these poisoned documents are inserted into a corpus, they can be accurately retrieved by any users, as long as attacker-specified queries are used. To understand this vulnerability, we discovered that the deviation from the query’s embedding to that of the poisoned document tends to follow a pattern in which the high similarity between the poisoned document and the query is retained, thereby enabling precise retrieval. Based on these findings, we develop a new detection-based defense to ensure the safe use of RAG. Through extensive experiments spanning various Q&A domains, we observed that our proposed method consistently achieves excellent detection rates in nearly all cases.

Xiaomeng Jin, Zhiqi Bu, Bhanukiran Vinzamuri, Anil Ramakrishna, Kai-Wei Chang, Volkan Cevher, Mingyi Hong. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

This paper introduces \textsc{InfantAgent-Next}, a generalist agent capable of interacting with computers in a multimodal manner, encompassing text, images, audio, and video. Unlike existing approaches that either build intricate workflows around a single large model or only provide workflow modularity, our agent integrates tool-based and pure vision agents within a highly modular architecture, enabling different models to collaboratively solve decoupled tasks in a step-by-step manner. Our generality is demonstrated by our ability to evaluate not only pure vision-based real-world benchmarks (i.e., OSWorld), but also more general or tool-intensive benchmarks (e.g., GAIA and SWE-Bench). Specifically, we achieve a $\mathbf{7.27\\%}$ accuracy gain over Claude-Computer-Use on OSWorld. Codes and evaluation scripts are included in the supplementary material and will be released as open-source.

Second-order optimization methods, such as cubic regularized Newton methods, are known for their rapid convergence rates; nevertheless, they become impractical in high-dimensional problems due to their substantial memory requirements and computational costs. One promising approach is to execute second-order updates within a lower-dimensional subspace, giving rise to subspace second-order methods. However, the majority of existing subspace second-order methods randomly select subspaces, consequently resulting in slower convergence rates depending on the problem’s dimension $d$. In this paper, we introduce a novel subspace cubic regularized Newton method that achieves a dimension-independent global convergence rate of $\mathcal{O}\left(\frac{1}{mk}+\frac{1}{k^2}\right)$ for solving convex optimization problems. Here, $m$ represents the subspace dimension, which can be significantly smaller than $d$. Instead of adopting a random subspace, our primary innovation involves performing the cubic regularized Newton update within the \emph{Krylov subspace} associated with the Hessian and the gradient of the objective function. This result marks the first instance of a dimension-independent convergence rate for a subspace second-order method. Furthermore, when specific spectral conditions of the Hessian are met, our method recovers the convergence rate of a full-dimensional cubic regularized Newton method. Numerical experiments show our method converges faster than existing random subspace methods, especially for high-dimensional problems.

Decentralized computation has received considerable research interest lately, due to its wide applications in information processing systems. However, one key requirement to establish convergence for almost all decentralized algorithms, for convex and non-convex problems alike, is that the loss function has Lipschitz-continuous gradient (LipGrad). This is a strong assumption, which does not hold for many practical problems, such as matrix/tensor factorization, neural network training, etc. On the contrary, in the centralized setting, one can utilize techniques such as the Bregman proximal gradient (BPG) method to deal with the lack of LipGrad. This work fills the gap between centralized and decentralized cases by developing a novel smoothed decentralized BPG algorithm to deal with a class of nonconvex decentralized problem, where the local problems do not have LipGrad objective functions. By leveraging the recent notion of relative smoothness and primal-dual error bounds, we show that the proposed algorithm achieves a certain ε-stationary solution by using $\mathcal{O}\left( {{\varepsilon ^{ - 2}}} \right)$ iterations, matching the rate of the centralized Bregman proximal gradient method. To our knowledge, this is the first decentralized algorithm that matches the centralized convergence rate bounds under the class of considered problems. Our numerical results on the decentralized quadratic regression example demonstrate the effectiveness of proposed algorithm.

Backdoor attacks pose a significant security risk to machine learning applications due to their stealthy nature and potentially serious consequences. Such attacks involve embedding triggers within a learning model with the intention of causing malicious behavior when an active trigger is present while maintaining regular functionality without it. This paper derives a fundamental understanding of backdoor attacks that applies to both discriminative and generative models, including diffusion models and large language models. We evaluate the effectiveness of any backdoor attack incorporating a constant trigger, by establishing tight lower and upper boundaries for the performance of the compromised model on both clean and backdoor test data. The developed theory answers a series of fundamental but previously underexplored problems, including (1) what are the determining factors for a backdoor attack's success, (2) what is the direction of the most effective backdoor attack, and (3) when will a human-imperceptible trigger succeed. We demonstrate the theory by conducting experiments using benchmark datasets and state-of-the-art backdoor attack scenarios. Our code is available \href{https://github.com/KeyWgh/DemystifyBackdoor}{here}.

Differentially Private Stochastic Gradient Descent with Gradient Clipping (DPSGD-GC) is a powerful tool for training deep learning models using sensitive data, providing both a solid theoretical privacy guarantee and high efficiency. However, existing research has shown that DPSGD-GC only converges when using large clipping thresholds that are dependent on problem-specific parameters that are often unknown in practice. Therefore, DPSGD-GC suffers from degraded performance due to the {\it constant} bias introduced by the clipping. In our work, we propose a new error-feedback (EF) DP algorithm as an alternative to DPSGD-GC, which offers a diminishing utility bound without inducing a constant clipping bias. More importantly, it allows for an arbitrary choice of clipping threshold that is independent of the problem. We establish an algorithm-specific DP analysis for our proposed algorithm, providing privacy guarantees based on R{\'e}nyi DP. And we demonstrate that under mild conditions, our algorithm can achieve the same utility bound as DPSGD without gradient clipping. Our empirical results on standard datasets show that the proposed algorithm achieves higher accuracies than DPSGD while maintaining the same level of DP guarantee.

Following the introduction of Adam, several novel adaptive optimizers for deep learning have been proposed. These optimizers typically excel in some tasks but may not outperform Adam uniformly across all tasks. In this work, we introduce Meta-Adaptive Optimizers (MADA), a unified optimizer framework that can generalize several known optimizers and dynamically learn the most suitable one during training. The key idea in MADA is to parameterize the space of optimizers and dynamically search through it using hyper-gradient descent during training. We empirically compare MADA to other popular optimizers on vision and language tasks, and find that MADA consistently outperforms Adam and other popular optimizers, and is robust against sub-optimally tuned hyper-parameters. MADA achieves a greater validation performance improvement over Adam compared to other popular optimizers during GPT-2 training and fine-tuning. We also propose AVGrad, a modification of AMSGrad that replaces the maximum operator with averaging, which is more suitable for hyper-gradient optimization. Finally, we provide a convergence analysis to show that parameterized interpolations of optimizers can improve their error bounds (up to constants), hinting at an advantage for meta-optimizers.

A key challenge in contrastive learning is to generate negative samples from a large sample set to contrast with positive samples, for learning better encoding of the data. These negative samples often follow a softmax distribution which are dynamically updated during the training process. However, sampling from this distribution is non-trivial due to the high computational costs in computing the partition function. In this paper, we propose an $\underline{\text{E}}$fficient $\underline{\text{M}}$arkov $\underline{\text{C}}$hain Monte Carlo negative sampling method for $\underline{\text{C}}$ontrastive learning (EMC$^2$). We follow the global contrastive learning loss as introduced in SogCLR, and propose EMC$^2$ which utilizes an adaptive Metropolis-Hastings subroutine to generate hardness-aware negative samples in an online fashion during the optimization. We prove that EMC$^2$ finds an $\mathcal{O}(1/\sqrt{T})$-stationary point of the global contrastive loss in $T$ iterations. Compared to prior works, EMC$^2$ is the first algorithm that exhibits global convergence (to stationarity) regardless of the choice of batch size while exhibiting low computation and memory cost. Numerical experiments validate that EMC$^2$ is effective with small batch training and achieves comparable or better performance than baseline algorithms. We report the results for pre-training image encoders on STL-10 and Imagenet-100.

In the evolving landscape of natural language processing (NLP), fine-tuning pre-trained Large Language Models (LLMs) with first-order (FO) optimizers like SGD and Adam has become standard. Yet, as LLMs grow in size, the substantial memory overhead from back-propagation (BP) for FO gradient computation presents a significant challenge. Addressing this issue is crucial, especially for applications like on-device training where memory efficiency is paramount. This paper proposes a shift towards BP-free, zeroth-order (ZO) optimization as a solution for reducing memory costs during LLM fine-tuning, building on the initial concept introduced by (Malladi et al., 2023). Unlike traditional ZO-SGD methods, ou让work expands the exploration to a wider array of ZO optimization techniques, through a comprehensive, first-of-its-kind benchmarking study across five LLM families, three task complexities, and five fine-tuning schemes. Our study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance. We further introduce novel enhancements to ZO optimization, including block-wise descent, hybrid training, and gradient sparsity. Our study offers a promising direction for achieving further memory-efficient LLM fine-tuning. Codes to reproduce all our experiments will be made public.

Privacy is a growing concern in modern deep-learning systems and applications. Differentially private (DP) training prevents the leakage of sensitive information in the collected training data from the trained machine learning models. DP optimizers, including DP stochastic gradient descent (DPSGD) and its variants, privatize the training procedure by gradient clipping and *DP noise* injection. However, in practice, DP models trained using DPSGD and its variants often suffer from significant model performance degradation. Such degradation prevents the application of DP optimization in many key tasks, such as foundation model pretraining. In this paper, we provide a novel *signal processing perspective* to the design and analysis of DP optimizers. We show that a ''frequency domain'' operation called *low-pass filtering* can be used to effectively reduce the impact of DP noise. More specifically, by defining the ''frequency domain'' for both the gradient and differential privacy (DP) noise, we have developed a new component, called DOPPLER. This component is designed for DP algorithms and works by effectively amplifying the gradient while suppressing DP noise within this frequency domain. As a result, it maintains privacy guarantees and enhances the quality of the DP-protected model. Our experiments show that the proposed DP optimizers with a low-pass filter outperform their counterparts without the filter on various models and datasets. Both theoretical and practical evidence suggest that the DOPPLER is effective in closing the gap between DP and non-DP training.

The superior performance of large foundation models can be attributed to the use of massive amounts of high-quality data. However, such datasets often contain sensitive, private and copyrighted material that requires formal protection. While differential privacy (DP) is a prominent method used to gauge the degree of security provided to large foundation models, its application in large foundation models has been met with limited success because there are often significant performance compromises when applying DP during the pre-training phase. Consequently, DP is more commonly implemented during the model fine-tuning stage, hence not capable of protecting a substantial portion of the data used during the initial pre-training process. In this work, we first provide a theoretical understanding of the efficacy of DP training by analyzing the per-iteration improvement of loss through the lens of the Hessian. We observe that DP optimizers' deceleration can be significantly mitigated by the use of limited public data, and thus propose the DP continual pre-training strategy. Our DP continual pre-training on vision models, using only 10% of public data, have achieved DP accuracy of 41.5% on ImageNet-21k (with epsilon=8) and non-DP accuracy of 55.7% on Places365 and 60.0% on iNaturalist-2021, which are on par with state-of-the-art standard pre-training and outperform existing DP pertained models. Our DP pre-trained models are released in *fastDP* library (https://github.com/awslabs/fast-differential-privacy/releases/tag/v2.1)

Large language models (LLMs) have shown impressive capabilities across various tasks. However, training LLMs from scratch requires significant computational power and extensive memory capacity. Recent studies have explored low-rank structures on weights for efficient fine-tuning in terms of parameters and memory, either through low-rank adaptation or factorization. While effective for fine-tuning, low-rank structures are generally less suitable for pretraining because they restrict parameters to a low-dimensional subspace. In this work, we propose to parameterize the weights as a sum of low-rank and sparse matrices for pretraining, which we call SLTrain. The low-rank component is learned via matrix factorization, while for the sparse component, we employ a simple strategy of uniformly selecting the sparsity support at random and learning only the non-zero entries with the fixed support. While being simple, the random fixed-support sparse learning strategy significantly enhances pretraining when combined with low-rank learning. Our results show that SLTrain adds minimal extra parameters and memory costs compared to pretraining with low-rank parameterization, yet achieves substantially better performance, which is comparable to full-rank training. Remarkably, when combined with quantization and per-layer updates, SLTrain can reduce memory requirements by up to 73% when pretraining the LLaMA 7B model.

Aligning human preference and value is an important requirement for contemporary foundation models. State-of-the-art techniques such as Reinforcement Learning from Human Feedback (RLHF) often consist of two stages: 1) supervised fine-tuning (SFT), where the model is fine-tuned by learning from human demonstration data; 2) Preference learning, where preference data is used to learn a reward model, which is in turn used by a reinforcement learning (RL) step to fine-tune the model. Such reward model serves as a proxy to human preference, and it is critical to guide the RL step towards improving the model quality. In this work, we argue that the SFT stage significantly benefits from learning a reward model as well. Instead of using the human demonstration data directly via supervised learning, we propose to leverage an Inverse Reinforcement Learning (IRL) technique to {\it simultaneously} build an reward model and a policy model. This approach leads to new SFT algorithms that are not only efficient to implement, but are robust to the presence of low-quality supervised learning data. Moreover, we discover a connection between the proposed IRL based approach, and a recent line of works called Self-Play Fine-tune (SPIN, \cite{chen2024self}). Theoretically, we show that the proposed algorithms converge to the stationary solutions of the IRL problem. Empirically, we align 1B and 7B models using proposed methods and evaluate them on a reward benchmark model and the HuggingFace Open LLM Leaderboard. The proposed methods show significant performance improvement over existing SFT approaches. Our results indicate that it is beneficial to leverage reward learning throughout the entire alignment process. Our code is available at \url{https://github.com/JasonJiaxiangLi/Reward_learning_SFT}.

While the Transformer architecture has achieved remarkable success across various domains, a thorough theoretical foundation explaining its optimization dynamics is yet to be fully developed. In this study, we aim to bridge this understanding gap by answering the following two core questions: (1) Which types of Transformer architectures allow Gradient Descent (GD) to achieve guaranteed convergence? and (2) Under what initial conditions and architectural specifics does the Transformer achieve rapid convergence during training? By analyzing the loss landscape of a single Transformer layer using Softmax and Gaussian attention kernels, our work provides concrete answers to these questions. Our findings demonstrate that, with appropriate weight initialization, GD can train a Transformer model (with either kernel type) to achieve a global optimal solution, especially when the input embedding dimension is large. Nonetheless, certain scenarios highlight potential pitfalls: training a Transformer using the Softmax attention kernel may sometimes lead to suboptimal local solutions. In contrast, the Gaussian attention kernel exhibits a much favorable behavior. Our empirical study further validate the theoretical findings.

Safeguarding intellectual property and preventing potential misuse of AI-generated images are of paramount importance. This paper introduces a robust and agile plug-and-play watermark detection framework, referred to as RAW. As a departure from existing encoder-decoder methods, which incorporate fixed binary codes as watermarks within latent representations, our approach introduces learnable watermarks directly into the original image data. Subsequently, we employ a classifier that is jointly trained with the watermark to detect the presence of the watermark. The proposed framework is compatible with various generative architectures and supports on-the-fly watermark injection after training. By incorporating state-of-the-art smoothing techniques, we show that the framework also provides provable guarantees regarding the false positive rate for misclassifying a watermarked image, even in the presence of adversarial attacks targeting watermark removal. Experiments on a diverse range of images generated by state-of-the-art diffusion models demonstrate substantially improved watermark encoding speed and watermark detection performance, under adversarial attacks, while maintaining image quality. Our code is publicly available [here](https://github.com/jeremyxianx/RAWatermark).

Diffusion models (DMs) have achieved remarkable success in text-to-image generation, but they also pose safety risks, such as the potential generation of harmful content and copyright violations. The techniques of machine unlearning, also known as concept erasing, have been developed to address these risks. However, these techniques remain vulnerable to adversarial prompt attacks, which can prompt DMs post-unlearning to regenerate undesired images containing concepts (such as nudity) meant to be erased. This work aims to enhance the robustness of concept erasing by integrating the principle of adversarial training (AT) into machine unlearning, resulting in the robust unlearning framework referred to as AdvUnlearn. However, achieving this effectively and efficiently is highly nontrivial. First, we find that a straightforward implementation of AT compromises DMs’ image generation quality post-unlearning. To address this, we develop a utility-retaining regularization on an additional retain set, optimizing the trade-off between concept erasure robustness and model utility in AdvUnlearn. Moreover, we identify the text encoder as a more suitable module for robustification compared to UNet, ensuring unlearning effectiveness. And the acquired text encoder can serve as a plug-and-play robust unlearner for various DM types. Empirically, we perform extensive experiments to demonstrate the robustness advantage of AdvUnlearn across various DM unlearning scenarios, including the erasure of nudity, objects, and style concepts. In addition to robustness, AdvUnlearn also achieves a balanced tradeoff with model utility. To our knowledge, this is the first work to systematically explore robust DM unlearning through AT, setting it apart from existing methods that overlook robustness in concept erasing. Codes are available at https://github.com/OPTML-Group/AdvUnlearn. Warning: This paper contains model outputs that may be offensive in nature.

Mathematical optimization is now widely regarded as an indispensable modeling and solution tool for the design of wireless communications systems. While optimization has played a significant role in the revolutionary progress in wireless communication and networking technologies from 1G to 5G and onto the future 6G, the innovations in wireless technologies have also substantially transformed the nature of the underlying mathematical optimization problems upon which the system designs are based and have sparked significant innovations in the development of methodologies to understand, to analyze, and to solve those problems. In this paper, we provide a comprehensive survey of recent advances in mathematical optimization theory and algorithms for wireless communication system design. We begin by illustrating common features of mathematical optimization problems arising in wireless communication system design. We discuss various scenarios and use cases and their associated mathematical structures from an optimization perspective. We then provide an overview of recently developed optimization techniques in areas ranging from nonconvex optimization, global optimization, and integer programming, to distributed optimization and learning-based optimization. The key to successful solution of mathematical optimization problems is in carefully choosing or developing suitable algorithms (or neural network architectures) that can exploit the underlying problem structure. We conclude the paper by identifying several open research challenges and outlining future research directions.

This work revisits the joint transmit beamforming and antenna selection problem. Existing approaches find approximate solutions to this NP-hard problem via various heuristics, e.g., convex/nonconvex relaxation, greedy method, and (deep) supervised learning. However, optimality (or even feasibility) of these heuristics is not guaranteed. To avoid sub-optimal solutions, an effective branch and bound (B&B) algorithm is proposed. B&B algorithms are ensured to return optimal solutions, but have scalability challenges. In order to enhance efficiency, a graph neural network (GNN)-based classfier is trained with imitation learning to accelerate the B&B algorithm— where the GNN is carefully designed to suit the dynamic nature of wireless communication scenarios. The GNN-based acceleration is shown to provably retain the optimality of B&B with high probability, while substantially reducing the computational burden, under reasonable conditions. Numerical experiments show that our GNN-based method always finds near-optimal and feasible solutions with significantly reduced complexity relative to the plain-vanilla B&B.

In this work, we propose algorithms for solving a class of Bilevel Optimization (BLO) problems, with applications in areas such as signal processing, networking and machine learning. Specifically, we develop a novel barrier-based gradient approximation algorithm that transforms the constrained BLO problem to a problem with only linear equality constraints in the LL task. For the reformulated problem, we compute the implicit gradient and develop a gradient-based scheme, involving only a single gradient descent step and the (approximate) solution of the linearly constrained strongly convex LL task at each iteration. We establish, under certain assumptions, the non-asymptotic convergence guarantees of the proposed method to stationary points. Finally, we perform a number of experiments that show the potential of the proposed algorithm.

This work develops analysis and algorithms for solving a class of bilevel optimization problems where the lower-level (LL) problems have linear constraints. Most of the existing approaches for constrained bilevel problems rely on value function-based approximate reformulations, which suffer from issues such as non-convex and non-differentiable constraints. In contrast, in this work, we develop an implicit gradient-based approach, which is easy to implement, and is suitable for machine learning applications. We first provide an in-depth understanding of the problem, by showing that the implicit objective for such problems is in general non-differentiable. However, if we add some small (linear) perturbation to the LL objective, the resulting implicit objective becomes differentiable almost surely. This key observation opens the door for developing (deterministic and stochastic) gradient-based algorithms similar to the state-of-the-art ones for unconstrained bi-level problems. We show that when the implicit function is assumed to be strongly-convex, convex, and weakly-convex, the resulting algorithms converge with guaranteed rate. Finally, we experimentally corroborate the theoretical findings and evaluate the performance of the proposed framework on numerical and adversarial learning problems.

Federated Learning (FL) is a distributed learning paradigm that allows multiple clients to learn a joint model by utilizing privately held data at each client. Significant research efforts have been devoted to develop advanced algorithms that deal with the situation where the data at individual clients have heterogeneous distributions. In this work, we show that data heterogeneity can be dealt from a different perspective. That is, by utilizing a certain overparameterized multi-layer neural network at each client, even the vanilla FedAvg (a.k.a. the Local SGD) algorithm can accurately optimize the training problem: When each client has a neural network with one wide layer of size $N$ (where $N$ is the number of total training samples), followed by layers of smaller widths, FedAvg converges linearly to a solution that achieves (almost) zero training loss, without requiring any assumptions on the clients’ data distributions. To our knowledge, this is the first work that demonstrates such resilience to data heterogeneity for FedAvg when trained on multi-layer neural networks. Our experiments also confirm that, neural networks of large size can achieve better and more stable performance for FL problems.

A poisoning backdoor attack is a rising security concern for deep learning. This type of attack can result in the backdoored model functioning normally most of the time but exhibiting abnormal behavior when presented with inputs containing the backdoor trigger, making it difficult to detect and prevent. In this work, we propose the adaptability hypothesis to understand when and why a backdoor attack works for general learning models, including deep neural networks, based on the theoretical investigation of classical kernel-based learning models. The adaptability hypothesis postulates that for an effective attack, the effect of incorporating a new dataset on the predictions of the original data points will be small, provided that the original data points are distant from the new dataset. Experiments on benchmark image datasets and state-of-the-art backdoor attacks for deep neural networks are conducted to corroborate the hypothesis. Our finding provides insight into the factors that affect the attack’s effectiveness and has implications for the design of future attacks and defenses.

Backdoor attacks involve inserting poisoned samples during training, resulting in a model containing a hidden backdoor that can trigger specific behaviors without impacting performance on normal samples. These attacks are challenging to detect, as the backdoored model appears normal until activated by the backdoor trigger, rendering them particularly stealthy. In this study, we devise a unified inference-stage detection framework to defend against backdoor attacks. We first rigorously formulate the inference-stage backdoor detection problem, encompassing various existing methods, and discuss several challenges and limitations. We then propose a framework with provable guarantees on the false positive rate or the probability of misclassifying a clean sample. Further, we derive the most powerful detection rule to maximize the detection power, namely the rate of accurately identifying a backdoor sample, given a false positive rate under classical learning scenarios. Based on the theoretically optimal detection rule, we suggest a practical and effective approach for real-world applications based on the latent representations of backdoored deep nets. We extensively evaluate our method on 14 different backdoor attacks using Computer Vision (CV) and Natural Language Processing (NLP) benchmark datasets. The experimental findings align with our theoretical results. We significantly surpass the state-of-the-art methods, e.g., up to 300\% improvement on the detection power as evaluated by AUCROC, over the state-of-the-art defense against advanced adaptive backdoor attacks.

Transformers are central in modern natural language processing and computer vision applications. Despite recent works devoted to reducing the quadratic cost of such models with respect to sequence length, dealing with ultra long sequences (e.g., $>$16K tokens) remains challenging. Applications such as answering questions based on a book or summarizing a scientific article are inefficient or infeasible. Here, we propose to significantly improve the efficiency of Transformers for ultra long sequences, by compressing the sequence into a much smaller representation at each layer. Specifically, by exploiting the fact that in many tasks, only a small subset of special tokens, which we call VIP-tokens, are most relevant to the final prediction, we propose a VIP-token centric compression (VCC) scheme which selectively compresses the sequence based on their impact on approximating the representation of the VIP-tokens. Compared with competitive baselines, our algorithm is not only efficient (achieving more than $3\times$ compute efficiency gain compared to baselines on 4K and 16K lengths), but also offers competitive/better performance on a large number of tasks. Further, we show that our algorithm scales to 128K tokens (or more) while consistently offering accuracy improvement. Code is available at https://github.com/mlpen/VCC.

Offline inverse reinforcement learning (Offline IRL) aims to recover the structure of rewards and environment dynamics that underlie observed actions in a fixed, finite set of demonstrations from an expert agent. Accurate models of expertise in executing a task has applications in safety-sensitive applications such as clinical decision making and autonomous driving. However, the structure of an expert's preferences implicit in observed actions is closely linked to the expert's model of the environment dynamics (i.e. the ``world''). Thus, inaccurate models of the world obtained from finite data with limited coverage could compound inaccuracy in estimated rewards. To address this issue, we propose a bi-level optimization formulation of the estimation task wherein the upper level is likelihood maximization based upon a conservative model of the expert's policy (lower level). The policy model is conservative in that it maximizes reward subject to a penalty that is increasing in the uncertainty of the estimated model of the world. We propose a new algorithmic framework to solve the bi-level optimization problem formulation and provide statistical and computational guarantees of performance for the associated optimal reward estimator. Finally, we demonstrate that the proposed algorithm outperforms the state-of-the-art offline IRL and imitation learning benchmarks by a large margin, over the continuous control tasks in MuJoCo and different datasets in the D4RL benchmark.

Massive data is often considered essential for deep learning applications, but it also incurs significant computational and infrastructural costs. Therefore, dataset pruning (DP) has emerged as an effective way to improve data efficiency by identifying and removing redundant training samples without sacrificing performance. In this work, we aim to address the problem of DP for transfer learning, i.e., how to prune a source dataset for improved pretraining efficiency and lossless finetuning accuracy on downstream target tasks. To our best knowledge, the problem of DP for transfer learning remains open, as previous studies have primarily addressed DP and transfer learning as separate problems. By contrast, we establish a unified viewpoint to integrate DP with transfer learning and find that existing DP methods are not suitable for the transfer learning paradigm. We then propose two new DP methods, label mapping and feature mapping, for supervised and self-supervised pretraining settings respectively, by revisiting the DP problem through the lens of source-target domain mapping. Furthermore, we demonstrate the effectiveness of our approach on numerous transfer learning tasks. We show that source data classes can be pruned by up to $40\%\sim 80\%$ without sacrificing the downstream performance, resulting in a significant $2\sim 5\times$ speed-up during the pretraining stage. Besides, our proposal exhibits broad applicability and can improve other computationally intensive transfer learning techniques, such as adversarial pretraining.

Finding the optimal beamformers in massive multiple-input multiple-output (MIMO) networks is challenging because of its non-convexity, and conventional optimization based algorithms suffer from high computational costs. Recently, deep learning based methods have been proposed because of their computational efficiency, but they typically can not generalize well when deployed in heterogeneous scenarios where the base stations (BSs) are equipped with different numbers of antennas and have different inter-BS distances. This paper proposes a novel deep learning based beamforming algorithm to address above challenges. Specifically, we consider the weighted sum rate (WSR) maximization problem in multi-input and single-output (MISO) interference channels, and propose a beamforming learning architecture by unfolding a parallel gradient projection algorithm. By leveraging the low-dimensional structures of the optimal beamforming solution, our constructed learning network can be made independent of the numbers of transmit antennas and BSs. Moreover, such a design can be further extended to a cooperative multicell network where users are jointly served by multiple BSs. Numerical results based on both synthetic and ray-tracing channel models show that the proposed neural network can achieve high WSRs with significantly reduced runtime, while exhibiting favorable generalization capability with respect to the antenna number, BS number and the inter-BS distance.

In this work, we develop an implicit gradient-type (IG-AL) algorithm for bilevel optimization with strongly convex linear inequality constrained lower-level problems. Many learning problems of interest, including problems in distributed optimization, machine learning, economics, and transport research are captured by the above formulation. The key characteristics of the proposed algorithm are: (i) the use of a primal-dual augmented Lagrangian method for solving the lower-level problem, and (ii) construction of an implicit gradient (derived using the KKT conditions of the lower-level problem) for solving the upper-level problem. Importantly, the proposed algorithm avoids the (expensive) projection step to a half-space inherent to gradient descent-based alternatives. The performance of the proposed algorithm is evaluated on a set of numerical experiments.

Supervised learning scenarios, where labels and features are possibly mismatched, have been an emerging concern in machine learning applications. For example, researchers often need to align heterogeneous data from multiple resources to the same entities without a unique identifier in the socioeconomic study. Such a mismatch problem can significantly affect the learning performance if it is not appropriately addressed. Due to the combinatorial nature of the mismatch problem, existing methods are often designed for small datasets and simple linear models but are not scalable to large-scale datasets and complex models. In this paper, we first present a new formulation of the mismatch problem that supports continuous optimization problems and allows for gradient-based methods. Moreover, we develop a computation and memory efficient method to process complex data and models. Empirical studies on synthetic and real-world data show significantly better performance of the proposed algorithms than state-of-the-art methods.

Providing privacy protection has been one of the primary motivations of Federated Learning (FL). Recently, there has been a line of work on incorporating the formal privacy notion of differential privacy with FL. To guarantee the client-level differential privacy in FL algorithms, the clients’ transmitted model updates have to be clipped before adding privacy noise. Such clipping operation is substantially different from its counterpart of gradient clipping in the centralized differentially private SGD and has not been well-understood. In this paper, we first empirically demonstrate that the clipped FedAvg can perform surprisingly well even with substantial data heterogeneity when training neural networks, which is partly because the clients’ updates become similar for several popular deep architectures. Based on this key observation, we provide the convergence analysis of a differential private (DP) FedAvg algorithm and highlight the relationship between clipping bias and the distribution of the clients’ updates. To the best of our knowledge, this is the first work that rigorously investigates theoretical and empirical issues regarding the clipping operation in FL algorithms.

In modern machine learning systems, distributed algorithms are deployed across applications to ensure data privacy and optimal utilization of computational resources. This work offers a fresh perspective to model, analyze, and design distributed optimization algorithms through the lens of stochastic multi-rate feedback control. We show that a substantial class of distributed algorithms—including popular Gradient Tracking for decentralized learning, and FedPD and Scaffold for federated learning—can be modeled as a certain discrete-time stochastic feedback-control system, possibly with multiple sampling rates. This key observation allows us to develop a generic framework to analyze the convergence of the entire algorithm class. It also enables one to easily add desirable features such as differential privacy guarantees, or to deal with practical settings such as partial agent participation, communication compression, and imperfect communication in algorithm design and analysis.

Adversarial training (AT) is a widely recognized defense mechanism to gain the robustness of deep neural networks against adversarial attacks. It is built on min-max optimization (MMO), where the minimizer (i.e., defender) seeks a robust model to minimize the worst-case training loss in the presence of adversarial examples crafted by the maximizer (i.e., attacker). However, the conventional MMO method makes AT hard to scale. Thus, Fast-AT and other recent algorithms attempt to simplify MMO by replacing its maximization step with the single gradient sign-based attack generation step. Although easy to implement, FAST-AT lacks theoretical guarantees, and its empirical performance is unsatisfactory due to the issue of robust catastrophic overfitting when training with strong adversaries. In this paper, we advance Fast-AT from the fresh perspective of bi-level optimization (BLO). We first show that the commonly-used Fast-AT is equivalent to using a stochastic gradient algorithm to solve a linearized BLO problem involving a sign operation. However, the discrete nature of the sign operation makes it difficult to understand the algorithm performance. Inspired by BLO, we design and analyze a new set of robust training algorithms termed Fast Bi-level AT (Fast-BAT), which effectively defends sign-based projected gradient descent (PGD) attacks without using any gradient sign method or explicit robust regularization. In practice, we show that our method yields substantial robustness improvements over multiple baselines across multiple models and datasets.

To regulate a social system comprised of self-interested agents, economic incentives are often required to induce a desirable outcome. This incentive design problem naturally possesses a bilevel structure, in which a designer modifies the payoffs of the agents with incentives while anticipating the response of the agents, who play a non-cooperative game that converges to an equilibrium. The existing bilevel optimization algorithms raise a dilemma when applied to this problem: anticipating how incentives affect the agents at equilibrium requires solving the equilibrium problem repeatedly, which is computationally inefficient; bypassing the time-consuming step of equilibrium-finding can reduce the computational cost, but may lead the designer to a sub-optimal solution. To address such a dilemma, we propose a method that tackles the designer’s and agents’ problems simultaneously in a single loop. Specifically, at each iteration, both the designer and the agents only move one step. Nevertheless, we allow the designer to gradually learn the overall influence of the incentives on the agents, which guarantees optimality after convergence. The convergence rate of the proposed scheme is also established for a broad class of games.

Bilevel optimization has been shown to be a powerful framework for formulating multi-task machine learning problems, e.g., reinforcement learning (RL) and meta-learning, where the decision variables are coupled in both levels of the minimization problems. In practice, the learning tasks would be located at different computing resource environments, and thus there is a need for deploying a decentralized training framework to implement multi-agent and multi-task learning. We develop a stochastic linearized augmented Lagrangian method (SLAM) for solving general nonconvex bilevel optimization problems over a graph, where both upper and lower optimization variables are able to achieve a consensus. We also establish that the theoretical convergence rate of the proposed SLAM to the Karush-Kuhn-Tucker (KKT) points of this class of problems is on the same order as the one achieved by the classical distributed stochastic gradient descent for only single-level nonconvex minimization problems. Numerical results tested on multi-agent RL problems showcase the superiority of SLAM compared with the benchmarks.

Decentralized optimization are playing an important role in applications such as training large machine learning models, among others. Despite its superior practical performance, there has been some lack of fundamental understanding about its theoretical properties. In this work, we address the following open research question: To train an overparameterized model over a set of distributed nodes, what is the {\it minimum} communication overhead (in terms of the bits got exchanged) that the system needs to sustain, while still achieving (near) zero training loss? We show that for a class of overparameterized models where the number of parameters $D$ is much larger than the total data samples $N$, the best possible communication complexity is ${\Omega}(N)$, which is independent of the problem dimension $D$. Further, for a few specific overparameterized models (i.e., the linear regression, and certain multi-layer neural network with one wide layer), we develop a set of algorithms which uses certain linear compression followed by adaptive quantization, and show that they achieve dimension independent, and sometimes near optimal, communication complexity. To our knowledge, this is the first time that dimension independent communication complexity has been shown for distributed optimization.

Inverse reinforcement learning (IRL) aims to recover the reward function and the associated optimal policy that best fits observed sequences of states and actions implemented by an expert. Many algorithms for IRL have an inherent nested structure: the inner loop finds the optimal policy given parametrized rewards while the outer loop updates the estimates towards optimizing a measure of fit. For high dimensional environments such nested-loop structure entails a significant computational burden. To reduce the computational burden of a nested loop, novel methods such as SQIL \cite{reddy2019sqil} and IQ-Learn \cite{garg2021iq} emphasize policy estimation at the expense of reward estimation accuracy. However, without accurate estimated rewards, it is not possible to do counterfactual analysis such as predicting the optimal policy under different environment dynamics and/or learning new tasks. In this paper we develop a novel {\em single-loop} algorithm for IRL that does not compromise reward estimation accuracy. In the proposed algorithm, each policy improvement step is followed by a stochastic gradient step for likelihood maximization. We show that the proposed algorithm provably converges to a stationary solution with a finite-time guarantee. If the reward is parameterized linearly we show the identified solution corresponds to the solution of the maximum entropy IRL problem. Finally, by using robotics control problems in Mujoco and their transfer settings, we show that the proposed algorithm achieves superior performance compared with other IRL and imitation learning benchmarks.

The deployment constraints in practical applications necessitate the pruning of large-scale deep learning models, i.e., promoting their weight sparsity. As illustrated by the Lottery Ticket Hypothesis (LTH), pruning also has the potential of improving their generalization ability. At the core of LTH, iterative magnitude pruning (IMP) is the predominant pruning method to successfully find ‘winning tickets’. Yet, the computation cost of IMP grows prohibitively as the targeted pruning ratio increases. To reduce the computation overhead, various efficient ‘one-shot’ pruning methods have been developed, but these schemes are usually unable to find winning tickets as good as IMP. This raises the question of how to close the gap between pruning accuracy and pruning efficiency? To tackle it, we pursue the algorithmic advancement of model pruning. Specifically, we formulate the pruning problem from a fresh and novel viewpoint, bi-level optimization (BLO). We show that the BLO interpretation provides a technically-grounded optimization base for an efficient implementation of the pruning-retraining learning paradigm used in IMP. We also show that the proposed bi-level optimization-oriented pruning method (termed BiP) is a special class of BLO problems with a bi-linear problem structure. By leveraging such bi-linearity, we theoretically show that BiP can be solved as easily as first-order optimization, thus inheriting the computation efficiency. Through extensive experiments on both structured and unstructured pruning with 5 model architectures and 4 data sets, we demonstrate that BiP can find better winning tickets than IMP in most cases, and is computationally as efficient as the one-shot pruning schemes, demonstrating $2-7\times$ speedup over IMP for the same level of model accuracy and sparsity.

Current deep neural networks (DNNs) are vulnerable to adversarial attacks, where adversarial perturbations to the inputs can change or manipulate classification. To defend against such attacks, an effective and popular approach, known as adversarial training (AT), has been shown to mitigate the negative impact of adversarial attacks by virtue of a min-max robust training method. While effective, it remains unclear whether it can successfully be adapted to the distributed learning context. The power of distributed optimization over multiple machines enables us to scale up robust training over large models and datasets. Spurred by that, we propose distributed adversarial training (DAT), a large-batch adversarial training framework implemented over multiple machines. We show that DAT is general, which supports training over labeled and unlabeled data, multiple types of attack generation methods, and gradient compression operations favored for distributed optimization. Theoretically, we provide, under standard conditions in the optimization theory, the convergence rate of DAT to the first-order stationary points in general non-convex settings. Empirically, we demonstrate that DAT either matches or outperforms state-of-the-art robust accuracies and achieves a graceful training speedup (e.g., on ResNet-50 under ImageNet). Codes are available at https://github.com/dat-2022/dat.

Efficiently finding First-order Nash Equilibria (FNE) in zero-sum games can be challenging, even in a two-player setting. This work proposes an algorithm for finding the FNEs of a two-player zero-sum game, in which the local cost functions can be non-convex, and the players only have access to local stochastic gradients. The proposed approach is based on reformulating the problem of interest as minimizing the Regularized Nikaido-Isoda (RNI) function. We show that the global minima of the RNI correspond to the set of FNEs, and that for certain classes of non-convex games the RNI minimization problem becomes convex. Moreover, we introduce a first-order (stochastic) optimization method, and establish its convergence to a neighborhood of a stationary solution of the RNI objective. The key in the analysis is to properly control the bias between the local stochastic gradient and the true one. Although the RNI function has been used in analyzing convex games, to our knowledge, this is the first time that the properties of the RNI formulation have been exploited to find FNEs for non-convex games in a stochastic setting.

This paper studies the generalization bounds for the empirical saddle point (ESP) solution to stochastic saddle point (SSP) problems. For SSP with Lipschitz continuous and strongly convex-strongly concave objective functions, we establish an $O\left(1/n\right)$ generalization bound by using a probabilistic stability argument. We also provide generalization bounds under a variety of assumptions, including the cases without strong convexity and without bounded domains. We illustrate our results in three examples: batch policy learning in Markov decision process, stochastic composite optimization problem, and mixed strategy Nash equilibrium estimation for stochastic games. In each of these examples, we show that a regularized ESP solution enjoys a near-optimal sample complexity. To the best of our knowledge, this is the first set of results on the generalization theory of ESP.

Canonical polyadic decomposition (CPD) has been a workhorse for multimodal data analytics. This work puts forth a stochastic algorithmic framework for CPD under β-divergence, which is well-motivated in statistical learning—where the Euclidean distance is typically not preferred. Despite the existence of a series of prior works addressing this topic, pressing computational and theoretical challenges, e.g., scalability and convergence issues, still remain. In this paper, a unified stochastic mirror descent framework is developed for large-scale β-divergence CPD. Our key contribution is the integrated design of a tensor fiber sampling strategy and a flexible stochastic Bregman divergence-based mirror descent iterative procedure, which significantly reduces the computation and memory cost per iteration for various β. Leveraging the fiber sampling scheme and the multilinear algebraic structure of low-rank tensors, the proposed lightweight algorithm also ensures global convergence to a stationary point under mild conditions. Numerical results on synthetic and real data show that our framework attains significant computational saving compared with state-of-the-art methods.

Spectrum cartography (SC) aims at estimating the multi-aspect (e.g., space, frequency, and time) interference level caused by multiple emitters from limited measurements. Early SC approaches rely on model assumptions about the radio map, e.g., sparsity and smoothness, which may be grossly violated under critical scenarios, e.g., in the presence of severe shadowing. More recent data-driven methods train deep generative networks to distill parsimonious representations of complex scenarios, in order to enhance performance of SC. The challenge is that the state space of this learning problem is extremely large—induced by different combinations of key problem constituents, e.g., the number of emitters, the emitters’ carrier frequencies, and the emitter locations. Learning over such a huge space can be costly in terms of sample complexity and training time; it also frequently leads to generalization problems. Our method integrates the favorable traits of model and data-driven approaches, which substantially ‘shrinks’ the state space. Specifically, the proposed learning paradigm only needs to learn a generative model for the radio map of a single emitter (as opposed to numerous combinations of multiple emitters), leveraging a nonnegative matrix factorization (NMF)-based emitter disaggregation process. Numerical evidence shows that the proposed method outperforms state-of-the-art purely model-driven and purely data-driven approaches.

There has been a growing interest in developing data-driven, in particular deep neural network (DNN) based methods for modern communication tasks. For a few popular tasks such as power control, beamforming, and MIMO detection, these methods achieve state-of-the-art performance while requiring less computational efforts, less channel state information (CSI), etc. However, it is often challenging for these approaches to learn in a dynamic environment where parameters such as CSIs keep changing.This work develops a methodology that enables data-driven methods to continuously learn and optimize in a dynamic environment. Specifically, we consider an "episodically dynamic" setting where the environment changes in "episodes", and in each episode the environment is stationary. We propose a continual learning (CL) framework for wireless systems, which can incrementally adapt the learning models to the new episodes, without forgetting models learned from the previous episodes. Our design is based on a novel min-max formulation which ensures certain "fairness" across different episodes. Finally, we demonstrate the effectiveness of the CL approach by customizing it to a popular DNN based model for power control, and testing using both synthetic and real data.

We consider a distributed non-convex optimization where a network of agents aims at minimizing a global function over the Stiefel manifold. The global function is represented as a finite sum of smooth local functions, where each local function is associated with one agent and agents communicate with each other over an undirected connected graph. The problem is non-convex as local functions are possibly non-convex (but smooth) and the Steifel manifold is a non-convex set. We present a decentralized Riemannian stochastic gradient method (DRSGD) with the convergence rate of $\mathcal{O}(1/\sqrt{K})$ to a stationary point. To have exact convergence with constant stepsize, we also propose a decentralized Riemannian gradient tracking algorithm (DRGTA) with the convergence rate of $\mathcal{O}(1/K)$ to a stationary point. We use multi-step consensus to preserve the iteration in the local (consensus) region. DRGTA is the first decentralized algorithm with exact convergence for distributed optimization on Stiefel manifold.

Federated Learning (FL) refers to the paradigm where multiple worker nodes (WNs) build a joint model by using local data. Despite extensive research, for a generic non-convex FL problem, it is not clear, how to choose the WNs' and the server's update directions, the minibatch sizes, and the local update frequency, so that the WNs use the minimum number of samples and communication rounds to achieve the desired solution. This work addresses the above question and considers a class of stochastic algorithms where the WNs perform a few local updates before communication. We show that when both the WN's and the server's directions are chosen based on certain stochastic momentum estimator, the algorithm requires $\tilde{\mathcal{O}}(\epsilon^{-3/2})$ samples and $\tilde{\mathcal{O}}(\epsilon^{-1})$ communication rounds to compute an $\epsilon$-stationary solution. To the best of our knowledge, this is the first FL algorithm that achieves such {\it near-optimal} sample and communication complexities simultaneously. Further, we show that there is a trade-off curve between local update frequencies and local minibatch sizes, on which the above sample and communication complexities can be maintained. {Finally, we show that for the classical FedAvg (a.k.a. Local SGD, which is a momentum-less special case of the STEM), a similar trade-off curve exists, albeit with worse sample and communication complexities. Our insights on this trade-off provides guidelines for choosing the four important design elements for FL algorithms, the update frequency, directions, and minibatch sizes to achieve the best performance.}

This paper proposes a new algorithm -- the \underline{S}ingle-timescale Do\underline{u}ble-momentum \underline{St}ochastic \underline{A}pprox\underline{i}matio\underline{n} (SUSTAIN) -- for tackling stochastic unconstrained bilevel optimization problems. We focus on bilevel problems where the lower level subproblem is strongly-convex and the upper level objective function is smooth. Unlike prior works which rely on \emph{two-timescale} or \emph{double loop} techniques, we design a stochastic momentum-assisted gradient estimator for both the upper and lower level updates. The latter allows us to control the error in the stochastic gradient updates due to inaccurate solution to both subproblems. If the upper objective function is smooth but possibly non-convex, we show that {SUSTAIN}~requires $O(\epsilon^{-3/2})$ iterations (each using $O(1)$ samples) to find an $\epsilon$-stationary solution. The $\epsilon$-stationary solution is defined as the point whose squared norm of the gradient of the outer function is less than or equal to $\epsilon$. The total number of stochastic gradient samples required for the upper and lower level objective functions matches the best-known complexity for single-level stochastic gradient algorithms. We also analyze the case when the upper level objective function is strongly-convex.

Modern neural networks are often quite wide, causing large memory and computation costs. It is thus of great interest to train a narrower network. However, training narrow neural nets remains a challenging task. We ask two theoretical questions: Can narrow networks have as strong expressivity as wide ones? If so, does the loss function exhibit a benign optimization landscape? In this work, we provide partially affirmative answers to both questions for 1-hidden-layer networks with fewer than $n$ (sample size) neurons when the activation is smooth. First, we prove that as long as the width $m \geq 2n/d$ (where $d$ is the input dimension), its expressivity is strong, i.e., there exists at least one global minimizer with zero training loss. Second, we identify a nice local region with no local-min or saddle points. Nevertheless, it is not clear whether gradient descent can stay in this nice region. Third, we consider a constrained optimization formulation where the feasible region is the nice local region, and prove that every KKT point is a nearly global minimizer. It is expected that projected gradient methods converge to KKT points under mild technical conditions, but we leave the rigorous convergence analysis to future work. Thorough numerical results show that projected gradient methods on this constrained formulation significantly outperform SGD for training narrow neural nets.

In this paper, we investigate the adaptive video delivery for multiple users over time-varying and mutually interfering multi-cell wireless networks. The key research challenge is to jointly design the physical-layer resource allocation scheme and application-layer rate adaptation logic, such that the users' long-term fair quality of experience (QoE) can be maximized. Due to the timescale mismatch between these two layers and the asynchrony of user requests, however, it is difficult to directly model the cross-layer stochastic control problem by using a reinforcement learning framework. To address this difficulty, we propose a novel two-level decision framework where an optimization-based beamforming scheme (performed at the base stations) and a deep reinforcement learning (DRL)-based rate adaptation scheme (performed at the user terminals) are, respectively, developed, such that a highly complex long-term multi-user QoE fairness problem is decomposed into some relatively simple problems and solved effectively. Our strategy represents a significant departure from the existing schemes with consideration of either a short-term multi-user QoE maximization or a long-term single-user point-to-point QoE maximization. Extensive simulations demonstrate that the proposed cross-layer design is effective and promising.

This paper discusses formulations and algorithms which allow a number of agents to collectively solve problems involving both (non-convex) minimization and (concave) maximization operations. These problems have a number of interesting applications in information processing and machine learning, and in particular can be used to model an adversary learning problem called network data poisoning. We develop a number of algorithms to efficiently solve these non-convex min-max optimization problems, by combining techniques such as gradient tracking in the decentralized optimization literature and gradient descent-ascent schemes in the min-max optimization literature. Also, we establish convergence to a first order stationary point under certain conditions. Finally, we perform experiments to demonstrate that the proposed algorithms are effective in the data poisoning attack.

In this work, we consider the use of model-driven deep learning (DL) techniques for signal detection in massive multiple-input multiple-output (MIMO) system. Massive MIMO promises improved spectral efficiency, coverage and reliability, compared to conventional MIMO systems. Unfortunately, these benefits usually come at the cost of significantly increased computational complexity. To address this difficulty, a learned conjugate gradient descent network, referred to as LcgNet, is presented by unfolding the iterative conjugate gradient descent (CG) detector. In the proposed network, instead of calculating the exact values of the scalar step-sizes for every problem instance, we explicitly learn their universal values. We show that the performance of the proposed network can be greatly improved by augmenting the dimensions of these step-sizes. Furthermore, due to the limited learnable parameters to be optimized, the proposed networks are easy and fast to train. Numerical results demonstrate that this approach can achieve superior performance over some state-of-the-art MIMO detectors such as the CG detector, the linear minimum mean squared error (LMMSE) detector etc., with much lower computational complexity.

In this paper, we study the problem of constrained min-max optimization in a black-box setting, where the desired optimizer cannot access the gradients of the objective function but may query its values. We present a principled optimization framework, integrating a zeroth-order (ZO) gradient estimator with an alternating projected stochastic gradient descent-ascent method, where the former only requires a small number of function queries and the later needs just one-step descent/ascent update. We show that the proposed framework, referred to as ZO-Min-Max, has a sublinear convergence rate under mild conditions and scales gracefully with problem size. We also explore a promising connection between black-box min-max optimization and black-box evasion and poisoning attacks in adversarial machine learning (ML). Our empirical evaluations on these use cases demonstrate the effectiveness of our approach and its scalability to dimensions that prohibit using recent black-box solvers.

Many modern large-scale machine learning problems benefit from decentralized and stochastic optimization. Recent works have shown that utilizing both decentralized computing and local stochastic gradient estimates can outperform state-of-the-art centralized algorithms, in applications involving highly non-convex problems, such as training deep neural networks. In this work, we propose a decentralized stochastic algorithm to deal with certain smooth non-convex problems where there are $m$ nodes in the system, and each node has a large number of samples (denoted as $n$). Differently from the majority of the existing decentralized learning algorithms for either stochastic or finite-sum problems, our focus is given to \emph{both} reducing the total communication rounds among the nodes, while accessing the minimum number of local data samples. In particular, we propose an algorithm named D-GET (decentralized gradient estimation and tracking), which jointly performs decentralized gradient estimation (which estimates the local gradient using a subset of local samples) \emph{and} gradient tracking (which tracks the global full gradient using local estimates). We show that to achieve certain $\epsilon$ stationary solution of the deterministic finite sum problem, the proposed algorithm achieves an $\mathcal{O}(mn^{1/2}\epsilon^{-1})$ sample complexity and an $\mathcal{O}(\epsilon^{-1})$ communication complexity. These bounds significantly improve upon the best existing bounds of $\mathcal{O}(mn\epsilon^{-1})$ and $\mathcal{O}(\epsilon^{-1})$, respectively. Similarly, for online problems, the proposed method achieves an $\mathcal{O}(m \epsilon^{-3/2})$ sample complexity and an $\mathcal{O}(\epsilon^{-1})$ communication complexity.

Recently, there is a growing interest in the study of median-based algorithms for distributed non-convex optimization. Two prominent examples include signSGD with majority vote, an effective approach for communication reduction via 1-bit compression on the local gradients, and medianSGD, an algorithm recently proposed to ensure robustness against Byzantine workers. The convergence analyses for these algorithms critically rely on the assumption that all the distributed data are drawn iid from the same distribution. However, in applications such as Federated Learning, the data across different nodes or machines can be inherently heterogeneous, which violates such an iid assumption. This work analyzes signSGD and medianSGD in distributed settings with heterogeneous data. We show that these algorithms are non-convergent whenever there is some disparity between the expected median and mean over the local gradients. To overcome this gap, we provide a novel gradient correction mechanism that perturbs the local gradients with noise, which we show can provably close the gap between mean and median of the gradients. The proposed methods largely preserve nice properties of these median-based algorithms, such as the low per-iteration communication complexity of signSGD, and further enjoy global convergence to stationary solutions. Our perturbation technique can be of independent interest when one wishes to estimate mean through a median estimator.

Deep learning models are increasingly popular in many machine learning applications where the training data may contain sensitive information. To provide formal and rigorous privacy guarantee, many learning systems now incorporate differential privacy by training their models with (differentially) private SGD. A key step in each private SGD update is gradient clipping that shrinks the gradient of an individual example whenever its l2 norm exceeds a certain threshold. We first demonstrate how gradient clipping can prevent SGD from converging to a stationary point. We then provide a theoretical analysis on private SGD with gradient clipping. Our analysis fully characterizes the clipping bias on the gradient norm, which can be upper bounded by the Wasserstein distance between the gradient distribution and a geometrically symmetric distribution. Our empirical evaluation further suggests that the gradient distributions along the trajectory of private SGD indeed exhibit such symmetric structure. Together, our results provide an explanation why private SGD with gradient clipping remains effective in practice despite its potential clipping bias. Finally, we develop a new perturbation-based technique that can provably correct the clipping bias even for instances with highly asymmetric gradient distributions.

This paper proposes two efficient algorithms for computing approximate second-order stationary points (SOSPs) of problems with generic smooth non-convex objective functions and generic linear constraints. While finding (approximate) SOSPs for the class of smooth non-convex linearly constrained problems is computationally intractable, we show that generic problem instances in this class can be solved efficiently. Specifically, for a generic problem instance, we show that certain strict complementarity (SC) condition holds for all Karush-Kuhn-Tucker (KKT) solutions. Based on this condition, we design an algorithm named Successive Negative-curvature grAdient Projection (SNAP), which performs either conventional gradient projection or some negative curvature-based projection steps to find SOSPs. SNAP is a second-order algorithm that requires $\widetilde{\mathcal{O}}(\max\{1/\epsilon^2_G,1/\epsilon^3_H\})$ iterations to compute an $(\epsilon_G,\epsilon_H)$-SOSP, where $\widetilde{\mathcal{O}}$ hides the iteration complexity for eigenvalue-decomposition. Building on SNAP, we propose a first-order algorithm, named SNAP$^+$, that requires $\mathcal{O}(1/\epsilon^{2.5})$ iterations to compute $(\epsilon, \sqrt{\epsilon})$-SOSP. The per-iteration computational complexities of our algorithms are polynomial in the number of constraints and problem dimension. To the best of our knowledge, this is the first time that first-order algorithms with polynomial per-iteration complexity and global sublinear rate are designed to find SOSPs of the important class of non-convex problems with linear constraints (almost surely).

This paper studies a gradient temporal difference (GTD) algorithm using neural network (NN) function approximators to minimize the mean squared Bellman error (MSBE). For off-policy learning, we show that the minimum MSBE problem can be recast into a min-max optimization involving a pair of over-parameterized primal-dual NNs. The resultant formulation can then be tackled using a neural GTD algorithm. We analyze the convergence of the proposed algorithm with a 2-layer ReLU NN architecture using $m$ neurons and prove that it computes an approximate optimal solution to the minimum MSBE problem as $m \rightarrow \infty$.

Accurate Time of Arrival (TOA) estimation has many use cases, including 5G initial access and localization. However, due to multipath propagation and noise, the correlation-based TOA estimation may not be accurate. In this paper, a deep learning based framework is proposed for preamble detection and TOA estimation without the need of knowing the transmit waveform. Extensive simulations on both synthetic data and real measured data show that the proposed method improves prediction accuracy by about three times while keeping the same computational complexity in comparison to the correlation method. It also provides 1000x computational reduction compared to the template matching method without loss of accuracy.

In this paper, we use the perturbed gradient based alternating minimization for solving a class of low-rank matrix factorization problems. Alternating minimization is a simple but popular approach which has been applied to problems in optimization, machine learning, data mining, and signal processing, etc. By leveraging the block structure of the problem, the algorithm updates two blocks of variables in an alternating manner. For the nonconvex optimization problem, it is well-known the alternating minimization algorithm converges to the first-order stationary solution with a global sublinear rate. In this paper, a perturbed alternating proximal point (PA-PP) algorithm is proposed, which 1) minimizes the smooth nonconvex problem by updating two blocks of variables alternatively and 2) adds some random noise occasionally under some conditions to extract the negative curvature of the second-order information of the objective function. We show that the proposed PA-PP is able to converge (with high probability) to the set of second-order stationary solutions (SS2) with a global sublinear rate, and as a consequence quickly finds global optimal solutions for the problems considered.

The min-max problem, also known as the saddle point problem, can be used to formulate a wide range of applications in signal processing and wireless communications. However, existing optimization theory and methods, which mostly deal with problems with certain convex-concave structure, are not applicable for the aforementioned applications, which oftentimes involve non-convexity. In this work, we consider a general block-wise one-sided non-convex min-max problem, in which the minimization problem consists of multiple blocks and is non-convex, while the maximization problem is (strongly) concave. We propose two simple algorithms, which alternatingly perform one gradient descent-type step for each minimization block and one gradient ascent-type step for the maximization problem. For the first time, we show that such simple alternating min-max algorithms converge to first-order stationary solutions. We conduct numerical tests on a robust learning problem, and a wireless communication problem in the presence of jammers, to validate the efficiency of the proposed algorithms.

In this paper, a gradient-based method for bound constrained non-convex problems is proposed. By leveraging both projected gradient descent and perturbed gradient descent, the proposed algorithm, named perturbed projected gradient descent (PP-GD), converges to some approximate second-order stationary (SS2) points (which satisfy certain approximate second-order necessary conditions) with provable convergence rate guarantees. The proposed algorithm is suitable for a large-scale problem since it only uses the gradient information of the objective function. It also seamlessly incorporates variable constraints such as nonnegativity, which is commonly seen in many practical machine learning problems. We provide a concrete theoretical analysis showing that PP-GD is able to obtain approximate second-order solutions by extracting the negative curvature of the objective function around the strict saddle points. Numerical results demonstrate that PP-GD indeed converges faster compared to other first-order methods in the presence of strict saddle points.

Alternating gradient descent (A-GD) is a simple but popular algorithm in machine learning, which updates two blocks of variables in an alternating manner using gradient descent steps. In this paper, we consider a smooth unconstrained nonconvex optimization problem, and propose a perturbed A-GD (PA-GD) which is able to converge (with high probability) to the second-order stationary points (SOSPs) with a global sublinear rate. Existing analysis on A-GD type algorithm either only guarantees convergence to first-order solutions, or converges to second-order solutions asymptotically (without rates). To the best of our knowledge, this is the first alternating type algorithm that takes $\mathcal{O}(\text{polylog}(d)/\epsilon^2)$ iterations to achieve an ($\epsilon,\sqrt{\epsilon}$)-SOSP with high probability, where polylog$(d)$ denotes the polynomial of the logarithm with respect to problem dimension $d$.

Graph neural networks (GNNs) which apply the deep neural networks to graph data have achieved significant performance for the task of semi-supervised node classification. However, only few work has addressed the adversarial robustness of GNNs. In this paper, we first present a novel gradient-based attack method that facilitates the difficulty of tackling discrete graph data. When comparing to current adversarial attacks on GNNs, the results show that by only perturbing a small number of edge perturbations, including addition and deletion, our optimization-based attack can lead to a noticeable decrease in classification performance. Moreover, leveraging our gradient-based attack, we propose the first optimization-based adversarial training for GNNs. Our method yields higher robustness against both different gradient based and greedy attack methods without sacrifice classification accuracy on original graph.

The adaptive momentum method (AdaMM), which uses past gradients to update descent directions and learning rates simultaneously, has become one of the most popular first-order optimization methods for solving machine learning problems. However, AdaMM is not suited for solving black-box optimization problems, where explicit gradient forms are difficult or infeasible to obtain. In this paper, we propose a zeroth-order AdaMM (ZO-AdaMM) algorithm, that generalizes AdaMM to the gradient-free regime. We show that the convergence rate of ZO-AdaMM for both convex and nonconvex optimization is roughly a factor of $O(\sqrt{d})$ worse than that of the first-order AdaMM algorithm, where $d$ is problem size. In particular, we provide a deep understanding on why Mahalanobis distance matters in convergence of ZO-AdaMM and other AdaMM-type methods. As a byproduct, our analysis makes the first step toward understanding adaptive learning rate methods for nonconvex constrained optimization.Furthermore, we demonstrate two applications, designing per-image and universal adversarial attacks from black-box neural networks, respectively. We perform extensive experiments on ImageNet and empirically show that ZO-AdaMM converges much faster to a solution of high accuracy compared with $6$ state-of-the-art ZO optimization methods.

Policy evaluation with smooth and nonlinear function approximation has shown great potential for reinforcement learning. Compared to linear function approxi- mation, it allows for using a richer class of approximation functions such as the neural networks. Traditional algorithms are based on two timescales stochastic approximation whose convergence rate is often slow. This paper focuses on an offline setting where a trajectory of $m$ state-action pairs are observed. We formulate the policy evaluation problem as a non-convex primal-dual, finite-sum optimization problem, whose primal sub-problem is non-convex and dual sub-problem is strongly concave. We suggest a single-timescale primal-dual gradient algorithm with variance reduction, and show that it converges to an $\epsilon$-stationary point using $O(m/\epsilon)$ calls (in expectation) to a gradient oracle.

Despite the empirical success of the actor-critic algorithm, its theoretical understanding lags behind. In a broader context, actor-critic can be viewed as an online alternating update algorithm for bilevel optimization, whose convergence is known to be fragile. To understand the instability of actor-critic, we focus on its application to linear quadratic regulators, a simple yet fundamental setting of reinforcement learning. We establish a nonasymptotic convergence analysis of actor- critic in this setting. In particular, we prove that actor-critic finds a globally optimal pair of actor (policy) and critic (action-value function) at a linear rate of convergence. Our analysis may serve as a preliminary step towards a complete theoretical understanding of bilevel optimization with nonconvex subproblems, which is NP-hard in the worst case and is often solved using heuristics.

In this paper, we propose a cross-layer decision framework for multiuser adaptive video delivery over time-varying and mutually interfering wireless cellular network. The key idea is to synthetically design the physical-layer optimization-based beamforming scheme (performed at the base stations) and the application-layer deep reinforcement learning (DRL)-based rate adaptation scheme (performed at the user terminals), so that a very complex multi-user overall fair long-term quality of experience (QoE) maximization problem can be decomposed to two layers and solved effectively. Extensive simulations show that the proposed cross-layer design is effective and promising.

In topic modeling, identifiability of the topics is an essential issue. Many topic modeling approaches have been developed under the premise that each topic has a characteristic anchor word that only appears in that topic. The anchor-word assumption is fragile in practice, because words and terms have multiple uses; yet it is commonly adopted because it enables identifiability guarantees. Remedies in the literature include using three- or higher-order word co-occurence statistics to come up with tensor factorization models, but such statistics need many more samples to obtain reliable estimates, and identifiability still hinges on additional assumptions, such as consecutive words being persistently drawn from the same topic. In this work, we propose a new topic identification criterion using second order statistics of the words. The criterion is theoretically guaranteed to identify the underlying topics even when the anchor-word assumption is grossly violated. An algorithm based on alternating optimization, and an efficient primal-dual algorithm are proposed to handle the resulting identification problem. The former exhibits high performance and is completely parameter-free; the latter affords up to 200 times speedup relative to the former, but requires step-size tuning and a slight sacrifice in accuracy. A variety of real text copora are employed to showcase the effectiveness of the approach, where the proposed anchor-free method demonstrates substantial improvements compared to a number of anchor-word based approaches under various evaluation metrics.

The sum-of-correlations (SUMCOR) generalized canonical correlation analysis (GCCA) aims at producing low-dimensional representations of multiview data via enforcing pairwise similarity of the reduced-dimension views. SUMCOR has been applied to a large variety of applications including blind separation, multilingual word embedding, and cross-modality retrieval. Despite the NP-hardness of SUMCOR, recent work has proposed effective algorithms for handling it at very large scale. However, the existing scalable algorithms are not easy to extend to incorporate structural regularization and prior information - which are critical for real-world applications where outliers and modeling mismatches are present. In this work, we propose a new computational framework for large-scale SUMCOR GCCA. The algorithm can easily incorporate a suite of structural regularizers which are frequently used in data analytics, has lightweight updates and low memory complexity, and can be easily implemented in a parallel fashion. The proposed algorithm is also guaranteed to converge to a Karush-Kuhn-Tucker (KKT) point of the regularized SUMCOR problem. Carefully designed simulations are employed to demonstrate the effectiveness of the proposed algorithm.

To support multiple on-demand services over several fixed communication networks, the network operators must allow flexible customization and fast provision of their network resources. One effective approach is network virtualization, whereby each service is mapped to a virtual subnetwork providing dedicated on-demand support. In practice, each service consists of a pre specified sequence of functions, called a service function chain (SFC). Moreover, each function in a SFC can only be provided by some given network nodes. Thus, to support a given service, we must select network function nodes according to the SFC, and determine the routing strategy through the function nodes in the specified order. A crucial problem that needs to be addressed is how to optimally allocate the network resources while satisfying multiple service requirements specified by the service function chains, subject to link and node capacity constraints. In this paper, we formulate the problem as a mixed binary linear program and establish its NP-hardness. Furthermore, we propose an efficient penalty successive upper bound minimization algorithm to solve the problem. We also present simulation results to demonstrate the effectiveness of the proposed algorithm.

In this work, we study two first-order primal-dual based algorithms, the Gradient Primal-Dual Algorithm (GPDA) and the Gradient Alternating Direction Method of Multipliers (GADMM), for solving a class of linearly constrained non-convex optimization problems. We show that with random initialization of the primal and dual variables, both algorithms are able to compute second-order stationary solutions (ss2) with probability one. This is the first result showing that primal-dual algorithm is capable of finding ss2 when only using first-order information; it also extends the existing results for first-order, but {primal-only} algorithms. An important implication of our result is that it also gives rise to the first global convergence result to the ss2, for two classes of unconstrained distributed non-convex learning problems over multi-agent networks.

Despite the success of single-agent reinforcement learning, multi-agent reinforcement learning (MARL) remains challenging due to complex interactions between agents. Motivated by decentralized applications such as sensor networks, swarm robotics, and power grids, we study policy evaluation in MARL, where agents with jointly observed state-action pairs and private local rewards collaborate to learn the value of a given policy. In this paper, we propose a double averaging scheme, where each agent iteratively performs averaging over both space and time to incorporate neighboring gradient information and local reward information, respectively. We prove that the proposed algorithm converges to the optimal solution at a global geometric rate. In particular, such an algorithm is built upon a primal-dual reformulation of the mean squared Bellman error minimization problem, which gives rise to a decentralized convex-concave saddle-point problem. To the best of our knowledge, the proposed double averaging primal-dual optimization algorithm is the first to achieve fast finite-time convergence on decentralized convex-concave saddle-point problems.

Symmetric nonnegative matrix factorization (SymNMF) plays an important role in applications of many data analytics problems such as community detection, document clustering and image segmentation. In this paper, we consider a stochastic SymNMF problem in which the observation matrix is generated in a random and sequential manner. We propose a stochastic nonconvex splitting method, which not only guarantees convergence to the set of stationary points of the problem (in the mean-square sense), but further achieves a sublinear convergence rate. Numerical results show that for clustering problems over both synthetic and real world datasets, the proposed algorithm converges quickly to the set of stationary points.

This paper studies the multiple-input multiple- output (MIMO) millimeter wave (mmWave) systems with non-linear power consumption model for 5G network. A new non-linear power consumption model is investigated, consisting of the power cost generated by the circuit and the non-linear power amplifiers. In this work, we aim to optimize the hybrid transceiver to maximize the system capacity subject to the resultant non-linear power constraint. In order to address this problem, we first transform the original optimization problem to a more tractable problem based on the weighted minimum mean squared error (WMMSE) approach. Then, we propose a novel transceiver design algorithm based on the penalty dual decomposition (PDD) optimization framework to address this problem. Moreover, a simplified algorithm is also proposed by using linear approximation. The effectiveness of the proposed algorithm is verified by simulation results.

Unlike dimensionality reduction (DR) tools for single-view data, e.g., principal component analysis (PCA), canonical correlation analysis (CCA) and generalized CCA (GCCA) are able to integrate information from multiple feature spaces of data. This is critical in multi-modal data fusion and analytics, where samples from a single view may not be enough for meaningful DR. In this work, we focus on a popular formulation of GCCA, namely, MAX-VAR GCCA. The classic MAX-VAR problem is optimally solvable via eigen-decomposition, but this solution has serious scalability issues. In addition, how to impose regularizers on the sought canonical components was unclear - while structure-promoting regularizers are often desired in practice. We propose an algorithm that can easily handle datasets whose sample and feature dimensions are both large by exploiting data sparsity. The algorithm is also flexible in incorporating regularizers on the canonical components. Convergence properties of the proposed algorithm are carefully analyzed. Numerical experiments are presented to showcase its effectiveness.

Symmetric non-negative matrix factorization (SymNMF) has important applications in data analytics problems such as document clustering, community detection and image segmentation. In this paper, we propose a novel nonconvex variable splitting method for solving SymNMF. Different from the existing works, we prove that the algorithm converges to the set of Karush-Kuhn-Tucker (KKT) points of the nonconvex SymNMF problem with a global sublinear convergence rate. We also show that the algorithm can be efficiently implemented in a distributed manner. Further, we provide sufficient conditions that guarantee the global and local optimality of the obtained solutions. Extensive numerical results performed on both synthetic and real data sets suggest that the proposed algorithm yields high quality of the solutions and converges quickly to the set of local minimum solutions compared with other algorithms.

Many problems of recent interest in signal processing, machine learning and wireless communications can be posed as nonconvex nonsmooth optimization problems. These problems are generally difficult to solve especially when the optimization variables are nonlinearly coupled in some nonconvex constraints. In this paper, we propose an algorithm named “penalty dual decomposition” (PDD) method, for the minimization of a nonconvex nonsmooth objective subject to nonconvex constraints. We show that the PDD converges to KKT solutions under certain constraint qualification condition. Simulations corroborate the excellent performance of the PDD method.

Traffic engineering (TE) problem is a central component of the next generation cloud-based wireless networks. In this paper, we study a new resource allocation scheme for effective traffic engineering under practical constraints such as the finite buffer size at each node. To reduce the computational effort required in the existing single-slot TE approaches and to deal with practical hardware limitations on link flows and buffers, we propose a two time-scale, low-complexity TE algorithm which incorporates a novel link scheduling component. The algorithm can be distributedly implemented. Simulation results demonstrate the effectiveness of the proposed algorithm.

In this paper we consider nonconvex optimization and learning over a network of distributed nodes. We develop a Proximal Primal-Dual Algorithm (Prox-PDA), which enables the network nodes to distributedly and collectively compute the set of first-order stationary solutions in a global sublinear manner [with a rate of $O(1/r)$, where $r$ is the iteration counter]. To the best of our knowledge, this is the first algorithm that enables distributed nonconvex optimization with global rate guarantees. Our numerical experiments also demonstrate the effectiveness of the proposed algorithm.

Most learning approaches treat dimensionality reduction (DR) and clustering separately (i.e., sequentially), but recent research has shown that optimizing the two tasks jointly can substantially improve the performance of both. The premise behind the latter genre is that the data samples are obtained via linear transformation of latent representations that are easy to cluster; but in practice, the transformation from the latent space to the data can be more complicated. In this work, we assume that this transformation is an unknown and possibly nonlinear function. To recover the `clustering-friendly’ latent representations and to better cluster the data, we propose a joint DR and K-means clustering approach in which DR is accomplished via learning a deep neural network (DNN). The motivation is to keep the advantages of jointly optimizing the two tasks, while exploiting the deep neural network’s ability to approximate any nonlinear function. This way, the proposed approach can work well for a broad class of generative models. Towards this end, we carefully design the DNN structure and the associated joint optimization criterion, and propose an effective and scalable algorithm to handle the formulated optimization problem. Experiments using different real datasets are employed to showcase the effectiveness of the proposed approach.

This work studies the joint transceiver design for a full-duplex (FD) cloud radio access network (C- RAN) with simultaneous wireless information and power transfer (SWIPT). In the considered network, a number of FD remote radio heads (RRHs) receive information from uplink users (UUs), while transmitting both information and energy to a set of half-duplex (HD) downlink users (DUs) with power splitting receivers. Based on the particular problem structure, a block coordinate descent (BCD) method is proposed to minimize the total transmission power subject to both uplink-downlink quality of service (QoS) constraints and energy harvesting (EH) constraints. Although the problem has complicated constraints coupling a set of transceivers, uplink transmit power levels, and receive power splitting ratios, we prove that the proposed BCD algorithm converges to a Karush-Kuhn- Tucker (KKT) solution. Simulation results validate the effectiveness of the proposed algorithm as compared with the traditional HD scheme.

To support multiple on-demand services over fixed communication networks, network operators must allow flexible customization and fast provision of their network resources. One effective approach to this end is network virtualization, whereby each service is mapped to a virtual subnetwork providing dedicated on-demand support to network users. In practice, each service consists of a prespecified sequence of functions, called a service function chain (SFC), while each service function in a SFC can only be provided by some given network nodes. Thus, to support a given service, we must select network function nodes according to the SFC and determine the routing strategy through the function nodes in a specified order. A crucial network slicing problem that needs to be addressed is how to optimally localize the service functions in a physical network as specified by the SFCs, subject to link and node capacity constraints. In this paper, we formulate the network slicing problem as a mixed binary linear program and establish its strong NP-hardness. Furthermore, we propose efficient penalty successive upper bound minimization (PSUM) and PSUM-R(ounding) algorithms, and two heuristic algorithms to solve the problem. Simulation results are shown to demonstrate the effectiveness of the proposed algorithms.

In this paper, we propose joint transceiver design algorithms for the full-duplex K -pair multiple-input multiple-output interference channel with simultaneous wireless information and power transfer. To mitigate and exploit the complex interference, we consider two important utility optimization problems, i.e., the sum power minimization problem and the sum-rate maximization problem. In the first problem, our aim is to minimize the total transmission power under both transmission rate and energy harvesting (EH) constraints. An iterative algorithm based on alternating optimization (AO) and with guaranteed monotonic convergence is proposed to successively optimize the transceiver coefficients. The algorithm consists of three main steps, where the concave-convex procedure (CCCP), the minimum mean-square error (MMSE) criterion, and the semidefinite relaxation technique are, respectively, employed to compute the vectors of power splitting ratios, the receiving matrices, and the transmitting beamforming vectors. Two simplified algorithms based on fixed beamformers, namely, the maximum ratio transmission and the maximum signal-to-interference-leakage beamformers are also proposed. In the second problem, our aim is to maximize the sum-rate under additional power and EH constraints. Due to the highly non-convex nature of this problem, we first reformulate it into an equivalent-weighted MMSE problem by introducing suitable weight factors, such that the global optima of the two problems are identical. Then, by utilizing the concept of AO and CCCP, we show that the equivalent problem can be efficiently solved. Again, with the aid of the fixed beamformers, two simplified algorithms are provided to reduce the computational complexity. Simulation results are presented to validate the effectiveness of the proposed algorithms.

The cyclic block coordinate descent-type (CBCD-type) methods have shown remarkable computational performance for solving strongly convex minimization problems. Typical applications include many popular statistical machine learning methods such as elastic-net regression, ridge penalized logistic regression, and sparse additive regression. Existing optimization literature has shown that the CBCD-type methods attain iteration complexity of O(p⋅\log(1/ε)), where εis a pre-specified accuracy of the objective value, and p is the number of blocks. However, such iteration complexity explicitly depends on p, and therefore is at least p times worse than those of gradient descent methods. To bridge this theoretical gap, we propose an improved convergence analysis for the CBCD-type methods. In particular, we first show that for a family of quadratic minimization problems, the iteration complexity of the CBCD-type methods matches that of the GD methods in term of dependency on p (up to a \log^2 p factor). Thus our complexity bounds are sharper than the existing bounds by at least a factor of p/\log^2p. We also provide a lower bound to confirm that our improved complexity bounds are tight (up to a \log^2 p factor) if the largest and smallest eigenvalues of the Hessian matrix do not scale with p. Finally, we generalize our analysis to other strongly convex minimization problems beyond quadratic ones.

Alternating direction method of multipliers (ADMM) has been recognized as an efficient approach for solving many large-scale learning problems over a computer cluster. However, traditional synchronized computation does not scale well with the problem size, as the speed of the algorithm is limited by the slowest workers. In this paper, we propose an asynchronous distributed ADMM (AD- ADMM) which can effectively improve the time efficiency of distributed optimization. Our main interest lies in characterizing the convergence conditions of the AD-ADMM, under the popular partially asynchronous model which is defined based on a maximum tolerable delay in the network. Specifically, by considering general and possibly non-convex cost functions, we show that the AD-ADMM converges to the set of Karush-Kuhn-Tucker (KKT) points as long as the algorithm parameters are chosen appropriately according to the network delay. We also show that the asynchrony of ADMM has to be handled with care, as a slightly different implementation can significantly jeopardize the algorithm convergence.

The nonnegative matrix factorization (NMF) has been a popular model for a wide range of signal processing and machine learning problems. It is usually formulated as a nonconvex cost minimization problem. This work settles the convergence issue of a popular algorithm based on the alternating direction method of multipliers proposed in Boyd et al 2011. We show that the algorithm converges globally to the set of KKT solutions whenever certain penalty parameter ρ satisfies ρ > 1. We further extend the algorithm and its analysis to the problem where the observation matrix contains missing values. Numerical experiments on real and synthetic data sets demonstrate the effectiveness of the algorithms under investigation.

We consider solving a convex, nonsmooth and stochastic optimization problem over a multi-agent network. Each agent has access to a local objective function and can communicate with its immediate neighbors only. We develop a dynamic stochastic proximal-gradient consensus (DySPGC) algorithm, featuring: i) it works for both the static and randomly time-varying networks; ii) it can deal with either the exact or the stochastic gradient information; iii) it has provable rate of convergence. Interestingly, the developed algorithm includes as special cases many existing (and seemingly unrelated) first-order algorithms for distributed optimization over static networks, such as the EXTRA (Shi et al 2014), the PG-EXTRA (Shi at 2015), the IC/IDC-ADMM (Chang et al 2014), and the DLM (Ling et al 2015). It is also closely related to the classical distributed gradient method.

This paper studies joint source transmit beamforming and relay amplification matrix design to achieve rate maximization for full-duplex (FD) MIMO amplify-and-forward (AF) relay systems with consideration of relay processing delay (RPD). The problem is difficult to solve due mainly to the self-interference constraint induced by the RPD. In this paper, we first propose a penalty-based algorithmic framework, called P-BSUM, for a class of constrained optimization problems with difficult equality constraints in addition to some convex constraints. We then apply the P-BSUM algorithm to the rate maximization problem and obtain a simple iterative algorithm. Finally, numerical results illustrate the efficiency of the proposed algorithm.

This paper considers multi-agent distributed optimization with quantized communication which is needed when inter-agent communications are subject to finite capacity and other practical constraints. To minimize the global objective formed by a sum of local convex functions, we develop a quantized distributed algorithm based on the alternating direction method of multipliers (ADMM). Under certain convexity assumptions, it is shown that the proposed algorithm converges to a consensus within log1+ηΩ iterations, where η > 0 depends on the network topology and the local objectives, and O is a polynomial fraction depending on the quantization resolution, the distance between initial and optimal variable values, the local objectives, and the network topology. We also obtain a tight upper bound on the consensus error which does not depend on the size of the network.

We study a stochastic and distributed algorithm for nonconvex problems whose objective consists a sum $N$ nonconvex $L_i/N$-smooth functions, plus a nonsmooth regularizer. The proposed NonconvEx primal-dual SpliTTing (NESTT) algorithm splits the problem into $N$ subproblems, and utilizes an augmented Lagrangian based primal-dual scheme to solve it in a distributed and stochastic manner. With a special non-uniform sampling, a version of NESTT achieves $\epsilon$-stationary solution using $\mathcal{O}((\sum_{i=1}^N\sqrt{L_i/N})^2/\epsilon)$ gradient evaluations, which can be up to $\mathcal{O}(N)$ times better than the (proximal) gradient descent methods. It also achieves Q-linear convergence rate for nonconvex $\ell_1$ penalized quadratic problems with polyhedral constraints. Further, we reveal a fundamental connection between {\it primal-dual} based methods and a few {\it primal only} methods such as IAG/SAG/SAGA.

In this paper, we propose a joint transceiver design algorithm for the full-duplex (FD) K-pair multiple- input multiple-output (MIMO) interference channel with simultaneous wireless information and power transfer (SWIPT). The aim is to minimize the total transmission power under both transmission rate and energy harvesting (EH) constraints. An iterative algorithm based on alternating optimization and with guaranteed monotonic convergence is proposed to successively optimize the transceiver coefficients. The algorithm consists of three main steps, aimed at successively optimizing: 1) the power splitting (PS) vectors of the EH nodes; 2) the receive beamforming vectors; 3) the transmit beamforming vectors.The first step is carried out based on concave-convex procedure (CCCP), the second step is based on the minimum mean square error (MMSE) criterion and the third step resorts to using semidefinite relaxation (SDR). Simulation results are presented to validate the effectiveness of the proposed algorithm.

We study the downlink linear precoder design problem in a multicell dense heterogeneous network (HetNet). The problem is formulated as a general sum-utility maximization (SUM) problem, which includes as special cases many practical precoder design problems such as multicell coordinated linear precoding, full and partial per-cell coordinated multipoint transmission, zero-forcing precoding, and joint BS clustering and beamforming/precoding. The SUM problem is difficult due to its nonconvexity and the tight coupling of the users' precoders. In this paper, we propose a novel convex approximation technique to approximate the original problem by a series of convex subproblems, each of which decomposes across all the cells. The convexity of the subproblems allows for efficient computation, while their decomposability leads to distributed implementation. Our approach hinges upon the identification of certain key convexity properties of the sum-utility objective, which allows us to transform the problem into a form that can be solved using a popular algorithmic framework called block successive upper-bound minimization (BSUM). Simulation experiments show that the proposed framework is effective for solving interference management problems in large HetNet.

Consider the joint power and admission control (JPAC) problem for a multiuser single-input single-output (SISO) interference channel. Most existing works on JPAC assume the perfect instantaneous channel state information (CSI). In this paper, we consider the JPAC problem with the imperfect CSI, i.e., we assume that only the channel distribution information (CDI) is available. We formulate the JPAC problem into a chance (probabilistic)-constrained program, where each link's SINR outage probability is enforced to be less than or equal to a specified tolerance. To circumvent the computational difficulty of the chance SINR constraints, we propose to use the sample (scenario) approximation scheme to convert them into finitely many simple linear constraints. Furthermore, we reformulate the sample approximation of the chance SINR-constrained JPAC problem as a composite group sparse minimization problem and then approximate it by a second-order cone program (SOCP). The solution of the SOCP approximation can be used to check the simultaneous supportability of all links in the network and to guide an iterative link removal procedure (the deflation approach). We exploit the special structure of the SOCP approximation and custom-design an efficient algorithm for solving it. Finally, we illustrate the effectiveness and efficiency of the proposed sample approximation-based deflation approaches by simulations.

In this paper, we analyze the behavior of the alternating direction method of multipliers (ADMM), for solving a family of nonconvex problems. Our focus is given to the well-known consensus and sharing problems, both of which have wide applications in signal processing. We show that in the presence of nonconvex objective function, classical ADMM is able to reach the set of stationary solutions for these problems, if the stepsize is chosen large enough. An interesting consequence of our analysis is that the ADMM is convergent for a family of sharing problems, regardless of the number of blocks or the convexity of the objective function. Our analysis is broadly applicable to many ADMM variants involving proximal update rules and various flexible block selection rules.

We consider the distributed network routing problem in a large-scale hierarchical network whereby the nodes are partitioned into subnetworks, each managed by a network controller (NC), and there is a central NC to coordinate the operation of the distributed NCs. We propose a semi-asynchronous routing algorithm for such a network, whereby the computation is distributed across the NCs and is parallel within each NC. A key feature of the algorithm is its ability to handle a certain degree of asynchronism: the distributed NCs can perform their local computation asynchronously at different processing speed. The efficiency of the proposed algorithm is validated through numerical experiments.

In this paper, we propose a beamforming algorithm for binaural hearing aids with enhanced noise suppression capability. The enhancement is based on incorporating a priori spatial information into the conventional multichannel Wiener filtering (MWF) approach for noise suppression. We develop a low complexity algorithm for the resulting quadratically constrained beamforming problem. Through numerical experiments, we demonstrate that the new algorithm can achieve better noise suppression performance than the existing beamforming algorithms under fairly realistic conditions. In addition, we propose two techniques to further reduce the algorithm's computational complexity and the communication overhead between two hearing aids without sacrificing the noise suppression performance.

In this work, we consider enhancing a target speech from a single-channel noisy observation corrupted by non-stationary noises at low signal-to-noise ratios (SNRs). We take a classification-based approach, where the objective is to estimate an Ideal Binary Mask (IBM) that classifies each time-frequency (T-F) unit of the noisy observation into one of the two categories: speech-dominant unit or noise-dominant unit. The estimated mask is used to binary weight the noisy mixture to obtain the enhanced speech. In the proposed system, the sparse non-negative matrix factorization (NMF) is used to extract features from the noisy observation, followed by a Deep Neural Network (DNN) for classification. Compared with several existing classification-based systems, the proposed system uses minimal speech-specific domain knowledge, but is able to achieve better performance in certain low SNR regions. Moreover, the proposed system outperforms the traditional statistical method, especially in terms of improving the intelligibility.

The iteration complexity of the block-coordinate descent (BCD) type algorithm has been under extensive investigation. It was recently shown that for convex problems the classical cyclic BCGD (block coordinate gradient descent) achieves an O(1/r) complexity (r is the number of passes of all blocks). However, such bounds are at least linearly depend on $K$ (the number of variable blocks), and are at least $K$ times worse than those of the gradient descent (GD) and proximal gradient (PG) methods.In this paper, we close such theoretical performance gap between cyclic BCD and GD/PG. First we show that for a family of quadratic nonsmooth problems, the complexity bounds for cyclic Block Coordinate Proximal Gradient (BCPG), a popular variant of BCD, can match those of the GD/PG in terms of dependency on $K$ (up to a \log^2(K) factor). Second, we establish an improved complexity bound for Coordinate Gradient Descent (CGD) for general convex problems which can match that of GD in certain scenarios. Our bounds are sharper than the known bounds as they are always at least $K$ times worse than GD. {Our analyses do not depend on the update order of block variables inside each cycle, thus our results also apply to BCD methods with random permutation (random sampling without replacement, another popular variant).

In a heterogeneous network (HetNet) with a large number of low power base stations (BSs), proper user-BS association and power control is crucial to achieving desirable system performance. In this paper, we systematically study the joint BS association and power allocation problem for a downlink cellular network under the max-min fairness criterion. First, we show that this problem is NP-hard. Second, we show that the upper bound of the optimal value can be easily computed, and propose a two-stage algorithm to find a high-quality suboptimal solution. Simulation results show that the proposed algorithm is near-optimal in the high-SNR regime. Third, we show that the problem under some additional mild assumptions can be solved to global optima in polynomial time by a semi-distributed algorithm. This result is based on a transformation of the original problem to an assignment problem with gains log(gij), where {gij} are the channel gains.

The multi-agent distributed consensus optimization problem arises in many engineering applications. Recently, the alternating direction method of multipliers (ADMM) has been applied to distributed consensus optimization which, referred to as the consensus ADMM (C-ADMM), can converge much faster than conventional consensus subgradient methods. However, C-ADMM can be computationally expensive when the cost function to optimize has a complicated structure or when the problem dimension is large. In this paper, we propose an inexact C-ADMM (IC-ADMM) where each agent only performs one proximal gradient (PG) update at each iteration. The PGs are often easy to obtain especially for structured sparse optimization problems. Convergence conditions for IC-ADMM are analyzed. Numerical results based on a sparse logistic regression problem show that IC-ADMM, though converges slower than the original C-ADMM, has a considerably reduced computational complexity.

In this paper, we consider a nonsmooth convex problem with linear coupling constraints. Problems of this form arise in many modern large-scale signal processing applications including the provision of smart grid networks. In this work, we propose a new class of algorithms called the block coordinate descent method of multipliers (BCDMM) to solve this family of problems. The BCDMM is a primal-dual type of algorithm. It optimizes an (approximate) augmented Lagrangian of the original problem one block variable per iteration, followed by a gradient update for the dual variable. We show that under certain regularity conditions, and when the order for which the block variables are either updated in a deterministic or a random fashion, the BCDMM converges to the set of optimal solutions. The effectiveness of the algorithm is illustrated using large-scale basis pursuit and smart grid problems.

We consider a heterogenous network (HetNet) consisting of a number of base stations (BSs) and network routers connected via a backhaul network. The optimal provision of such networks requires proper resource allocation across the radio access links in conjunction with appropriate traffic engineering within the backhaul network. In this paper we propose an efficient distributed algorithm for the joint resource allocation across the wireless links and the flow control within the backhaul network. The proposed algorithm, which maximizes the minimum rate among all the users and/or flows, is based on a decomposition approach that leverages both the Alternating Direction Method of Multipliers (ADMM) and the WMMSE algorithm, and is shown to be globally convergent to a stationary solution of the joint flow control and resource allocation problem. Moreover, this algorithm is easily parallelizable and can be extended to the multi-antenna scenario.

In this work, we consider the joint day-ahead power bidding and load scheduling problem for the smart grid system, in the presence of uncertain energy demand and renewable energy generation. We formulate the problem as a convex stochastic program in which the renewable energy generation and energy demand are modeled as random variables. The objective is to minimize the cost in the day-ahead market as well as the cost due to real-time power imbalance, by simultaneously selecting: 1) the amount of power to buy in the day-ahead market and 2) the schedule for the controllable load. We propose a stochastic alternating direction method of multipliers (S AD-MM) to solve the resulting convex stochastic optimization problem and analyze its convergence. The effectiveness of the proposed approach is demonstrated via numerical experiments using real solar power data.

We consider a cloud-based heterogeneous network of base stations (BSs) connected via a backhaul network of routers and wired/wireless links with limited capacity. The optimal provision of such networks requires proper resource allocation across the radio access links in conjunction with appropriate traffic engineering within the backhaul network. In this paper, we propose an efficient algorithm for joint resource allocation across the wireless links and flow control over the entire network. The proposed algorithm, which maximizes the min-rate among all the transmitted commodities, is based on a decomposition approach that leverages both the alternating direction method of multipliers (ADMM) and the weighted-MMSE (WMMSE) algorithm. We show that this algorithm is easily parallelizable and converges globally to a stationary solution of the joint optimization problem. The proposed algorithm can also be extended to networks with multi-antenna nodes and other utility functions.

We consider the downlink coordinated beamforming problem in a cellular network in which the base stations (BSs) are equipped with multiple antennas and each user is equipped with a single antenna. The BSs cooperate in sharing their local interference information, and they aim to maximize the sum-rate of the users in the network. A decentralized interference pricing beamforming (IPBF) algorithm is proposed to identify the coordinated beamformer, where a BS is penalized according to the interference it creates to its peers. We show that the decentralized pricing mechanism converges to an interference equilibrium, which is a KKT point of the sum-rate maximization problem. The proofs rely on the identification of rank-1 solutions of each BSs' interference-penalized rate maximization problem. Numerical results show that the proposed iterative mechanism reduces significantly the exchanged information with respect to other state-of-the-art beamforming algorithms with very little sum-rate loss. The version of the algorithm that limits the coordination to a cluster of base stations (IPBF-L) is shown to have very small sum-rate loss with respect to the full coordinated algorithm with much less backhaul information exchange.

In this paper, we consider the robust secure beamformer design for multiple-input-single-output wiretap channels. Assuming that the eavesdroppers' channels are only partially available at the transmitter, we seek to maximize the secrecy rate under the transmit power and the secrecy rate outage probability constraint. The outage probability constraint requires that the secrecy rate exceed certain thresholds with high probability. Therefore, including such constraint in the design naturally ensures the desired robustness. Unfortunately, the presence of the probabilistic constraints makes the problem nonconvex and, hence, difficult to solve. In this paper, we investigate the outage probability constrained secrecy rate maximization problem using a novel two-step approach. Under a wide range of uncertainty models, our developed algorithms can obtain high-quality solutions, sometimes even exact global solutions, for the robust secure beamformer design problem. Simulation results are presented to verify the effectiveness and robustness of the proposed algorithms.

This paper introduces Truthful Multichannel Auction (TMCA), an auction design for the allocation of wireless channels to several bidders with private information about their channel valuation. Channel allocations are subject to interference constraints in the form of a conflict graph. In contrast to other channel auctions, TMCA allows for variable (instead of fixed) marginal valuations. In TMCA, it is a dominant strategy for bidders to truthfully reveal their channel valuations, which in turn guarantees the implementation of highly efficient allocations in polynomial time. This paper also shows the expected revenue of the auctioneer can be maximized by imposing a reserve price.

Loudness restoration approaches to hearing aid fitting prescribe gain and compression so as to restore the loudness perceived by a hearing-impaired listener to that perceived by a listener with normal-hearing. Restoring the loudness perception to normal is complicated by the spread of excitation at high stimulus levels that causes intense stimuli at low frequencies to be “heard” and to contribute to the perceived loudness at high frequencies, producing excess loudness growth and poor sound quality. We apply derivative-free optimization algorithms to find a configuration of hearing aid gain and compression parameters that restores specific loudness perception of a hearing impaired listener to that of a normal hearing listener, while simultaneously minimizing the across-frequency spreading of excitation, and ensuring the feasibility of the resulting hearing aid parameters.

This paper considers transmit covariance optimization for a multi-input multi-output (MIMO) Gaussian wiretap channel. Specifically, we aim to maximize the MIMO secrecy capacity by judiciously designing the transmit covariance under the sum power and per-antenna power constraints. The MIMO secrecy capacity maximization (SCM) problem is nonconvex, and so far there is no tractable solution available. We propose an alternating optimization (AO) approach to handle the SCM problem. In particular, our development consists of two steps: First, we show that the SCM problem can be reexpressed to a form that can be conveniently processed by AO. Second, we develop a custom-designed fast algorithm for each AO iteration. Interestingly, with this fast implementation, the overall AO algorithm can be viewed as performing iterative reweighting and water-filling. Finally, the convergence of the proposed algorithm to a stationary solution of SCM is shown, and numerical results are provided to demonstrate its efficacy.

In a densely deployed Heterogeneous network (HetNet), the number of pico/micro base stations (BS) can be comparable or more than the number of the users. To reduce the operational overhead of the HetNet, selection of serving BSs becomes an important design issue. In this work, we propose to jointly optimize the transceiver and active BSs to trade off the overall spectrum efficiency with the operational overhead. We formulate this problem as a regularized sum rate maximization problem and solve it using sparse optimization techniques. The proposed algorithm is guaranteed to converge to a local optimal solution. The efficiency and the efficacy of the algorithm are demonstrated via realistic numerical simulations.

In this paper, a novel algorithm named Sparsity-based Wiener plus Dictionary Learning (SWDL) is proposed for single channel speech enhancement. SWDL combines both Wiener filter and dictionary learning technique. The Wiener filter is used to ensure the enhanced speech is statistically optimal, while the dictionary learning technique is used to improve the enhanced speech quality and intelligibility by utilizing speech-specific information. Such information is incorporated in the pre-trained speech dictionary that can sparsely represent the clean speech spectra. When applied to the TIM-IT database, SWDL outperforms the Log Mean Square-Error Short-Time Spectra Amplitude estimator (LSTSA) according to four different objective metrics measuring speech quality and intelligibility. Subjective tests also show that SWDL produces better speech quality and intelligibility than LSTSA.

In this paper, we consider the single-channel speech enhancement problem, in which a clean speech signal needs to be estimated from a noisy observation. To capture the characteristics of both the noise and speech signals, we combine the well-known Short-Time-Spectrum-Amplitude (STSA) estimator with a machine learning based technique called Multi-frame Sparse Dictionary Learning (MSDL). The former utilizes statistical information for denoising, while the latter helps better preserve speech, especially its temporal structure. The proposed algorithm, named STSA-MSDL, outperforms standard statistical algorithms such as the Wiener filter, STSA estimator, as well as dictionary based algorithms when applied to the TIMIT database, using four different objective metrics that measure speech intelligibility, speech distortion, background noise reduction, and the overall quality.

We consider the interference management problem in a multicell MIMO heterogeneous network. Within each cell there is a large number of distributed micro/pico base stations (BSs) that can be potentially coordinated for joint transmission. To reduce coordination overhead, we consider user-centric BS clustering so that each user is served by only a small number of (potentially overlapping) BSs. Thus, given the channel state information, our objective is to jointly design the BS clustering and the linear beamformers for all BSs in the network. In this paper, we formulate this problem from a {sparse optimization} perspective, and propose an efficient algorithm that is based on iteratively solving a sequence of group LASSO problems. A novel feature of the proposed algorithm is that it performs BS clustering and beamformer design jointly rather than separately as is done in the existing approaches for partial coordinated transmission. Moreover, the cluster size can be controlled by adjusting a single penalty parameter in the nonsmooth regularized utility function. The convergence of the proposed algorithm (to a stationary solution) is guaranteed, and its effectiveness is demonstrated via extensive simulation.

In a wireless system with a large number of distributed nodes, the quality of communication can be greatly improved by pooling the nodes to perform joint transmission/reception. In this paper, we consider the problem of optimally selecting a subset of nodes from potentially a large number of candidates to form a virtual multi-antenna system, while at the same time designing their joint linear transmission strategies. We focus on two specific application scenarios: 1) multiple single antenna transmitters cooperatively transmit to a receiver; 2) a single transmitter transmits to a receiver with the help of a number of cooperative relays. We formulate the joint node selection and beamforming problems as cardinality constrained optimization problems with both discrete variables (used for selecting cooperative nodes) and continuous variables (used for designing beamformers). For each application scenario, we first characterize the computational complexity of the joint optimization problem, and then propose novel semi-definite relaxation (SDR) techniques to obtain approximate solutions. We show that the new SDR algorithms have a guaranteed approximation performance in terms of the gap to global optimality, regardless of channel realizations. The effectiveness of the proposed algorithms is demonstrated via numerical experiments.

This paper considers transmit optimization in multi-input multi-output (MIMO) wiretap channels, wherein we aim at maximizing the secrecy capacity or rate of an MIMO channel overheard by one or multiple eavesdroppers. Such optimization problems are nonconvex, and appear to be difficult especially in the multi-eavesdropper scenario. In this paper, we propose an alternating optimization (AO) approach to tackle these secrecy optimization problems. We first consider the secrecy capacity maximization (SCM) problem in the single eavesdropper scenario. An AO algorithm is derived through a judicious SCM reformulation. The algorithm conducts some kind of reweighting and water-filling in an alternating fashion, and thus is computationally efficient to implement. We also prove that the AO algorithm is guaranteed to converge to a Karush-Kuhn-Tucker (KKT) point of the SCM problem. Then, we turn our attention to the multiple eavesdropper scenario, where the artificial noise (AN)-aided secrecy rate maximization (SRM) problem is considered. Although the AN-aided SRM problem has a more complex problem structure than the previous SCM, we show that AO can be extended to deal with the former, wherein the problem is handled by solving convex problems in an alternating fashion. Again, the resulting AO method is proven to have KKT point convergence guarantee. For fast implementation, a custom-designed AO algorithm based on smoothing and projected gradient is also derived. The secrecy rate performance and computational efficiency of the proposed algorithms are demonstrated by simulations.

Consider the linear transceiver design problem for a multi-user single-input multi-output (SIMO) interference channel. Assuming perfect channel knowledge, we formulate this problem as one of maximizing the minimum signal to interference plus noise ratio (SINR) among all the users, subject to individual power constraints at each transmitter. We prove in this letter that the max-min fairness linear transceiver design problem for the SIMO interference channel can be solved to global optimality in polynomial time. We further propose a low-complexity inexact cyclic coordinate ascent algorithm (ICCAA) to solve this problem. Numerical simulations show the proposed algorithm can efficiently find the global optimal solution of the considered problem.

We consider the problem of weighted sum rate optimization in a MIMO interfering multiple access channel (IMAC). We propose to jointly optimize the users' linear procoders as well as their base station (BS) associations. This approach enables the users to avoid congested BSs and can improve system performance as well as user fairness. We formulate the problem into a noncooperative game, and develop an algorithm that allows the players to distributedly reach the Nash Equilibrium (NE) of the game. We show that every NE of the game is a stationary solution of the weighted sum rate optimization problem, and propose an algorithm to compute the NE of the game. Simulation results show that the proposed algorithm performs well in the presence of BS congestion.

We consider a resource management problem in a multi-cell downlink OFDMA network whereby the goal is to find the optimal combination of (i) assignment of users to base stations and (ii) resource allocation strategies at each base station. Efficient resource management protocols must rely on users truthfully reporting privately held information such as downlink channel states. However, individual users can manipulate the resulting resource allocation (by misreporting their private information) if by doing so they can improve their payoff. Therefore, it is of interest to design efficient resource management protocols that are strategy-proof, i.e. it is in the users' best interests to truthfully report their private information. Unfortunately, we show that the implementation of any protocol that is efficient and strategy-proof is NP-hard. Thus, we propose a computationally tractable strategy-proof mechanism that is approximately efficient, i.e. the solution obtained yields at least 1/2 of the optimal throughput. Simulations are provided to illustrate the effectiveness of the proposed mechanism.

Spectrum management has been identified as a crucial step towards enabling the technology of the cognitive radio network (CRN). Most of the current works dealing with spectrum management in the CRN focus on a single task of the problem, e.g., spectrum sensing, spectrum decision, spectrum sharing or spectrum mobility. In this work, we argue that for certain network configurations, jointly performing several tasks of the spectrum management improves the spectrum efficiency. Specifically, we study the uplink resource management problem in a CRN where there exist multiple cognitive users (CUs) and access points (APs), with each AP operates on a set of non-overlapping channels. The CUs, in order to maximize their uplink transmission rates, have to associate to a suitable AP (spectrum decision), and to share the channels belong to this AP with other CUs (spectrum sharing). These tasks are clearly interdependent, and the problem of how they should be carried out efficiently and distributedly is still open in the literature. In this work we formulate this joint spectrum decision and spectrum sharing problem into a non-cooperative game, in which the feasible strategy of a player contains a discrete variable and a continuous vector. The structure of the game is hence very different from most non-cooperative spectrum management game proposed in the literature. We provide characterization of the Nash Equilibrium (NE) of this game, and present a set of novel algorithms that allow the CUs to distributively and efficiently select the suitable AP and share the channels with other CUs. Finally, we study the properties of the proposed algorithms as well as their performance via extensive simulations.

Previous localization solutions in wireless sensor networks mainly focus on using various techniques to estimate node positions. In this paper, we argue that quantifying the uncertainty of these estimates is equally important in practice. By using the quantitative uncertainty of measurements and estimates, we can derive more accurate estimates by better fusing the measurements, provide confidence information for confidence-based applications, and know how to select the best anchor nodes so as to minimize the total mean square errors of the whole network. This paper quantifies the estimation uncertainty as an error covariance matrix, and presents an efficient incremental centralized algorithm---INOVA and a decentralized algorithm---OSE-COV for calculating the error covariance matrix. Furthermore, we present how to use the error covariance matrix to infer the confidence region of each node's estimate, and provide an optimal strategy for the anchor selection problem. Extensive simulation results show that INOVA significantly improves the computation efficiency when the network changes dynamically; the confidence region inference is accurate when the measurement number to node number ratio is more than 2; and the optimal anchor selection strategy reduces the total mean square error by four times as much as the variation-based algorithm in best case.

Population Monte Carlo is a statistical method that is used for generation of samples approximately from a target distribution. The method is iterative in nature and is based on the principle of importance sampling. In this paper, we show that in problems where some of the parameters are conditionally linear on the remaining parameters, we can improve the computational efficiency of population Monte Carlo by generating samples of the nonlinear parameters only and marginalizing the linear parameters. We demonstrate the marginalized population Monte Carlo on the problem of frequency estimation of closely spaced sinusoids.