Predicting in-hospital mortality for ICU patients is critical for improving clinical outcomes. Deep learning models have achieved remarkable accuracy but often lack explainability. To address this, we propose$X-M M P$, an eXplainable Multimodal Mortality Predictor based on transformers for heterogeneous ICU data. X-MMP integrates tabular time-series, vital signs, and clinical notes, and introduces a Layer-Wise Relevance Propagation (LRP) method to generate interpretable saliency maps across modalities. Evaluated on the multimodal MIMICIII dataset, X-MMP achieves competitive predictive performance and generates clinically meaningful explanations. Code: https://github.com/lixingqiao/XAI-ICU.

Prompt learning has attracted widespread attention in adapting vision-language models to downstream tasks. Existing methods largely rely on optimization strategies to ensure the task-awareness of learnable prompts. Due to the scarcity of task-specific data, overfitting is prone to occur. The resulting prompts often do not generalize well or exhibit limited task-awareness. To address this issue, we propose a novel Task-Aware Clustering (TAC) framework for prompting vision-language models, which increases the task-awareness of learnable prompts by introducing task-aware pre-context. The key ingredients are as follows: (a) generating task-aware pre-context based on task-aware clustering that can preserve the backbone structure of a downstream task with only a few clustering centers, (b) enhancing the task-awareness of learnable prompts by enabling them to interact with task-aware pre-context via the well-pretrained encoders, and (c) preventing the visual task-aware pre-context from interfering the interaction between patch embeddings by masked attention mechanism. Extensive experiments are conducted on benchmark datasets, covering the base-to-novel, domain generalization, and cross-dataset transfer settings. Ablation studies validate the effectiveness of key ingredients. Comparative results show the superiority of our TAC over competitive counterparts. The code is available at https://github.com/FushengHao/TAC.

Decentralized learning offers a promising approach to crowdsource data consumptions and computational workloads across geographically distributed compute interconnected through peer-to-peer networks, accommodating the exponentially increasing demands. However, proper incentives are still in absence, considerably discouraging participation. Our vision is that a fair incentive mechanism relies on fair attribution of contributions to participating nodes, which faces non-trivial challenges arising from the localized connections making influence ``cascade'' in a decentralized network. To overcome this, we design the first method to estimate Data Influence CascadE (DICE) in a decentralized environment. Theoretically, the framework derives tractable approximations of influence cascade over arbitrary neighbor hops, suggesting the influence cascade is determined by an interplay of data, communication topology, and the curvature of loss landscape.DICE also lays the foundations for applications including selecting suitable collaborators and identifying malicious behaviors. Project page is available at https://raiden-zhu.github.io/blog/2025/DICE.

Massive human-related data is collected to train neural networks for computer vision tasks. A major conflict is exposed relating to software engineers between better developing AI systems and distancing from the sensitive training data. To reconcile this conflict, the paper proposes an efficient privacy-preserving learning paradigm, where images are encrypted to become ``human-imperceptible, machine-recognizable'' via one of the two encryption strategies: (1) random shuffling equally-sized patches and (2) mixing-up sub-patches. Then, minimal adaptations are made to vision transformer to enable it to learn on the encrypted images for vision tasks, including image classification and object detection. Extensive experiments on ImageNet and COCO show that the proposed paradigm achieves comparable accuracy with the competitive methods. Decrypting the encrypted images requires solving an NP-hard jigsaw puzzle or ill-posed inverse problem, which is empirically shown intractable to be recovered by various attackers, including the powerful vision transformer-based attacker. We thus show that the proposed paradigm can ensure the encrypted images have become human-imperceptible while preserving machine-recognizable information.

Sign languages are the primary languages of the deaf community as well as hearing individuals who are unable to speak, which engage the visual-manual modality to convey meanings. In recent years, there has been an explosive growth of sign language videos available from video streaming and social media service platforms. Given the size of these corpora, sign language users often face significant challenges in effectively acquiring the information they need. Therefore, we propose a novel deep learning architecture, namely Graph Traverse Reference Network (GTRN), allowing visual signing queries to retrieve relevant sign language videos (documents) from a large corpus. GTRN introduces a traverse graph, which provides coarse-to-fine reference information in a hierarchical manner from frame-level to body-part-level observations. A reference-based attention is devised to obtain the embedding for a visual input of each level, which allows the computations to be allocated and processed at difference locations regarding local devices and central servers. A contrastive learning strategy optimizes GTRN in pursuit of a joint latent space for the queries and the documents by their meanings. Moreover, GTRN is compatible to leverage existing general visual representation foundation models, by which their resulted embeddings are used as the frame-level reference of GTRN. To the best of our knowledge, it is one of the first studies on using visual signing queries for retrieving sign language videos in a real-world setting and comprehensive experiments were conducted which demonstrated the effectiveness of our proposed method. • A deep network formulates sign language embeddings for video queries and documents. • A reference-based attention method and contrastive strategy for embedding alignment. • Comprehensive experiments and analyses on a real-world sign language corpus.

With the increasing penetration of machine learning applications in critical decision-making areas, calls for algorithmic fairness are more prominent. Although there have been various modalities to improve algorithmic fairness through learning with fairness constraints, their performance does not generalize well in the test set. A performance-promising fair algorithm with better generalizability is needed. This article proposes a novel adaptive reweighing method to eliminate the impact of the distribution shifts between training and test data on model generalizability. Most previous reweighing methods propose to assign a unified weight for each (sub)group. Rather, our method granularly models the distance from the sample predictions to the decision boundary. Our adaptive reweighing method prioritizes samples closer to the decision boundary and assigns a higher weight to improve the generalizability of fair classifiers. Extensive experiments are performed to validate the generalizability of our adaptive priority reweighing method for accuracy and fairness measures (i.e., equal opportunity, equalized odds, and demographic parity) in tabular benchmarks. We also highlight the performance of our method in improving the fairness of language and vision models. The code is available at https://github.com/che2198/APW .

This article studies the generalization of neural networks (NNs) by examining how a network changes when trained on a training sample with or without out-of-distribution (OoD) examples. If the network's predictions are less influenced by fitting OoD examples, then the network learns attentively from the clean training set. A new notion, dataset-distraction stability, is proposed to measure the influence. Extensive CIFAR-10/100 experiments on the different VGG, ResNet, WideResNet, ViT architectures, and optimizers show a negative correlation between the dataset-distraction stability and generalizability. With the distraction stability, we decompose the learning process on the training set into multiple learning processes on the subsets of drawn from simpler distributions, i.e., distributions of smaller intrinsic dimensions (IDs), and furthermore, a tighter generalization bound is derived. Through attentive learning, miraculous generalization in deep learning can be explained and novel algorithms can also be designed.

This article discovers that the neural network (NN) with lower decision boundary (DB) variability has better generalizability. Two new notions, algorithm DB variability and -data DB variability, are proposed to measure the DB variability from the algorithm and data perspectives. Extensive experiments show significant negative correlations between the DB variability and the generalizability. From the theoretical view, two lower bounds based on algorithm DB variability are proposed and do not explicitly depend on the sample size. We also prove an upper bound of order based on data DB variability. The bound is convenient to estimate without the requirement of labels and does not explicitly depend on the network size which is usually prohibitively large in deep learning.

This paper presents the first theoretical guarantee for Bayesian bilevel optimization (BBO) that we term for the prevalent bilevel framework combining Bayesian optimization at the outer level to tune hyperparameters, and the inner-level stochastic gradient descent (SGD) for training the model. We prove sublinear regret bounds suggesting simultaneous convergence of the inner-level model parameters and outer-level hyperparameters to optimal configurations for generalization capability. A pivotal, technical novelty in the proofs is modeling the excess risk of the SGD-trained parameters as evaluation noise during Bayesian optimization. Our theory implies the inner unit horizon, defined as the number of SGD iterations, shapes the convergence behavior of BBO. This suggests practical guidance on configuring the inner unit horizon to enhance training efficiency and model performance.

This paper studies how to tune the stepping schedule in diffusion models, which is mostly fixed in current practice, lacking theoretical foundations and assurance of optimal performance at the chosen discretization points. In this paper, we advocate the use of adaptive time-stepping schedules and design two algorithms with an optimized sampling error bound $EB$: (1) for continuous diffusion, we treat $EB$ as the loss function to discretization points and run gradient descent to adjust them; and (2) for discrete diffusion, we propose a greedy algorithm that adjusts only one discretization point to its best position in each iteration. We conducted extensive experiments that show (1) improved generation ability in well-trained models, and (2) premature though usable generation ability in under-trained models. The code is submitted and will be released publicly.

Many machine learning problems can be formulated as non-convex multi-player games. Due to non-convexity, it is challenging to obtain the existence condition of the global Nash equilibrium (NE) and design theoretically guaranteed algorithms. This paper studies a class of non-convex multi-player games, where players' payoff functions consist of canonical functions and quadratic operators. We leverage conjugate properties to transform the complementary problem into a variational inequality (VI) problem using a continuous pseudo-gradient mapping. We prove the existence condition of the global NE as the solution to the VI problem satisfies a duality relation. We then design an ordinary differential equation to approach the global NE with an exponential convergence rate. For practical implementation, we derive a discretized algorithm and apply it to two scenarios: multi-player games with generalized monotonicity and multi-player potential games. In the two settings, step sizes are required to be O(1/k) and O(1/√k) to yield the convergence rates of O(1/ k) and O(1/√k), respectively. Extensive experiments on robust neural network training and sensor network localization validate our theory. Our code is available at https://github.com/GuanpuChen/Global-NE.

Mobile-centric AI applications have high requirements for the resource-efficiency of model inference. Input filtering is a promising approach to eliminate redundancy so as to reduce the cost of inference. Previous efforts have tailored effective solutions for many applications, but left two essential questions unanswered: (1)theoretical filterability of an inference workloadto guide the application of input filtering techniques, thereby avoiding the trial-and-error cost for resource-constrained mobile applications; (2)robust discriminability of feature embeddingto allow input filtering to be widely effective for diverse inference tasks and input content. To answer them, we first formulate the input filtering problem and theoretically compare the hypothesis complexity of inference models and input filters to understand the optimization potential. Then we propose the first end-to-end learnable input filtering framework that covers most state-of-the-art methods and surpasses them in feature embedding with robust discriminability. We design and implementInFithat supports different input modalities and mobile-centric deployments. Comprehensive evaluations confirm our theoretical results and show thatInFioutperforms strong baselines in applicability, accuracy, and efficiency.InFican achieve 8.5× throughput and save 95% bandwidth, while keeping over 90% accuracy, for a video analytics application on mobile platforms.

Online convex optimization in distributed systems has shown great promise in collaboratively learning on data streams with massive learners, such as in collaborative coordination in robot and IoT networks. When implemented in communication-constrained networks like robot and IoT networks, two critical yet distinct objectives in distributed online convex optimization (DOCO) are minimizing the overall regret and the communication cost. Achieving both objectives simultaneously is challenging, especially when the number of learners$n$and learning time$T$are prohibitively large. To address this challenge, we propose novel algorithms in typical adversarial and stochastic settings. Our algorithms significantly reduce the communication complexity of the algorithms with the state-of-the-art regret by a factor of$\mathcal {O}(n^{2})$and$\tilde{\mathcal {O}}(\sqrt{nT})$in adversarial and stochastic settings, respectively. We are the first to achieve nearly optimal regret and communication complexity simultaneously up to polylogarithmic factors. We validate our algorithms through experiments on real-world datasets in classification tasks. Our algorithms with appropriate parameters can achieve$90\%\sim 99\%$communication saving with close accuracy over existing methods in most cases. The code is available athttps://github.com/GGBOND121382/Communication-Efficient_Regret-Optimal_DOCO.

Transformer-based text-to-image synthesis generates images from abstractive textual conditions and achieves prompt results. Since transformer-based models predict visual tokens step by step in testing, where the early error is hard to be corrected and would be propagated. To alleviate this issue, the common practice is drawing multi-paths from the transformer-based models and re-ranking the multi-images decoded from multi-paths to find the best one and filter out others. Therefore, the computing procedure of excluding images may be inefficient. To improve the effectiveness and efficiency of decoding, we exploit a reject decoding algorithm with tiny multi-modal models to enlarge the searching space and exclude the useless paths as early as possible. Specifically, we build tiny multi-modal models to evaluate the similarities between the partial paths and the caption at multi scales. Then, we propose a reject decoding algorithm to exclude some lowest quality partial paths at the inner steps. Thus, under the same computing load as the original decoding, we could search across more multi-paths to improve the decoding efficiency and synthesizing quality. The experiments conducted on the MS-COCO dataset and large-scale datasets show that the proposed reject decoding algorithm can exclude the useless paths and enlarge the searching paths to improve the synthesizing quality by consuming less time.

Learning to optimize (L2O) has gained increasing popularity, which automates the design of optimizers by data-driven approaches. However, current L2O methods often suffer from poor generalization performance in at least two folds: (i) applying the L2O-learned optimizer to unseen optimizees, in terms of lowering their loss function values (optimizer generalization, or “generalizable learning of optimizers”); and (ii) the test performance of an optimizee (itself as a machine learning model), trained by the optimizer, in terms of the accuracy over unseen data (optimizee generalization, or “learning to generalize”). While the optimizer generalization has been recently studied, the optimizee generalization (or learning to generalize) has not been rigorously studied in the L2O context, which is the aim of this paper. We first theoretically establish an implicit connection between the local entropy and the Hessian, and hence unify their roles in the handcrafted design of generalizable optimizers as equivalent metrics of the landscape flatness of loss functions. We then propose to incorporate these two metrics as flatness-aware regularizers into the L2O framework in order to meta-train optimizers to learn to generalize, and theoretically show that such generalization ability can be learned during the L2O meta-training process and then transformed to the optimizee loss function. Extensive experiments consistently validate the effectiveness of our proposals with substantially improved generalization on multiple sophisticated L2O models and diverse optimizees.

"A picture is worth a thousand words", significantly beyond mere a categorization. Accompanied by that, many patches of the image could have completely irrelevant meanings with the categorization if they were independently observed. This could significantly reduce the efficiency of a large family of few-shot learning algorithms, which have limited data and highly rely on the comparison of image patches. To address this issue, we propose a Class-aware Patch Embedding Adaptation (CPEA) method to learn "class-aware embeddings" of the image patches. The key idea of CPEA is to integrate patch embeddings with class-aware embeddings to make them class-relevant. Furthermore, we define a dense score matrix between class-relevant patch embeddings across images, based on which the degree of similarity between paired images is quantified. Visualization results show that CPEA concentrates patch embeddings by class, thus making them class-relevant. Extensive experiments on four benchmark datasets, miniImageNet, tieredImageNet, CIFAR-FS, and FC-100, indicate that our CPEA significantly outperforms the existing state-of-the-art methods. The source code is available at https://github.com/FushengHao/CPEA.

Additive models have been burgeoning in data analysis due to their flexible representation and desirable interpretability. However, most existing approaches are constructed under empirical risk minimization (ERM), and thus perform poorly in situations where average performance is not a suitable criterion for the problems of interest, e.g., data with complex non-Gaussian noise, imbalanced labels or both of them. In this paper, a novel class of sparse additive models is proposed under tilted empirical risk minimization (TERM), which addresses the deficiencies in ERM by imposing tilted impact on individual losses, and is flexibly capable of achieving a variety of learning objectives, e.g., variable selection, robust estimation, imbalanced classification and multiobjective learning. On the theoretical side, a learning theory analysis which is centered around the generalization bound and function approximation error bound (under some specific data distributions) is conducted rigorously. On the practical side, an accelerated optimization algorithm is designed by integrating Prox-SVRG and random Fourier acceleration technique. The empirical assessments verify the competitive performance of our approach on both synthetic and real data.

Decentralized stochastic gradient descent (D-SGD) allows collaborative learning on massive devices simultaneously without the control of a central server. However, existing theories claim that decentralization invariably undermines generalization. In this paper, we challenge the conventional belief and present a completely new perspective for understanding decentralized learning. We prove that D-SGD implicitly minimizes the loss function of an average-direction Sharpness-aware minimization (SAM) algorithm under general non-convex non-$\beta$-smooth settings. This surprising asymptotic equivalence reveals an intrinsic regularization-optimization trade-off and three advantages of decentralization: (1) there exists a free uncertainty evaluation mechanism in D-SGD to improve posterior estimation; (2) D-SGD exhibits a gradient smoothing effect; and (3) the sharpness regularization effect of D-SGD does not decrease as total batch size increases, which justifies the potential generalization benefit of D-SGD over centralized SGD (C-SGD) in large-batch scenarios.

This paper studies multiparty learning, aiming to learn a model using the private data of different participants. Model reuse is a promising solution for multiparty learning, assuming that a local model has been trained for each party. Considering the potential sample selection bias among different parties, some heterogeneous model reuse approaches have been developed. However, although pre-trained local classifiers are utilized in these approaches, the characteristics of the local data are not well exploited. This motivates us to estimate the density of local data and design an auxiliary model together with the local classifiers for reuse. To address the scenarios where some local models are not well pre-trained, we further design a multiparty cross-entropy loss for calibration. Upon existing works, we address a challenging problem of heterogeneous model reuse from a decision theory perspective and take advantage of recent advances in density estimation. Experimental results on both synthetic and benchmark data demonstrate the superiority of the proposed method.

Vision Transformer (ViT) has achieved remarkable performance in computer vision. However, positional encoding in ViT makes it substantially difficult to learn the intrinsic equivariance in data. Ini- tial attempts have been made on designing equiv- ariant ViT but are proved defective in some cases in this paper. To address this issue, we design a Group Equivariant Vision Transformer (GE-ViT) via a novel, effective positional encoding opera- tor. We prove that GE-ViT meets all the theoreti- cal requirements of an equivariant neural network. Comprehensive experiments are conducted on standard benchmark datasets, demonstrating that GE-ViT significantly outperforms non-equivariant self-attention networks. The code is available at https://github.com/ZJUCDSYangKaifan/GEVit.

Complex-valued neural networks (CVNNs) have been widely applied in various fields, primarily in signal processing and image recognition. Few studies have focused on the generalisation of CVNNs, although it is vital to ensure the performance of CVNNs on unseen data. This study is the first to prove a generalisation bound for complex-valued neural networks. The bounds increase as the spectral complexity increases, with the dominant factor being the product of the spectral norms of the weight matrices. Furthermore, this work provides a generalisation bound for CVNNs trained on sequential data, which is also affected by the spectral complexity. Theoretically, these bounds are derived using the Maurey Sparsification Lemma and Dudley entropy integral. We conducted empirical experiments on various datasets including MNIST, ashionMNIST, CIFAR-10, CIFAR-100, Tiny ImageNet, and IMDB by training complex-valued convolutional neural networks. The Spearman rank-order correlation coefficient and the corresponding p-values on these datasets provide strong proof of the statistically significant correlation between the spectral complexity of a network and its generalisation ability, as measured by the spectral norm product of the weight matrices. The code is available at https://github.com/LeavesLei/cvnn_generalization.

Multimodal fusion and multitask learning are two vital topics in machine learning. Despite the fruitful progress, existing methods for both problems are still brittle to the same challenge-it remains dilemmatic to integrate the common information across modalities (resp. tasks) meanwhile preserving the specific patterns of each modality (resp. task). Besides, while they are actually closely related to each other, multimodal fusion and multitask learning are rarely explored within the same methodological framework before. In this paper, we propose Channel-Exchanging-Network (CEN) which is self-adaptive, parameter-free, and more importantly, applicable for multimodal and multitask dense image prediction. At its core, CEN adaptively exchanges channels between subnetworks of different modalities. Specifically, the channel exchanging process is self-guided by individual channel importance that is measured by the magnitude of Batch-Normalization (BN) scaling factor during training. For the application of dense image prediction, the validity of CEN is tested by four different scenarios: multimodal fusion, cycle multimodal fusion, multitask learning, and multimodal multitask learning. Extensive experiments on semantic segmentation via RGB-D data and image translation through multi-domain input verify the effectiveness of CEN compared to state-of-the-art methods. Detailed ablation studies have also been carried out, which demonstrate the advantage of each component we propose. Our code is available at https://github.com/yikaiw/CEN.

Key semantics can come from everywhere on an image. Semantic alignment is a key part of few-shot learning but still remains challenging. In this paper, we design a Mixer-Based Semantic Spread (MBSS) algorithm that employs amixermodule to spread the key semantic on the whole image, so that one can directly compare the processed image pairs. We first adopt a convolutional neural network to extract features from both support and query images and separate each of them into multiple Local Descriptor-based Representations (LDRs). The LDRs are then fed into themixerfor semantic spread, where every LDR attracts complementary information from its peers. In this way, the objective semantic is made spread on the whole image in a data-driven manner. The overall pipeline is supervised by a voting-based loss, guaranteeing a goodmixer. Visualization results validate the feasibility of ourmixer. Comprehensive experiments on three benchmark datasets, miniImageNet, tieredImageNet, and CUB, show that our algorithm achieves the state-of-the-art performance in both 5-way 1-shot and 5-way 5-shot settings.

Text-based image manipulation is a popular subject and has many applications. However, it is a challenging task because there is no ground-truth edited dataset and textual descriptions have abstractive and ambiguous properties. To alleviate the difficult issues, we propose a manipulation framework consisting of the proposal attentional GANs, language-related semantic mask, and language-guided ranker. Specially, we construct an editing proposal generator to generate the suitable edited proposals with and without semantic conditions, which supports the reorganization of sub-generators to output proposals in various aspects as many as possible. To distinguish the text-relevant and the text-irrelevant regions, we introduce a language-related semantic mask based on the source image and target caption. Then, we exploit a language-guided ranker to retrieve the best edited result from the edited proposals through using the multi-modal similarity and the language-related semantic mask. Extensive experiments on widely-used datasets demonstrate that our model could manipulate images interactively and improve the editing quality effectively.

Deep neural networks (DNNs) have greatly contributed to the performance gains in semantic segmentation. Nevertheless, training DNNs generally requires large amounts of pixel-level labeled data, which is expensive and time-consuming to collect in practice. To mitigate the annotation burden, this paper proposes a self-ensembling generative adversarial network (SE-GAN) exploiting cross-domain data for semantic segmentation. In SE-GAN, a teacher network and a student network constitute a self-ensembling model for generating semantic segmentation maps, which together with a discriminator, forms a GAN. Despite its simplicity, we find SE-GAN can significantly boost the performance of adversarial training and enhance the stability of the model, the latter of which is a common barrier shared by most adversarial training-based methods. We theoretically analyze SE-GAN and provide an$\mathcal {O}(1/\sqrt{N})$generalization bound ($N$is the training sample size), which suggests controlling the discriminator's hypothesis complexity to enhance the generalizability. Accordingly, we choose a simple network as the discriminator. Extensive and systematic experiments in two standard settings demonstrate that the proposed method significantly outperforms current state-of-the-art approaches.

Invariance to diverse types of image corruption, such as noise, blurring, or colour shifts, is essential to establish robust models in computer vision. Data augmentation has been the major approach in improving the robustness against common corruptions. However, the samples produced by popular augmentation strategies deviate significantly from the underlying data manifold. As a result, performance is skewed toward certain types of corruption. To address this issue, we propose a multi-source vicinal transfer augmentation (VITA) method for generating diverse on-manifold samples. The proposed VITA consists of two complementary parts: tangent transfer and integration of multi-source vicinal samples. The tangent transfer creates initial augmented samples for improving corruption robustness. The integration employs a generative model to characterize the underlying manifold built by vicinal samples, facilitating the generation of on-manifold samples. Our proposed VITA significantly outperforms the current state-of-the-art augmentation methods, demonstrated in extensive experiments on corruption benchmarks.

Existing Scene Text Recognition (STR) methods typically use a language model to optimize the joint probability of the 1D character sequence predicted by a visual recognition (VR) model, which ignore the 2D spatial context of visual semantics within and between character instances, making them not generalize well to arbitrary shape scene text. To address this issue, we make the first attempt to perform textual reasoning based on visual semantics in this paper. Technically, given the character segmentation maps predicted by a VR model, we construct a subgraph for each instance, where nodes represent the pixels in it and edges are added between nodes based on their spatial similarity. Then, these subgraphs are sequentially connected by their root nodes and merged into a complete graph. Based on this graph, we devise a graph convolutional network for textual reasoning (GTR) by supervising it with a cross-entropy loss. GTR can be easily plugged in representative STR models to improve their performance owing to better textual reasoning. Specifically, we construct our model, namely S-GTR, by paralleling GTR to the language model in a segmentation-based STR baseline, which can effectively exploit the visual-linguistic complementarity via mutual learning. S-GTR sets new state-of-the-art on six challenging STR benchmarks and generalizes well to multi-linguistic datasets. Code is available at https://github.com/adeline-cs/GTR.

Semi-supervised video object segmentation (VOS) refers to segmenting the target object in remaining frames given its annotation in the first frame, which has been actively studied in recent years. The key challenge lies in finding effective ways to exploit the spatio-temporal context of past frames to help learn discriminative target representation of current frame. In this paper, we propose a novel Siamese network with a specifically designed interactive transformer, called SITVOS, to enable effective context propagation from historical to current frames. Technically, we use the transformer encoder and decoder to handle the past frames and current frame separately, i.e., the encoder encodes robust spatio-temporal context of target object from the past frames, while the decoder takes the feature embedding of current frame as the query to retrieve the target from the encoder output. To further enhance the target representation, a feature interaction module (FIM) is devised to promote the information flow between the encoder and decoder. Moreover, we employ the Siamese architecture to extract backbone features of both past and current frames, which enables feature reuse and is more efficient than existing methods. Experimental results on three challenging benchmarks validate the superiority of SITVOS over state-of-the-art methods. Code is available at https://github.com/LANMNG/SITVOS.

3D object detection is a crucial research topic in computer vision, which usually uses 3D point clouds as input in conventional setups. Recently, there is a trend of leveraging multiple sources of input data, such as complementing the 3D point cloud with 2D images that often have richer color and fewer noises. However, due to the heterogeneous geometrics of the 2D and 3D representations, it prevents us from applying off-the-shelf neural networks to achieve multimodal fusion. To that end, we propose Bridged Transformer (BrT), an end-to-end architecture for 3D object detection. BrT is simple and effective, which learns to identify 3D and 2D object bounding boxes from both points and image patches. A key element of BrT lies in the utilization of object queries for bridging 3D and 2D spaces, which unifies different sources of data representations in Transformer. We adopt a form of feature aggregation realized by point-to-patch projections which further strengthen the interaction between images and points. Moreover, BrT works seamlessly for fusing the point cloud with multi-view images. We experimentally show that BrT surpasses state-of-the-art methods on SUN RGB-D and ScanNetV2 datasets.

Object-guided text-to-image synthesis aims to generate images from natural language descriptions built by two-step frameworks, i.e., the model generates the layout and then synthesizes images from the layout and captions. However, such frameworks have two issues: 1) complex structure, since generating language-related layout is not a trivial task; 2) error propagation, because the inappropriate layout will mislead the image synthesis and is hard to be revised. In this paper, we propose an object-guided joint-decoding module to simultaneously generate the image and the corresponding layout. Specially, we present the joint-decoding transformer to model the joint probability on images tokens and the corresponding layouts tokens, where layout tokens provide additional observed data to model the complex scene better. Then, we describe a novel Layout-Vqgan for layout encoding and decoding to provide more information about the complex scene. After that, we present the detail-enhanced module to enrich the language-related details based on two facts: 1) visual details could be omitted in the compression of VQGANs; 2) the joint-decoding transformer would not have sufficient generating capacity. The experiments show that our approach is competitive with previous object-centered models and can generate diverse and high-quality objects under the given layouts.

Personalized federated learning is proposed to handle the data heterogeneity problem amongst clients by learning dedicated tailored local models for each user. However, existing works are often built in a centralized way, leading to high communication pressure and high vulnerability when a failure or an attack on the central server occurs. In this work, we propose a novel personalized federated learning framework in a decentralized (peer-to-peer) communication protocol named DisPFL, which employs personalized sparse masks to customize sparse local models on the edge. To further save the communication and computation cost, we propose a decentralized sparse training technique, which means that each local model in DisPFL only maintains a fixed number of active parameters throughout the whole local training and peer-to-peer communication process. Comprehensive experiments demonstrate that DisPFL significantly saves the communication bottleneck for the busiest node among all clients and, at the same time, achieves higher model accuracy with less computation cost and communication rounds. Furthermore, we demonstrate that our method can easily adapt to heterogeneous local clients with varying computation complexities and achieves better personalized performances.

This paper studies the algorithmic stability and generalizability of decentralized stochastic gradient descent (D-SGD). We prove that the consensus model learned by D-SGD is $\mathcal{O}{(m/N\unaryplus1/m\unaryplus\lambda^2)}$-stable in expectation in the non-convex non-smooth setting, where $N$ is the total sample size of the whole system, $m$ is the worker number, and $1\unaryminus\lambda$ is the spectral gap that measures the connectivity of the communication topology. These results then deliver an $\mathcal{O}{(1/N\unaryplus{({(m^{-1}\lambda^2)}^{\frac{\alpha}{2}}\unaryplus m^{\unaryminus\alpha})}/{N^{1\unaryminus\frac{\alpha}{2}}})}$ in-average generalization bound, which is non-vacuous even when $\lambda$ is closed to $1$, in contrast to vacuous as suggested by existing literature on the projected version of D-SGD. Our theory indicates that the generalizability of D-SGD has a positive correlation with the spectral gap, and can explain why consensus control in initial training phase can ensure better generalization. Experiments of VGG-11 and ResNet-18 on CIFAR-10, CIFAR-100 and Tiny-ImageNet justify our theory. To our best knowledge, this is the first work on the topology-aware generalization of vanilla D-SGD. Code is available at \url{https://github.com/Raiden-Zhu/Generalization-of-DSGD}.

Multi-source domain adaptation has attracted great attention in machine learning community. Most of these methods focus on weighting the predictions produced by the adaptation networks of different domains. Thus the domain shifts between certain of domains and target domain are not effectively relieved, resulting in that these domains are not fully exploited and even may have a negative influence on multi-source domain adaptation task. To address such challenge, we propose a multi-source domain adaptation method to gradually improve the adaptation ability of each source domain by producing more high-confident pseudo-labels with self-paced learning for conditional distribution alignment. The proposed method first trains several separate domain branch networks with single domains and an ensemble branch network with all domains. Then we obtain some high-confident pseudo-labels with the branch networks and learn the branch specific pseudo-labels with self-paced learning. Each branch network reduces the domain gap by aligning the conditional distribution with its branch specific pseudo-labels and the pseudo-labels provided by all branch networks. Experiments on Office31, Office-Home and DomainNet show that the proposed method outperforms the state-of-the-art methods.

Mobile-centric AI applications put forward high requirements for resource-efficiency of model inference. Input filtering is a promising approach to eliminate the redundancy in the input so as to reduce the cost of inference. Previous efforts have tailored effective solutions for many applications, but left two essential questions unanswered: (1) theoretical filterability of an inference workload to guide the application of input filtering techniques, thereby avoiding the trial-and-error cost for resource-constrained mobile applications; (2) robust discriminability of feature embedding to allow input filtering to be widely effective for diverse inference tasks and input content. To answer these questions, we first provide a generic formalization of the input filtering problem and theoretically compare the hypothesis complexity of inference models and their input filters to understand the optimization potential of applying input filtering. Then we propose the first end-to-end learnable input filtering framework that covers most state-of-the-art methods and surpasses them in feature embedding with robust discriminability. Based on our framework, we design and implement an input filtering system InFi supporting six input modalities. InFi is the first to support text and sensor signal inputs and model partitioning deployments widely adopted by under-resourced mobile systems. Comprehensive evaluations confirm our theoretical results and show that InFi outperforms strong baselines in applicability, accuracy, and efficiency, owing to its generality and end-to-end learnability. InFi can achieve 8.5X throughput and save 95% bandwidth, while keeping over 90% accuracy, for a video analytics app on mobile platforms.

Designing and analyzing model-based RL (MBRL) algorithms with guaranteed monotonic improvement has been challenging, mainly due to the interdependence between policy optimization and model learning. Existing discrepancy bounds generally ignore the impacts of model shifts, and their corresponding algorithms are prone to degrade performance by drastic model updating. In this work, we first propose a novel and general theoretical scheme for a non-decreasing performance guarantee of MBRL. Our follow-up derived bounds reveal the relationship between model shifts and performance improvement. These discoveries encourage us to formulate a constrained lower-bound optimization problem to permit the monotonicity of MBRL. A further example demonstrates that learning models from a dynamically-varying number of explorations benefit the eventual returns. Motivated by these analyses, we design a simple but effective algorithm CMLO (Constrained Model-shift Lower-bound Optimization), by introducing an event-triggered mechanism that flexibly determines when to update the model. Experiments show that CMLO surpasses other state-of-the-art methods and produces a boost when various policy optimization methods are employed.

Designing an incentive-compatible auction mechanism that maximizes the auctioneer's revenue while minimizes the bidders’ ex-post regret is an important yet intricate problem in economics. Remarkable progress has been achieved through learning the optimal auction mechanism by neural networks. In this paper, we consider the popular additive valuation and symmetric valuation setting; i.e., the valuation for a set of items is defined as the sum of all items’ valuations in the set, and the valuation distribution is invariant when the bidders and/or the items are permutated. We prove that permutation-equivariant neural networks have significant advantages: the permutation-equivariance decreases the expected ex-post regret, improves the model generalizability, while maintains the expected revenue invariant. This implies that the permutation-equivariance helps approach the theoretically optimal dominant strategy incentive compatible condition, and reduces the required sample complexity for desired generalization. Extensive experiments fully support our theory. To our best knowledge, this is the first work towards understanding the benefits of permutation-equivariance in auction mechanisms.

Data augmentation is a critical contributing factor to the success of deep learning but heavily relies on prior domain knowledge which is not always available. Recent works on automatic data augmentation learn a policy to form a sequence of augmentation operations, which are still pre-defined and restricted to limited options. In this paper, we show that a prior-free autonomous data augmentation's objective can be derived from a representation learning principle that aims to preserve the minimum sufficient information of the labels. Given an example, the objective aims at creating a distant ``hard positive example'' as the augmentation, while still preserving the original label. We then propose a practical surrogate to the objective that can be optimized efficiently and integrated seamlessly into existing methods for a broad class of machine learning tasks, e.g., supervised, semi-supervised, and noisy-label learning. Unlike previous works, our method does not require training an extra generative model but instead leverages the intermediate layer representations of the end-task model for generating data augmentations. In experiments, we show that our method consistently brings non-trivial improvements to the three aforementioned learning tasks from both efficiency and final performance, either or not combined with pre-defined augmentations, e.g., on medical images when domain knowledge is unavailable and the existing augmentation techniques perform poorly. Code will be released publicly.

Few-shot learning aims to recognize novel classes from only a few labeled training examples. Aligning semantically relevant local regions has shown promise in effectively comparing a query image with support images. However, global information is usually overlooked in the existing approaches, resulting in a higher possibility of learning semantics unrelated to the global information. To address this issue, we propose a Global-Local Interplay Metric Learning (GLIML) framework to employ the interplay between global features and local features to guide semantic alignment. We first design a Global-Local Information Concurrent Learning (GLICL) module to extract both global features and local features and perform global-local interplay. We then design a Global-Local Information Cross-Covariance Estimator (GLICCE) to learn the similarity on the global-local interplay, in contrast to the current practice where only local features are considered. Visualizations show that the global-local interplay decreases (1) the weights placed on the semantics that are irrelevant to the global information and (2) the variability of the learned features within every class in the feature space. Quantitative experiments on three benchmark datasets demonstrate that GLIML achieves state-of-the-art performance while maintaining high efficiency.

Detection transformers have recently shown promising object detection results and attracted increasing attention. However, how to develop effective domain adaptation techniques to improve its cross-domain performance remains unexplored and unclear. In this paper, we delve into this topic and empirically find that direct feature distribution alignment on the CNN backbone only brings limited improvements, as it does not guarantee domain-invariant sequence features in the transformer for prediction. To address this issue, we propose a novel Sequence Feature Alignment (SFA) method that is specially designed for the adaptation of detection transformers. Technically, SFA consists of a domain query-based feature alignment (DQFA) module and a token-wise feature alignment (TDA) module. In DQFA, a novel domain query is used to aggregate and align global context from the token sequence of both domains. DQFA reduces the domain discrepancy in global feature representations and object relations when deploying in the transformer encoder and decoder, respectively. Meanwhile, TDA aligns token features in the sequence from both domains, which reduces the domain gaps in local and instance-level feature representations in the transformer encoder and decoder, respectively. Besides, a novel bipartite matching consistency loss is proposed to enhance the feature discriminability for robust object detection. Experiments on three challenging benchmarks show that SFA outperforms state-of-the-art domain adaptive object detection methods. Code has been made available at: https://github.com/encounter1997/SFA.

This paper studies the relationship between generalization and privacy preservation of machine learning in two steps. We first establish an alignment between the two facets for any learning algorithm. We prove that $(\varepsilon, \delta)$-differential privacy implies an on-average generalization bound for a multi-sample-set learning algorithm, which further leads to a high-probability bound for any learning algorithm. We then investigate how the iterative nature shared by most learning algorithms influences privacy preservation and further generalization. Three composition theorems are proved to approximate the differential privacy of an iterative algorithm through the differential privacy of its every iteration. Integrating the above two steps, we eventually deliver generalization bounds for iterative learning algorithms. Our results are strictly tighter than the existing works. Particularly, our generalization bounds do not rely on the model size which is prohibitively large in deep learning. Experiments of MLP, VGG, and ResNet on MNIST, CIFAR-10, and CIFAR-100 are in full agreement with our theory. The theory applies to a wide spectrum of learning algorithms. In this paper, it is applied to the Gaussian mechanism as an example.

Deep learning is often criticized by two serious issues that rarely exist in natural nervous systems: overfitting and catastrophic forgetting. It can even memorize randomly labeled data, which has little knowledge behind the instance-label pairs. When a deep network continually learns over time by accommodating new tasks, it usually quickly overwrites the knowledge learned from previous tasks. Referred to as the neural variability, it is well known in neuroscience that human brain reactions exhibit substantial variability even in response to the same stimulus. This mechanism balances accuracy and plasticity/flexibility in the motor learning of natural nervous systems. Thus, it motivates us to design a similar mechanism, named artificial neural variability (ANV), that helps artificial neural networks learn some advantages from "natural" neural networks. We rigorously prove that ANV plays as an implicit regularizer of the mutual information between the training data and the learned model. This result theoretically guarantees ANV a strictly improved generalizability, robustness to label noise, and robustness to catastrophic forgetting. We then devise a neural variable risk minimization (NVRM) framework and neural variable optimizers to achieve ANV for conventional network architectures in practice. The empirical studies demonstrate that NVRM can effectively relieve overfitting, label noise memorization, and catastrophic forgetting at negligible costs.

Residual connections significantly boost the performance of deep neural networks. However, few theoretical results address the influence of residuals on the hypothesis complexity and the generalization ability of deep neural networks. This article studies the influence of residual connections on the hypothesis complexity of the neural network in terms of the covering number of its hypothesis space. We first present an upper bound of the covering number of networks with residual connections. This bound shares a similar structure with that of neural networks without residual connections. This result suggests that moving a weight matrix or nonlinear activation from the bone to a vine would not increase the hypothesis space. Afterward, an O(1 / √N) margin-based multiclass generalization bound is obtained for ResNet, as an exemplary case of any deep neural network with residual connections. Generalization guarantees for similar state-of-the-art neural network architectures, such as DenseNet and ResNeXt, are straightforward. According to the obtained generalization bound, we should introduce regularization terms to control the magnitude of the norms of weight matrices not to increase too much, in practice, to ensure a good generalization ability, which justifies the technique of weight decay.

Recently, deep learning based video super-resolution (SR) methods have achieved promising performance. To simultaneously exploit the spatial and temporal information of videos, employing 3-dimensional (3D) convolutions is a natural approach. However, straight utilizing 3D convolutions may lead to an excessively high computational complexity which restricts the depth of video SR models and thus undermine the performance. In this paper, we present a novel fast spatio-temporal residual network (FSTRN) to adopt 3D convolutions for the video SR task in order to enhance the performance while maintaining a low computational load. Specifically, we propose a fast spatio-temporal residual block (FRB) that divide each 3D filter to the product of two 3D filters, which have considerably lower dimensions. Furthermore, we design a cross-space residual learning that directly links the low-resolution space and the high-resolution space, which can greatly relieve the computational burden on the feature fusion and up-scaling parts. Extensive evaluations and comparisons on benchmark datasets validate the strengths of the proposed approach and demonstrate that the proposed network significantly outperforms the current state-of-the-art methods.

Few-shot learning aims to learn latent patterns from few training examples and has shown promises in practice. However, directly calculating the distances between the query image and support image in existing methods may cause ambiguity because dominant objects can locate anywhere on images. To address this issue, this paper proposes a Semantic Alignment Metric Learning (SAML) method for few-shot learning that aligns the semantically relevant dominant objects through a "collect-and-select'' strategy. Specifically, we first calculate a relation matrix (RM) to "`collect" the distances of each local region pairs of the 3D tensor extracted from a query image and the mean tensor of the support images. Then, the attention technique is adapted to "select" the semantically relevant pairs and put more weights on them. Afterwards, a multi-layer perceptron (MLP) is utilized to map the reweighted RMs to their corresponding similarity scores. Theoretical analysis demonstrates the generalization ability of SAML and gives a theoretical guarantee. Empirical results demonstrate that semantic alignment is achieved. Extensive experiments on benchmark datasets validate the strengths of the proposed approach and demonstrate that SAML significantly outperforms the current state-of-the-art methods. The source code is available at https://github.com/haofusheng/SAML.

Inpainting methods aim to restore missing parts of corrupted images and play a critical role in many computer vision applications, such as object removal and image restoration. Although existing methods perform well on images with small holes, restoring large holes remains elusive. To address this issue, this paper proposes a Progressive Reconstruction of Visual Structure (PRVS) network that progressively reconstructs the structures and the associated visual feature. Specifically, we design a novel Visual Structure Reconstruction (VSR) layer to entangle reconstructions of the visual structure and visual feature, which benefits each other by sharing parameters. We repeatedly stack four VSR layers in both encoding and decoding stages of a U-Net like architecture to form the generator of a generative adversarial network (GAN) for restoring images with either small or large holes. We prove the generalization error upper bound of the PRVS network is O(1\sqrt(N)), which theoretically guarantees its performance. Extensive empirical evaluations and comparisons on Places2, Paris Street View and CelebA datasets validate the strengths of the proposed approach and demonstrate that the model outperforms current state-of-the-art methods. The source code package is available at https://github.com/jingyuanli001/PRVS-Image-Inpainting.

Deep neural networks have received dramatic success based on the optimization method of stochastic gradient descent (SGD). However, it is still not clear how to tune hyper-parameters, especially batch size and learning rate, to ensure good generalization. This paper reports both theoretical and empirical evidence of a training strategy that we should control the ratio of batch size to learning rate not too large to achieve a good generalization ability. Specifically, we prove a PAC-Bayes generalization bound for neural networks trained by SGD, which has a positive correlation with the ratio of batch size to learning rate. This correlation builds the theoretical foundation of the training strategy. Furthermore, we conduct a large-scale experiment to verify the correlation and training strategy. We trained 1,600 models based on architectures ResNet-110, and VGG-19 with datasets CIFAR-10 and CIFAR-100 while strictly control unrelated variables. Accuracies on the test sets are collected for the evaluation. Spearman's rank-order correlation coefficients and the corresponding $p$ values on 164 groups of the collected data demonstrate that the correlation is statistically significant, which fully supports the training strategy.