Shaoan Xie, Lingjing Kong, Xiangchen Song, Xinshuai Dong, Guangyi Chen, Eric P. Xing, Kun Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Contrastive Language-Image Pre-training (CLIP) [37] has emerged as a pivotal model in computer vision and multimodal learning, achieving state-of-the-art performance at aligning visual and textual representations through contrastive learning. However, CLIP struggles with potential information misalignment in many image-text datasets and suffers from entangled representation. On the one hand, short captions for a single image in datasets like MSCOCO may describe disjoint regions in the image, leaving the model uncertain about which visual features to retain or disregard. On the other hand, directly aligning long captions with images can lead to the retention of entangled details, preventing the model from learning disentangled, atomic concepts – ultimately limiting its generalization on certain downstream tasks involving short prompts.In this paper, we establish theoretical conditions that enable flexible alignment between textual and visual representations across varying levels of granularity. Specifically, our framework ensures that a model can not only preserve cross-modal semantic information in its entirety but also disentangle visual representations to capture finegrained textual concepts. Building on this foundation, we introduce SmartCLIP, a novel approach that identifies and aligns the most relevant visual and textual representations in a modular manner. Superior performance across various tasks demonstrates its capability to handle information misalignment and supports our identification theory. The code is available at https://github.com/MidPush/SmartCLIP.

Disentangled representation learning aims to uncover the latent variables underlying observed data, yet identifying these variables under mild assumptions remains challenging. Some methods rely on sufficient changes in the distribution of latent variables indicated by auxiliary variables, such as domain indices, but acquiring enough domains is often impractical. Alternative approaches exploit the structural sparsity assumption on mixing processes, but this constraint may not hold in practice. Interestingly, we find that these two seemingly unrelated assumptions can actually complement each other. Specifically, when conditioned on auxiliary variables, the sparse mixing process induces independence between latent and observed variables, which simplifies the mapping from estimated to true latent variables and hence compensates for deficiencies of auxiliary variables. Building on this insight, we propose an identifiability theory with less restrictive constraints regarding the auxiliary variables and the sparse mixing process, enhancing applicability to real-world scenarios. Additionally, we develop a generative model framework incorporating a domain encoding network and a sparse mixing constraint and provide two implementations based on variational autoencoders and generative adversarial networks. Experiment results on synthetic and real-world datasets support our theoretical results.

Temporally causal representation learning aims to identify the latent causal process from time series observations, but most methods require the assumption that the latent causal processes do not have instantaneous relations. Although some recent methods achieve identifiability in the instantaneous causality case, they require either interventions on the latent variables or grouping of the observations, which are in general difficult to obtain in real-world scenarios. To fill this gap, we propose an \textbf{ID}entification framework for instantane\textbf{O}us \textbf{L}atent dynamics (\textbf{IDOL}) by imposing a sparse influence constraint that the latent causal processes have sparse time-delayed and instantaneous relations. Specifically, we establish identifiability results of the latent causal process based on sufficient variability and the sparse influence constraint by employing contextual information of time series data. Based on these theories, we incorporate a temporally variational inference architecture to estimate the latent variables and a gradient-based sparsity regularization to identify the latent causal process. Experimental results on simulation datasets illustrate that our method can identify the latent causal process. Furthermore, evaluations on multiple human motion forecasting benchmarks with instantaneous dependencies indicate the effectiveness of our method in real-world settings.

Identifying time-delayed latent causal process is crucial for understanding temporal dynamics and enabling downstream reasoning. While recent methods have made progress in identifying latent time-delayed causal processes, they cannot address the dynamics in which the influence of some latent factors on both the subsequent latent states and the observed data can become inactive or irrelevant at different time steps. Therefore, we introduce intermittent temporal latent processes, where: (1) any subset of latent factors may be missing during nonlinear data generation at any time step, and (2) the active latent factors at each step are unknown. This framework encompasses both nonstationary and stationary transitions, accommodating changing or consistent active factors over time. Our work shows that under certain assumptions, the latent causal variables are block-wise identifiable. With further conditional independence assumption, each latent variable can even be recovered up to component-wise transformations. Using this identification theory, we propose an unsupervised approach, InterLatent, to reliably uncover the representations of the intermittent temporal latent process. The experimental findings on both synthetic and real-world datasets verify our theoretical claims.

Prevalent in biomedical applications (e.g., human phenotype research), multimodal datasets can provide valuable insights into the underlying physiological mechanisms. However, current machine learning (ML) models designed to analyze these datasets often lack interpretability and identifiability guarantees, which are essential for biomedical research. Recent advances in causal representation learning have shown promise in identifying interpretable latent causal variables with formal theoretical guarantees. Unfortunately, most current work on multimodal distributions either relies on restrictive parametric assumptions or yields only coarse identification results, limiting their applicability to biomedical research that favors a detailed understanding of the mechanisms. In this work, we aim to develop flexible identification conditions for multimodal data and principled methods to facilitate the understanding of biomedical datasets. Theoretically, we consider a nonparametric latent distribution (c.f., parametric assumptions in previous work) that allows for causal relationships across potentially different modalities. We establish identifiability guarantees for each latent component, extending the subspace identification results from previous work. Our key theoretical contribution is the structural sparsity of causal connections between modalities, which, as we will discuss, is natural for a large collection of biomedical systems. Empirically, we present a practical framework to instantiate our theoretical insights. We demonstrate the effectiveness of our approach through extensive experiments on both numerical and synthetic datasets. Results on a real-world human phenotype dataset are consistent with established biomedical research, validating our theoretical and methodological framework.

The autoregressive decoding for text generation in large language models (LLMs), while widely used, is inherently suboptimal due to the lack of a built-in mechanism to perform refinement and/or correction of the generated content. In this paper, we consider optimality in terms of the joint probability over the generated response, when jointly considering all tokens at the same time. We theoretically characterize the potential deviation of the autoregressively generated response from its globally optimal counterpart that is of the same length. Our analysis suggests that we need to be cautious when noticeable uncertainty arises during text generation, which may signal the sub-optimality of the generation history. To address the pitfall of autoregressive decoding for text generation, we propose an approach that incorporates a sliding reflection window and a pausing criterion, such that refinement and generation can be carried out interchangeably as the decoding proceeds. Our selective refinement framework strikes a balance between efficiency and optimality, and our extensive experimental results demonstrate the effectiveness of our approach.

A central problem in unsupervised domain adaptation is determining what to transfer from labeled source domains to an unlabeled target domain. To handle high-dimensional observations (e.g., images), a line of approaches use deep learning to learn latent representations of the observations, which facilitate knowledge transfer in the latent space. However, existing approaches often rely on restrictive assumptions to establish identifiability of the joint distribution in the target domain, such as independent latent variables or invariant label distributions, limiting their real-world applicability. In this work, we propose a general domain adaptation framework that learns compact latent representations to capture distribution shifts relative to the prediction task and address the fundamental question of what representations should be learned and transferred. Notably, we first demonstrate that learning representations based on all the predictive information, i.e., the label's Markov blanket in terms of the learned representations, is often underspecified in general settings. Instead, we show that, interestingly, general domain adaptation can be achieved by partitioning the representations of Markov blanket into those of the label's parents, children, and spouses. Moreover, its identifiability guarantee can be established. Building on these theoretical insights, we develop a practical, nonparametric approach for domain adaptation in a general setting, which can handle different types of distribution shifts.

Concept learning seeks to extract semantic and interpretable representations of atomic concepts from high-dimensional data such as images and text, which can be instrumental to a variety of downstream tasks (e.g., image generation/editing). Despite its importance, the theoretical foundations for learning atomic concepts and their interactions, especially from multimodal distributions, remain underexplored. In this work, we establish fundamental conditions for learning atomic multimodal concepts and their underlying interactions With identfiability guarantees. We formulate concept learning as a latent variable identification problem, representing atomic concepts in each modality as latent variables, with a graphical model to specify their interactions across modalities. Our theoretical contribution is to provide component-wise identifiability of atomic concepts under flexible, nonparametric conditions that accommodate both continuous and discrete modalities. Building on these theoretical insights, we demonstrate the practical utility of our theory in a downstream task text-to-image (T2I) generation. We develop a principled T2I model that explicitly learns atomic textual and visual concepts with sparse connections between them, allowing us to achieve image generation and editing at the atomic concept level. Empirical evaluations show that our model outperforms existing methods in T2I generation tasks, offering superior controllability and interpretability.

Causal Representation Learning (CRL) aims to uncover the data-generating process and identify the underlying causal variables and relations, whose evaluation remains inherently challenging due to the requirement of known ground-truth causal variables and causal structure. Existing evaluations often rely on either simplistic synthetic datasets or downstream performance on real-world tasks, generally suffering a dilemma between realism and evaluative precision. In this paper, we introduce a new benchmark for CRL using high-fidelity simulated visual data that retains both realistic visual complexity and, more importantly, access to ground-truth causal generating processes. The dataset comprises around 200 thousand images and 3 million video frames across 24 sub-scenes in four domains: static image generation, dynamic physical simulations, robotic manipulations, and traffic situation analysis. These scenarios range from static to dynamic settings, simple to complex structures, and single to multi-agent interactions, offering a comprehensive testbed that hopefully bridges the gap between rigorous evaluation and real-world applicability. In addition, we provide flexible access to the underlying causal structures, allowing users to modify or configure them to align with the required assumptions in CRL, such as available domain labels, temporal dependencies, or intervention histories. Leveraging this benchmark, we evaluated representative CRL methods across diverse paradigms and offered empirical insights to assist practitioners and newcomers in choosing or extending appropriate CRL frameworks to properly address specific types of real problems that can benefit from the CRL perspective. Welcome to visit our: Project page: https://causal-verse.github.io/ , Dataset: https://huggingface.co/CausalVerse

Vision-Language Models (VLMs) integrate visual knowledge with the analytical capabilities of Large Language Models (LLMs) through supervised visual instruction tuning, using image-question-answer triplets. However, the potential of VLMs trained without supervised instruction remains largely unexplored. This study validates that VLMs possess inherent self-refinement capabilities, enabling them to generate high-quality supervised data without external inputs and thereby learn autonomously. Specifically, to stimulate the self-refinement ability of VLMs, we propose a self-refinement framework based on a Triangular Consistency principle: within the image-query-answer triangle, any masked elements should be consistently and accurately reconstructed. The framework involves three steps: (1) We enable the instruction generation ability of VLMs by adding multi-task instruction tuning like image$\rightarrow$question-answer or image-answer$\rightarrow$question. (2) We generate image-query-answer triplets from unlabeled images and use the Triangular Consistency principle for filtering. (3) The model is further updated using the filtered synthetic data. To investigate the underlying mechanisms behind this self-refinement capability, we conduct a theoretical analysis from a causal perspective. Using the widely recognized LLaVA-1.5 as our baseline, our experiments reveal that the model can autonomously achieve consistent, though deliberately modest, improvements across multiple benchmarks without any external supervision, such as human annotations or environmental feedback. We expect that the insights of this study on the self-refinement ability of VLMs can inspire future research on the learning mechanism of VLMs. Code is available at https://github.com/dengyl20/SRF-LLaVA-1.5.

Modeling hierarchical latent dynamics behind time series data is critical for capturing temporal dependencies across multiple levels of abstraction in real-world tasks. However, existing temporal causal representation learning methods fail to capture such dynamics, as they fail to recover the joint distribution of hierarchical latent variables from \textit{single-timestep observed variables}. Interestingly, we find that the joint distribution of hierarchical latent variables can be uniquely determined using three conditionally independent observations. Building on this insight, we propose a Causally Hierarchical Latent Dynamic (CHiLD) identification framework. Our approach first employs temporal contextual observed variables to identify the joint distribution of multi-layer latent variables. Sequentially, we exploit the natural sparsity of the hierarchical structure among latent variables to identify latent variables within each layer. Guided by the theoretical results, we develop a time series generative model grounded in variational inference. This model incorporates a contextual encoder to reconstruct multi-layer latent variables and normalize flow-based hierarchical prior networks to impose the independent noise condition of hierarchical latent dynamics. Empirical evaluations on both synthetic and real-world datasets validate our theoretical claims and demonstrate the effectiveness of CHiLD in modeling hierarchical latent dynamics.

This paper is concerned with online time series forecasting, where unknown distribution shifts occur over time, i.e., latent variables influence the mapping from historical to future observations. To develop an automated way of online time series forecasting, we propose a Theoretical framework for Online Time-series forecasting (TOT in short) with theoretical guarantees. Specifically, we prove that supplying a forecaster with latent variables tightens the Bayes risk—the benefit endures under estimation uncertainty of latent variables and grows as the latent variables achieve a more precise identifiability. To better introduce latent variables into online forecasting algorithms, we further propose to identify latent variables with minimal adjacent observations. Based on these results, we devise a model-agnostic blueprint by employing a temporal decoder to match the distribution of observed variables and two independent noise estimators to model the causal inference of latent variables and mixing procedures of observed variables, respectively. Experiment results on synthetic data support our theoretical claims. Moreover, plug-in implementations built on several baselines yield general improvement across multiple benchmarks, highlighting the effectiveness in real-world applications.

Vector-Quantized Variational Autoencoders (VQVAEs) have enabled strong performance in generative modeling by mapping continuous data to learnable codes. In this work, we identify a surprising yet consistent phenomenon that we term \emph{dimensional collapse}: despite using high-dimensional embeddings, VQVAEs tend to compress their representations into a much smaller subspace, typically only 4 to 10 dimensions. We provide an in-depth analysis of this phenomenon and reveal its relation to model performance and learning dynamics. Interestingly, VQVAEs naturally gravitate toward this low-dimensional regime, and enforcing higher-dimensional usage (e.g., via rank regularization) could lead to degraded performance. To overcome this low-dimensionality limitation, we propose \textbf{Divide-and-Conquer VQ (DCVQ)}, which partitions the latent space into multiple low-dimensional subspaces, each quantized independently. By design, each subspace respects the model’s preference for low dimensionality, while their combination expands the overall capacity. Our results show that DCVQ overcomes the inherent dimensional bottleneck and achieves improved reconstruction quality across image datasets.

Multi-modal time series data is common in web technologies like the Internet of Things (IoT). Existing methods for multi-modal time series representation learning aim to disentangle the modality-shared and modality-specific latent variables. Although achieving notable performances on downstream tasks, they usually assume an orthogonal latent space. However, the modality-specific and modality-shared latent variables might be dependent on real-world scenarios. Therefore, we propose a general generation process, where the modality-shared and modality-specific latent variables are dependent, and further develop a Multi-modAl TEmporal Disentanglement (MATE) model. Specifically, our MATE model is built on a temporally variational inference architecture with the modality-shared and modality-specific prior networks for the disentanglement of latent variables. Furthermore, we establish identifiability results to show that the extracted representation is disentangled. More specifically, we first achieve the subspace identifiability for modality-shared and modality-specific latent variables by leveraging the pairing of multi-modal data. Then we establish the component-wise identifiability of modality-specific latent variables by employing sufficient changes of historical latent variables. Extensive experimental studies on 12 datasets show a general improvement in different downstream tasks, highlighting the effectiveness of our method in real-world scenarios.

Although graph neural networks have achieved great success in the task of molecular property prediction in recent years, their generalization ability under out-of-distribution (OOD) settings is still under-explored. Most of the existing methods rely on learning discriminative representations for prediction, often assuming that the underlying semantic components are correctly identified. However, this assumption does not always hold, leading to potential misidentifications that affect model robustness. Different from these discriminative-based methods, we propose a generative model to ensure the Semantic-Components Identifiability, named SCI. We demonstrate that the latent variables in this generative model can be explicitly identified into semantic-relevant (SR) and semantic-irrelevant (SI) components, which contributes to better OOD generalization by involving minimal change properties of causal mechanisms. Specifically, we first formulate the data generation process from the atom level to the molecular level, where the latent space is split into SI substructures, SR substructures, and SR atom variables. Sequentially, to reduce misidentification, we restrict the minimal changes of the SR atom variables and add a semantic latent substructure regularization to mitigate the variance of the SR substructure under augmented domain changes. Under mild assumptions, we prove the block-wise identifiability of the SR substructure and the comment-wise identifiability of SR atom variables. Experimental studies achieve state-of-the-art performance and show general improvement on 21 datasets in 3 mainstream benchmarks. Moreover, the visualization results of the proposed SCI method provide insightful case studies and explanations for the prediction results.

Large-scale pre-trained Vision Language Models (VLMs) have proven effective for zero-shot classification. Despite the success, most traditional VLMs-based methods are restricted by the assumption of partial source supervision or ideal target vocabularies, which rarely satisfy the open-world scenario. In this paper, we aim at a more challenging setting, Realistic Zero-Shot Classification, which assumes no annotation but instead a broad vocabulary. To address the new problem, we propose the Self Structural Semantic Alignment (S3A) framework, which extracts the structural semantic information from unlabeled data while simultaneously self-learning. Our S3A framework adopts a unique Cluster-Vote-Prompt-Realign (CVPR) algorithm, which iteratively groups unlabeled data to derive structural semantics for pseudo-supervision. Our CVPR algorithm includes iterative clustering on images, voting within each cluster to identify initial class candidates from the vocabulary, generating discriminative prompts with large language models to discern confusing candidates, and realigning images and the vocabulary as structural semantic alignment. Finally, we propose to self-train the CLIP image encoder with both individual and structural semantic alignment through a teacher-student learning strategy. Our comprehensive experiments across various generic and fine-grained benchmarks demonstrate that the S3A method substantially improves over existing VLMs-based approaches, achieving a more than 15% accuracy improvement over CLIP on average. Our codes, models, and prompts are publicly released at https://github.com/sheng-eatamath/S3A.

In this paper, we investigate a new problem called narrative action evaluation (NAE). NAE aims to generate professional commentary that evaluates the execution of an action. Unlike traditional tasks such as score-based action qual-ity assessment and video captioning involving superficial sentences, NAE focuses on creating detailed narratives in natural language. These narratives provide intricate descriptions of actions along with objective evaluations. NAE is a more challenging task because it requires both narrative flex-ibility and evaluation rigor. One existing possible solution is to use multi-task learning, where narrative language and evaluative information are predicted separately. However, this approach results in reduced performance for individual tasks because of variations between tasks and differences in modality between language information and evaluation information. To address this, we propose a prompt-guided multimodal interaction framework. This framework utilizes a pair of transformers to facilitate the interaction between different modalities of information. It also uses prompts to transform the score regression task into a video-text matching task, thus enabling task interactivity. To support further research in this field, we re-annotate the MTL-AQA and FineGym datasets with high-quality and comprehensive action narration. Additionally, we establish benchmarks for NAE. Extensive experiment results prove that our method outperforms separate learning methods and naive multi-task learning methods. Data and code are released at here.

To better interpret the intrinsic mechanism of large language models (LLMs), recent studies focus on monosemanticity on its basic units.A monosemantic neuron is dedicated to a single and specific concept, which forms a oneto-one correlation between neurons and concepts.Despite extensive research in monosemanticity probing, it remains unclear whether monosemanticity is beneficial or harmful to model capacity.To explore this question, we revisit monosemanticity from the feature decorrelation perspective and advocate for its encouragement.We experimentally observe that the current conclusion by Wang et al. ( 2024), which suggests that decreasing monosemanticity enhances model performance, does not hold when the model changes.Instead, we demonstrate that monosemanticity consistently exhibits a positive correlation with model capacity, in the preference alignment process.Consequently, we apply feature correlation as a proxy for monosemanticity and incorporate a feature decorrelation regularizer into the dynamic preference optimization process.The experiments show that our method not only enhances representation diversity but also improves preference alignment performance 1 .

Composing simple actions into complex actions is crucial yet challenging. Existing methods largely rely on language annotations to discern composable latent semantics, which is costly and labor-intensive. In this study, we introduce a novel framework to generate compositional actions without language auxiliaries. Our approach consists of three components: Action Coupling, Conditional Action Generation, and Decoupling Refinement. Action Coupling integrates two sub-actions to generate pseudo-training examples. Then, a conditional generative model, CVAE is employed to facilitate the diverse generation. Decoupling Refinement leverages a self-supervised pre-trained model MAE to ensure semantic consistency between sub-actions and compositional actions. Due to the lack of existing datasets containing both sub-actions and compositional actions, we create two new datasets, named HumanAct-C and UESTC-C. Both qualitative and quantitative assessments are conducted to show our efficacy.1

Current approaches to Video Question Answering (VideoQA) primarily focus on cross-modality matching, which is limited by the requirement for extensive data annotations and the insufficient capacity for causal reasoning (e.g. attributing accidents). To address these challenges, we introduce a causal framework for video reasoning, termed Learning Latent Causal Processes (LLCP). At the heart of LLCP lies a multivariate generative model designed to analyze the spatial-temporal dynamics of objects within events. Leveraging the inherent modularity of causal mechanisms, we train the model through self-supervised local auto-regression eliminating the need for annotated question-answer pairs. During inference, the model is applied to answer two types of reasoning questions: accident attribution, which infers the cause from observed effects, and counterfactual prediction, which predicts the effects of counterfactual conditions given the factual evidence. In the first scenario, we identify variables that deviate from the established distribution by the learned model, signifying the root cause of accidents. In the second scenario, we replace embeddings of previous variables with counterfactual ones, enabling us to forecast potential developments. Once we have identified these cause/effect variables, natural language answers are derived through a combination of grammatical parsing and a pre-trained vision-language model. We assess the efficacy of LLCP on both synthetic and real-world data, demonstrating comparable performance to supervised methods despite our framework using no paired textual annotations.

Conventional causal discovery approaches, which seek to uncover causal relationships among measured variables, are typically fragile to the presence of latent variables. While various methods have been developed to address this confounding issue, they often rely on strong assumptions about the underlying causal structure. In this paper, we consider a general scenario where measured and latent variables collectively form a partially observed causally sufficient linear system and latent variables may be anywhere in the causal structure. We theoretically show that with the aid of high-order statistics, the causal graph is (almost) fully identifiable if, roughly speaking, each latent set has a sufficient number of pure children, which can be either latent or measured. Naturally, LiNGAM, a model without latent variables, is encompassed as a special case. Based on the identification theorem, we develop a principled algorithm to identify the causal graph by testing for statistical independence involving only measured variables in specific manners. Experimental results show that our method effectively recovers the causal structure, even when latent variables are influenced by measured variables.

Conventional causal discovery methods rely on centralized data, which is inconsistent with the decentralized nature of data in many real-world situations. This discrepancy has motivated the development of federated causal discovery (FCD) approaches. However, existing FCD methods may be limited by their potentially restrictive assumptions of identifiable functional causal models or homogeneous data distributions, narrowing their applicability in diverse scenarios. In this paper, we propose a novel FCD method attempting to accommodate arbitrary causal models and heterogeneous data. We first utilize a surrogate variable corresponding to the client index to account for the data heterogeneity across different clients. We then develop a federated conditional independence test (FCIT) for causal skeleton discovery and establish a federated independent change principle (FICP) to determine causal directions. These approaches involve constructing summary statistics as a proxy of the raw data to protect data privacy. Owing to the nonparametric properties, FCIT and FICP make no assumption about particular functional forms, thereby facilitating the handling of arbitrary causal models. We conduct extensive experiments on synthetic and real datasets to show the efficacy of our method. The code is available at https://github.com/lokali/FedCDH.git.

Identifying the underlying time-delayed latent causal processes in sequential data is vital for grasping temporal dynamics and making downstream reasoning. While some recent methods can robustly identify these latent causal variables, they rely on strict assumptions about the invertible generation process from latent variables to observed data. However, these assumptions are often hard to satisfy in real-world applications containing information loss. For instance, the visual perception process translates a 3D space into 2D images, or the phenomenon of persistence of vision incorporates historical data into current perceptions. To address this challenge, we establish an identifiability theory that allows for the recovery of independent latent components even when they come from a nonlinear and non-invertible mix. Using this theory as a foundation, we propose a principled approach, CaRiNG, to learn the Causal Representation of Non-invertible Generative temporal data with identifiability guarantees. Specifically, we utilize temporal context to recover lost latent information and apply the conditions in our theory to guide the training process. Through experiments conducted on synthetic datasets, we validate that our CaRiNG method reliably identifies the causal process, even when the generation process is non-invertible. Moreover, we demonstrate that our approach considerably improves temporal understanding and reasoning in practical applications.

Few-shot action recognition aims at quickly adapting a pre-trained model to the novel data with a distribution shift using only a limited number of samples. Key challenges include how to identify and leverage the transferable knowledge learned by the pre-trained model. We therefore propose CDTD, or Causal Domain-Invariant Temporal Dynamics for knowledge transfer. To identify the temporally invariant and variant representations, we employ the causal representation learning methods for unsupervised pertaining, and then tune the classifier with supervisions in next stage. Specifically, we assume the domain information can be well estimated and the pre-trained temporal dynamic generation and transition models can be well transferred. During adaptation, we fix the transferable temporal dynamics and update the image encoder and domain estimator. The efficacy of our approach is revealed by the superior accuracy of CDTD over leading alternatives across standard few-shot action recognition datasets.

Learning concepts from natural high-dimensional data (e.g., images) holds potential in building human-aligned and interpretable machine learning models. Despite its encouraging prospect, formalization and theoretical insights into this crucial task are still lacking. In this work, we formalize concepts as discrete latent causal variables that are related via a hierarchical causal model that encodes different abstraction levels of concepts embedded in high-dimensional data (e.g., a dog breed and its eye shapes in natural images). We formulate conditions to facilitate the identification of the proposed causal model, which reveals when learning such concepts from unsupervised data is possible. Our conditions permit complex causal hierarchical structures beyond latent trees and multi-level directed acyclic graphs in prior work and can handle high-dimensional, continuous observed variables, which is well-suited for unstructured data modalities such as images. We substantiate our theoretical claims with synthetic data experiments. Further, we discuss our theory's implications for understanding the underlying mechanisms of latent diffusion models and provide corresponding empirical evidence for our theoretical insights.

Canonical work handling distribution shifts typically necessitates an entire target distribution that lands inside the training distribution. However, practical scenarios often involve only a handful target samples, potentially lying outside the training support, which requires the capability of extrapolation. In this work, we aim to provide a theoretical understanding of when extrapolation is possible and offer principled methods to achieve it without requiring an on-support target distribution. To this end, we formulate the extrapolation problem with a latent-variable model that embodies the minimal change principle in causal mechanisms. Under this formulation, we cast the extrapolation problem into a latent-variable identification problem. We provide realistic conditions on shift properties and the estimation objectives that lead to identification even when only one off-support target sample is available, tackling the most challenging scenarios. Our theory reveals the intricate interplay between the underlying manifold's smoothness and the shift properties. We showcase how our theoretical results inform the design of practical adaptation algorithms. Through experiments on both synthetic and real-world data, we validate our theoretical findings and their practical implications.

Many causal discovery methods typically rely on the assumption of independent noise, yet real-life situations often involve deterministic relationships. In these cases, observed variables are represented as deterministic functions of their parental variables without noise. When determinism is present, constraint-based methods encounter challenges due to the violation of the faithfulness assumption. In this paper, we find, supported by both theoretical analysis and empirical evidence, that score-based methods with exact search can naturally address the issues of deterministic relations under rather mild assumptions. Nonetheless, exact score-based methods can be computationally expensive. To enhance the efficiency and scalability, we develop a novel framework for causal discovery that can detect and handle deterministic relations, called Determinism-aware Greedy Equivalent Search (DGES). DGES comprises three phases: (1) identify minimal deterministic clusters (i.e., a minimal set of variables with deterministic relationships), (2) run modified Greedy Equivalent Search (GES) to obtain an initial graph, and (3) perform exact search exclusively on the deterministic cluster and its neighbors. The proposed DGES accommodates both linear and nonlinear causal relationships, as well as both continuous and discrete data types. Furthermore, we investigate the identifiability conditions of DGES. We conducted extensive experiments on both simulated and real-world datasets to show the efficacy of our proposed method.

Causal Temporal Representation Learning (Ctrl) methods aim to identify the temporal causal dynamics of complex nonstationary temporal sequences. Despite the success of existing Ctrl methods, they require either directly observing the domain variables or assuming a Markov prior on them. Such requirements limit the application of these methods in real-world scenarios when we do not have such prior knowledge of the domain variables. To address this problem, this work adopts a sparse transition assumption, aligned with intuitive human understanding, and presents identifiability results from a theoretical perspective. In particular, we explore under what conditions on the significance of the variability of the transitions we can build a model to identify the distribution shifts. Based on the theoretical result, we introduce a novel framework, *Causal Temporal Representation Learning with Nonstationary Sparse Transition* (CtrlNS), designed to leverage the constraints on transition sparsity and conditional independence to reliably identify both distribution shifts and latent factors. Our experimental evaluations on synthetic and real-world datasets demonstrate significant improvements over existing baselines, highlighting the effectiveness of our approach.

Recent works have introduced GNN-to-MLP knowledge distillation (KD) frameworks to combine both GNN's superior performance and MLP's fast inference speed. However, existing KD frameworks are primarily designed for node classification within single graphs, leaving their applicability to graph classification largely unexplored. Two main challenges arise when extending KD for node classification to graph classification: (1) The inherent sparsity of learning signals due to soft labels being generated at the graph level; (2) The limited expressiveness of student MLPs, especially in datasets with limited input feature spaces. To overcome these challenges, we introduce MuGSI, a novel KD framework that employs Multi-granularity Structural Information for graph classification. Specifically, we propose multi-granularity distillation loss in MuGSI to tackle the first challenge. This loss function is composed of three distinct components: graph-level distillation, subgraph-level distillation, and node-level distillation. Each component targets a specific granularity of the graph structure, ensuring a comprehensive transfer of structural knowledge from the teacher model to the student model. To tackle the second challenge, MuGSI proposes to incorporate a node feature augmentation component, thereby enhancing the expressiveness of the student MLPs and making them more capable learners. We perform extensive experiments across a variety of datasets and different teacher/student model architectures. The experiment results demonstrate the effectiveness, efficiency, and robustness of MuGSI. Codes are publicly available at: https://github.com/tianyao-aka/MuGSI.

Source-free object detection (SFOD) aims to transfer a detector pre-trained on a label-rich source domain to an unlabeled target domain without seeing source data. While most existing SFOD methods generate pseudo labels via a source-pretrained model to guide training, these pseudo labels usually contain high noises due to heavy domain discrepancy. In order to obtain better pseudo supervisions, we divide the target domain into source-similar and source-dissimilar parts and align them in the feature space by adversarial learning.Specifically, we design a detection variance-based criterion to divide the target domain. This criterion is motivated by a finding that larger detection variances denote higher recall and larger similarity to the source domain. Then we incorporate an adversarial module into a mean teacher framework to drive the feature spaces of these two subsets indistinguishable. Extensive experiments on multiple cross-domain object detection datasets demonstrate that our proposed method consistently outperforms the compared SFOD methods. Our implementation is available at https://github.com/ChuQiaosong.

The indeterminate nature of human motion requires trajectory prediction systems to use a probabilistic model to formulate the multi-modality phenomenon and infer a finite set of future trajectories. However, the inference processes of most existing methods rely on Monte Carlo random sampling, which is insufficient to cover the realistic paths with finite samples, due to the long tail effect of the predicted distribution. To promote the sampling process of stochastic prediction, we propose a novel method, called BOsampler, to adaptively mine potential paths with Bayesian optimization in an unsupervised manner, as a sequential design strategy in which new prediction is dependent on the previously drawn samples. Specifically, we model the trajectory sampling as a Gaussian process and construct an acquisition function to measure the potential sampling value. This acquisition function applies the original distribution as prior and encourages exploring paths in the long-tail region. This sampling method can be integrated with existing stochastic predictive models without retraining. Experimental results on various baseline methods demonstrate the effectiveness of our method. The source code is released in this link.

Masked autoencoder (MAE), a simple and effective self-supervised learning framework based on the reconstruction of masked image regions, has recently achieved prominent success in a variety of vision tasks. Despite the emergence of intriguing empirical observations on MAE, a theoretically principled understanding is still lacking. In this work, we formally characterize and justify existing empirical in-sights and provide theoretical guarantees of MAE. We formulate the underlying data-generating process as a hierarchical latent variable model, and show that under reasonable assumptions, MAE provably identifies a set of latent variables in the hierarchical model, explaining why MAE can extract high-level information from pixels. Further, we show how key hyperparameters in MAE (the masking ratio and the patch size) determine which true latent variables to be recovered, therefore influencing the level of semantic information in the representation. Specifically, extremely large or small masking ratios inevitably lead to low-level representations. Our theory offers coherent explanations of existing empirical observations and provides insights for potential empirical improvements and fundamental limitations of the masked-reconstruction paradigm. We conduct extensive experiments to validate our theoretical insights.

Although existing semi-supervised learning models achieve remarkable success in learning with unannotated in-distribution data, they mostly fail to learn on unlabeled data sampled from novel semantic classes due to their closed-set assumption. In this work, we target a pragmatic but under-explored Generalized Novel Category Discovery (GNCD) setting. The GNCD setting aims to categorize unlabeled training data coming from known and novel classes by leveraging the information of partially labeled known classes. We propose a two-stage Contrastive Affinity Learning method with auxiliary visual Prompts, dubbed PromptCAL, to address this challenging problem. Our approach discovers reliable pairwise sample affinities to learn better semantic clustering of both known and novel classes for the class token and visual prompts. First, we propose a discriminative prompt regularization loss to reinforce semantic discriminativeness of prompt-adapted pre-trained vision transformer for refined affinity relationships. Besides, we propose contrastive affinity learning to calibrate semantic representations based on our iterative semi-supervised affinity graph generation method for semantically-enhanced supervision. Extensive experimental evaluation demonstrates that our PromptCAL method is more effective in discovering novel classes even with limited annotations and surpasses the current state-of-the-art on generic and fine-grained benchmarks (e.g., with nearly 11% gain on CUB-200, and 9% on ImageNet-100) on overall accuracy. Our code is available at https://github.com/sheng-eatamath/PromptCAL.

Video-language pre-trained models have shown remarkable success in guiding video question-answering (VideoQA) tasks. However, due to the length of video sequences, training large-scale video-based models incurs considerably higher costs than training image-based ones. This motivates us to leverage the knowledge from image-based pretraining, despite the obvious gaps between image and video domains. To bridge these gaps, in this paper, we propose Tem-adapter, which enables the learning of temporal dynamics and complex semantics by a visual Temporal Aligner and a textual Semantic Aligner. Unlike conventional pretrained knowledge adaptation methods that only concentrate on the downstream task objective, the Temporal Aligner introduces an extra language-guided autoregressive task aimed at facilitating the learning of temporal dependencies, with the objective of predicting future states based on historical clues and language guidance that describes event progression. Besides, to reduce the semantic gap and adapt the textual representation for better event description, we introduce a Semantic Aligner that first designs a template to fuse question and answer pairs as event descriptions and then learns a Transformer decoder with the whole video sequence as guidance for refinement. We evaluate Tem-adapter and different pre-train transferring methods on two VideoQA benchmarks, and the significant performance improvement demonstrates the effectiveness of our method.1

Graph neural networks aim to learn representations for graph-structured data and show impressive performance in node classification. Recently, many methods have studied the representations of GNNs from the perspective of optimization goals and spectral graph theory. However, the feature space that dominates representation learning has not been systematically studied in graph neural networks. In this paper, we propose to fill this gap by analyzing the feature space of both spatial and spectral models. We decompose graph neural networks into determined feature spaces and trainable weights, providing the convenience of studying the feature space explicitly using matrix space analysis. In particular, we find theoretically that the feature space tends to be linearly correlated due to repeated aggregations. In this case, the feature space is bounded by the poor representation of shared weights or the limited dimensionality of node attributes in existing models, leading to poor performance. Motivated by these findings, we propose 1) feature subspaces flattening and 2) structural principal components to expand the feature space. Extensive experiments verify the effectiveness of our proposed more comprehensive feature space, with comparable inference time to the baseline, and demonstrate its efficient convergence capability.

Multi-source domain adaptation (MSDA) methods aim to transfer knowledge from multiple labeled source domains to an unlabeled target domain. Although current methods achieve target joint distribution identifiability by enforcing minimal changes across domains, they often necessitate stringent conditions, such as an adequate number of domains, monotonic transformation of latent variables, and invariant label distributions. These requirements are challenging to satisfy in real-world applications. To mitigate the need for these strict assumptions, we propose a subspace identification theory that guarantees the disentanglement of domain-invariant and domain-specific variables under less restrictive constraints regarding domain numbers and transformation properties and thereby facilitating domain adaptation by minimizing the impact of domain shifts on invariant variables. Based on this theory, we develop a Subspace Identification Guarantee (SIG) model that leverages variational inference. Furthermore, the SIG model incorporates class-aware conditional alignment to accommodate target shifts where label distributions change with the domain. Experimental results demonstrate that our SIG model outperforms existing MSDA techniques on various benchmark datasets, highlighting its effectiveness in real-world applications.

In unsupervised causal representation learning for sequential data with time-delayed latent causal influences, strong identifiability results for the disentanglement of causally-related latent variables have been established in stationary settings by leveraging temporal structure. However, in nonstationary setting, existing work only partially addressed the problem by either utilizing observed auxiliary variables (e.g., class labels and/or domain indexes) as side information or assuming simplified latent causal dynamics. Both constrain the method to a limited range of scenarios. In this study, we further explored the Markov Assumption under time-delayed causally related process in nonstationary setting and showed that under mild conditions, the independent latent components can be recovered from their nonlinear mixture up to a permutation and a component-wise transformation, without the observation of auxiliary variables. We then introduce NCTRL, a principled estimation framework, to reconstruct time-delayed latent causal variables and identify their relations from measured sequential data only. Empirical evaluations demonstrated the reliable identification of time-delayed latent causal influences, with our methodology substantially outperforming existing baselines that fail to exploit the nonstationarity adequately and then, consequently, cannot distinguish distribution shifts.

Human behavior has the nature of indeterminacy, which requires the pedestrian trajectory prediction system to model the multi-modality of future motion states. Unlike existing stochastic trajectory prediction methods which usually use a latent variable to represent multi-modality, we explicitly simulate the process of human motion variation from indeterminate to determinate. In this paper, we present a new framework to formulate the trajectory prediction task as a reverse process of motion indeterminacy diffusion (MID), in which we progressively discard indeterminacy from all the walkable areas until reaching the desired trajectory. This process is learned with a parameterized Markov chain conditioned by the observed trajectories. We can adjust the length of the chain to control the degree of indeterminacy and balance the diversity and determinacy of the predictions. Specifically, we encode the history behavior information and the social interactions as a state embedding and devise a Transformer-based diffusion model to capture the temporal dependencies of trajectories. Extensive experiments on the human trajectory prediction benchmarks including the Stanford Drone and ETH/UCY datasets demonstrate the superiority of our method. Code is available at https://github.com/gutianpei/MID.

Recent progress has shown that large-scale pre-training using contrastive image-text pairs can be a promising alternative for high-quality visual representation learning from natural language supervision. Benefiting from a broader source of supervision, this new paradigm exhibits impressive transferability to downstream classification tasks and datasets. However, the problem of transferring the knowledge learned from image-text pairs to more complex dense prediction tasks has barely been visited. In this work, we present a new framework for dense prediction by implicitly and explicitly leveraging the pre-trained knowledge from CLIP. Specifically, we convert the original image-text matching problem in CLIP to a pixel-text matching problem and use the pixel-text score maps to guide the learning of dense prediction models. By further using the contextual information from the image to prompt the language model, we are able to facilitate our model to better exploit the pretrained knowledge. Our method is model-agnostic, which can be applied to arbitrary dense prediction systems and various pre-trained visual backbones including both CLIP models and ImageNet pre-trained models. Extensive experiments demonstrate the superior performance of our methods on semantic segmentation, object detection, and instance segmentation tasks. Code is available at https://github.com/raoyongming/DenseCLIP.

Most existing action quality assessment methods rely on the deep features of an entire video to predict the score, which is less reliable due to the non-transparent inference process and poor interpretability. We argue that understanding both high-level semantics and internal temporal structures of actions in competitive sports videos is the key to making predictions accurate and interpretable. Towards this goal, we construct a new fine-grained dataset, called FineDiving, developed on diverse diving events with detailed annotations on action procedures. We also propose a procedure-aware approach for action quality assessment, learned by a new Temporal Segmentation Attention module. Specifically, we propose to parse pairwise query and exemplar action instances into consecutive steps with diverse semantic and temporal correspondences. The procedure-aware cross-attention is proposed to learn embeddings between query and exemplar steps to discover their semantic, spatial, and temporal correspondences, and further serve for fine-grained contrastive regression to derive a reliable scoring mechanism. Extensive experiments demonstrate that our approach achieves substantial improvements over the state-of-the-art methods with better interpretability. The dataset and code are available at https://github.com/xujinglin/FineDiving.

Unsupervised domain adaptation is critical to many real-world applications where label information is unavailable in the target domain. In general, without further assumptions, the joint distribution of the features and the label is not identifiable in the target domain. To address this issue, we rely on a property of minimal changes of causal mechanisms across domains to minimize unnecessary influences of domain shift. To encode this property, we first formulate the data generating process using a latent variable model with two partitioned latent subspaces: invariant components whose distributions stay the same across domains, and sparse changing components that vary across domains. We further constrain the domain shift to have a restrictive influence on the changing components. Under mild conditions, we show that the latent variables are partially identifiable, from which it follows that the joint distribution of data and labels in the target domain is also identifiable. Given the theoretical insights, we propose a practical domain adaptation framework, called iMSDA. Extensive experimental results reveal that iMSDA outperforms state-of-the-art domain adaptation algorithms on benchmark datasets, demonstrating the effectiveness of our framework.

Recently in the field of unsupervised representation learning, strong identifiability results for disentanglement of causally-related latent variables have been established by exploiting certain side information, such as class labels, in addition to independence. However, most existing work is constrained by functional form assumptions such as independent sources or further with linear transitions, and distribution assumptions such as stationary, exponential family distribution. It is unknown whether the underlying latent variables and their causal relations are identifiable if they have arbitrary, nonparametric causal influences in between. In this work, we establish the identifiability theories of nonparametric latent causal processes from their nonlinear mixtures under fixed temporal causal influences and analyze how distribution changes can further benefit the disentanglement. We propose TDRL, a principled framework to recover time-delayed latent causal variables and identify their relations from measured sequential data under stationary environments and under different distribution shifts. Specifically, the framework can factorize unknown distribution shifts into transition distribution changes under fixed and time-varying latent causal relations, and under global changes in observation. Through experiments, we show that time-delayed latent causal influences are reliably identified and that our approach considerably outperforms existing baselines that do not correctly exploit this modular representation of changes.

How do humans localize unintentional action like " A boy falls down while playing skateboard "? Cognitive science shows that an 18-month-old baby understands the intention by observing the actions and comparing the feedback. Motivated by this evidence, we propose a causal inference approach that constructs a video pool containing intentional knowledge, conducts the counterfactual intervention to observe intentional action, and compares the unintentional action with intentional action to achieve localization. Specifically, we first build a video pool, where each video contains the same action content as an original unintentional action video. Then we conduct the counterfactual intervention to generate counterfactual examples. We further maximize the difference between the predictions of factual unintentional action and counterfactual intentional action to train the model. By disentangling the effects of different clues on the model prediction, we encourage the model to highlight the intention clue and alleviate the negative effect brought by the training bias of the action content clue. We evaluate our approach on a public unintentional action dataset and achieve consistent improvements on both unintentional action recognition and localization tasks.

Humans have the inherent advantage of understanding action intention, while it is an enormous challenge to train the machine to localize unintentional action in videos due to the lack of reliable annotations for stable training. The annotations of unintentional action are unreliable since different annotators are affected by their subjective appraisals and intrinsic ambiguity, which brings heavy difficulties for the training. To address this issue, we propose a probabilistic framework for unintentional action localization by modeling the uncertainty of annotations. Our framework consists of two main components, including Temporal Label Aggregation (TLA) and Dense Probabilistic Localization (DPL). We first formulate each annotated failure moment as a temporal label distribution. Then we propose a TLA component to aggregate temporal label distributions of different failure moments in an online manner and generate dense probabilistic supervision. Based on TLA, We further develop a DPL component to jointly train three heads (i.e., probabilistic dense classification, probabilistic temporal detection, and probabilistic regression) with different supervision granularities and make them highly collaborative. We evaluate our approach on the largest unintentional action dataset OOPS and demonstrate that our approach can achieve significant improvement over the baseline and state-of-the-art methods.

Forecasting human trajectories in complex dynamic environments plays a critical role in autonomous vehicles and intelligent robots. Most existing methods learn to predict future trajectories by behavior clues from history trajectories and interaction clues from environments. However, the inherent bias between training and deployment environments is ignored. Hence, we propose a counterfactual analysis method for human trajectory prediction to investigate the causality between the predicted trajectories and input clues and alleviate the negative effects brought by the environment bias. We first build a causal graph for trajectory forecasting with history trajectory, future trajectory, and the environment interactions. Then, we cut off the inference from environment to trajectory by constructing the counterfactual intervention on the trajectory itself. Finally, we compare the factual and counterfactual trajectory clues to alleviate the effects of environment bias and highlight the trajectory clues. Our counterfactual analysis is a plug-and-play module that can be applied to any baseline prediction methods including RNN- and CNN-based ones. We show that our method achieves consistent improvement for different baselines and obtains the state-of-the-art results on public pedestrian trajectory forecasting benchmarks.1

Trajectory prediction is confronted with the dilemma to capture the multi-modal nature of future dynamics with both diversity and accuracy. In this paper, we present a distribution discrimination (DisDis) method to predict personalized motion patterns by distinguishing the potential distributions. Motivated by that the motion pattern of each person is personalized due to his/her habit, our DisDis learns the latent distribution to represent different motion patterns and optimize it by the contrastive discrimination. This distribution discrimination encourages latent distributions to be more discriminative. Our method can be integrated with existing multi-modal stochastic predictive models as a plug-and-play module to learn the more discriminative latent distribution. To evaluate the latent distribution, we further propose a new metric, probability cumulative minimum distance (PCMD) curve, which cumulatively calculates the minimum distance on the sorted probabilities. Experimental results on the ETH and UCY datasets show the effectiveness of our method.1

Attention mechanism has demonstrated great potential in fine-grained visual recognition tasks. In this paper, we present a counterfactual attention learning method to learn more effective attention based on causal inference. Unlike most existing methods that learn visual attention based on conventional likelihood, we propose to learn the attention with counterfactual causality, which provides a tool to measure the attention quality and a powerful supervisory signal to guide the learning process. Specifically, we analyze the effect of the learned visual attention on network prediction through counterfactual intervention and maximize the effect to encourage the network to learn more useful attention for fine-grained image recognition. Empirically, we evaluate our method on a wide range of fine-grained recognition tasks where attention plays a crucial role, including fine-grained image categorization, person re-identification, and vehicle re-identification. The consistent improvement on all benchmarks demonstrates the effectiveness of our method.1

Humans can easily understand whether a person’s action is intentional or not. However, it is very challenging to teach a machine to recognize this due to the lack of referable comparisons and reliable annotations. Given a video with unintentional action, the annotations are usually unreliable due to the intrinsic ambiguity from multiple annotators and the subjective appraisals. To address this problem, we propose a new framework which online aggregates multiple probabilistic labels for unintentional action localization. Specifically, we first model the uncertainty of annotations with a temporal probability distribution, and then develop a label attention model to aggregate the reliable annotations in an online manner. We evaluate our method on the public OOPS dataset where each video contains multiple annotations of unintentional action and our experimental results show that mining reliable supervision information from multiple unreliable annotations achieves significant improvements over the baseline methods.

In this paper, we propose an attention pyramid method for person re-identification. Unlike conventional attention-based methods which only learn a global attention map, our attention pyramid exploits the attention regions in a multi-scale manner because human attention varies with different scales. Our attention pyramid imitates the process of human visual perception which tends to notice the foreground person over the cluttered background, and further focus on the specific color of the shirt with close observation. Specifically, we describe our attention pyramid by a "split-attend-merge-stack" principle. We first split the features into multiple local parts and learn the corresponding attentions. Then, we merge local attentions and stack these merged attentions with the residual connection as an attention pyramid. The proposed attention pyramid is a lightweight plug-and-play module that can be applied to off-the-shelf models. We implement our attention pyramid method in two different attention mechanisms including: channel-wise attention and spatial attention. We evaluate our method on four large-scale person re-identification benchmarks including Market-1501, DukeMTMC, CUHK03, and MSMT17. Experimental results demonstrate the superiority of our method, which outperforms the state-of-the-art methods by a large margin with limited computationa cost. Code is available at https://github.com/CHENGY12/APNet.

In this paper, we propose to learn recurrent 3D attention (A3D) for video-based person re-identification. Attention model plays a key role in both spatial and temporal domains for video representation. Most existing methods apply spatial attention model to extract feature from a single image and aggregate image features with attentive temporal pooling or RNN. However, the inherent consistencies and correlations between spatial and temporal clues are not leveraged. Our A3D method aims to utilize the joint constraints of temporal and spatial attentions to enhance the robustness of attention model. Towards this goal, we treat the pedestrian video as a unified 3D bin where the temporal domain is denoted as an additional dimension. Then we develop an attention agent to iteratively select the locations of the salient spatial-temporal parts in the 3D bin. In addition, we formulate our sequential 3D attention learning as a Markov Decision Process and train the representation network and attention detector with the policy gradient method in an end-to-end manner. We evaluate the proposed method on three challenging datasets including iLIDS-VID, PRID-2011 and the large-scale MARS dataset, and consistently improve the performance in comparison with the state-of-the-art methods.

In this paper, we propose a self-critical attention learning method for person re-identification. Unlike most existing methods which train the attention mechanism in a weakly-supervised manner and ignore the attention confidence level, we learn the attention with a critic which measures the attention quality and provides a powerful supervisory signal to guide the learning process. Moreover, the critic model facilitates the interpretation of the effectiveness of the attention mechanism during the learning process, by estimating the quality of the attention maps. Specifically, we jointly train our attention agent and critic in a reinforcement learning manner, where the agent produces the visual attention while the critic analyzes the gain from the attention and guides the agent to maximize this gain. We design spatial- and channel-wise attention models with our critic module and evaluate them on three popular benchmarks including Market-1501, DukeMTMC-ReID, and CUHK03. The experimental results demonstrate the superiority of our method, which outperforms the state-of-the-art methods by a large margin of 5.9%/2.1%, 6.3%/3.0%, and 10.5%/9.5% on mAP/Rank-1, respectively.

In this paper, we present a deep meta metric learning (DMML) approach for visual recognition. Unlike most existing deep metric learning methods formulating the learning process by an overall objective, our DMML formulates the metric learning in a meta way, and proves that softmax and triplet loss are consistent in the meta space. Specifically, we sample some subsets from the original training set and learn metrics across different subsets. In each sampled sub-task, we split the training data into a support set as well as a query set, and learn the set-based distance, instead of sample-based one, to verify the query cell from multiple support cells. In addition, we introduce hard sample mining for set-based distance to encourage the intra-class compactness. Experimental results on three visual recognition applications including person re-identification, vehicle re-identification and face verification show that the proposed DMML method outperforms most existing approaches.

In this paper, we present a spatial-temporal attention-aware learning (STAL) method for video-based person re-identification. Most existing person re-identification methods aggregate image features identically to represent persons, which are extracted from the same receptive field across video frames. However, the image quality may be varying for different spatial regions and changing over time, which shall contribute to person representation and matching adaptively. Our STAL method aims to attend to the salient parts of persons in videos jointly in both spatial and temporal domains. To achieve this, we slice the video into multiple spatial-temporal units which preserve the body structure of a person and develop a joint spatial-temporal attention model to learn the quality scores of these units. We evaluate the proposed method on three challenging datasets including iLIDS-VID, PRID-2011, and the large-scale MARS dataset, and consistently improve the rank-1 accuracy by a large margin of 5.7%, 0.9%, and 6.6% respectively, in comparison with the state-of-the-art methods.

This paper presents a localized multi-kernel discriminative canonical correlation analysis (LMKDCCA) approach for video-based person re-identification, which aims to match persons from pedestrian videos captured by non-overlapping cameras. Unlike conventional methods, our approach models each pedestrian video as a point on the Riemannian manifold and learns similarity over these points under the multiple kernel learning framework. For each given person video, we first represent it as a symmetric positive definite (SPD) matrix which lies on a Riemannian manifold and compute the similarity of multiple SPDs. Then, we develop an LMKDCCA algorithm to learn a nonlinear distance metric which effectively combines these SPDs to exploit complementary information for similarity measure. Experimental results on the iLIDS-VID and PRID 2011 datasets show that our approach achieves the state-of-the-arts.