Despite the great performance of deep learning models in many areas, they still make mistakes and underperform on certain subsets of data, i.e. error slices. Given a trained model, it is important to identify its semantically coherent error slices that are easy to interpret, which is referred to as the error slice discovery problem. However, there is no proper metric of slice coherence without relying on extra information like predefined slice labels. Current evaluation of slice coherence requires access to predefined slices formulated by metadata like attributes or subclasses. Its validity heavily relies on the quality and abundance of metadata, where some possible patterns could be ignored. Besides, current algorithms cannot directly incorporate the constraint of coherence into their optimization objective due to absence of an explicit coherence metric, which could potentially hinder their effectiveness. In this paper, we propose manifold compactness, a coherence metric without reliance on extra information by incorporating the data geometry property into its design, and experiments on typical datasets empirically validate the rationality of the metric. Then we develop Manifold Compactness based error Slice Discovery (MCSD), a novel algorithm that directly treats risk and coherence as the optimization objective, and is flexible to be applied to models of various tasks. Extensive experiments on the benchmark and case studies on other typical datasets demonstrate the superiority of MCSD.

Although neural networks achieve promising performance in many tasks, they may still fail when encountering some examples and bring about risks to applications. To discover risky samples, previous literature attempts to search for patterns of risky samples within existing datasets or inject perturbation into them. Yet in this way the diversity of risky samples is limited by the coverage of existing datasets. To overcome this limitation, recent works adopt diffusion models to produce new risky samples beyond the coverage of existing datasets. However, these methods struggle in the conformity between generated samples and expected categories, which could introduce label noise and severely limit their effectiveness in applications. To address this issue, we propose RiskyDiff that incorporates the embeddings of both texts and images as implicit constraints of category conformity. We also design a conformity score to further explicitly strengthen the category conformity, as well as introduce the mechanisms of embedding screening and risky gradient guidance to boost the risk of generated samples. Extensive experiments reveal that RiskyDiff greatly outperforms existing methods in terms of the degree of risk, generation quality, and conformity with conditioned categories. We also empirically show the generalization ability of the models can be enhanced by augmenting training data with generated samples of high conformity.

Large-scale color datasets exhibit significant sparsity in name-color correspondences, substantially impeding the effectiveness of conventional methodologies. We propose a contrastive learning-based framework for color name generation and recommendation that addresses sparsity through negative sampling, supporting two core tasks: color-to-name recommendation and name-to-color generation. Our framework employs a multi-task contrastive learning architecture comprising three key components: (1) a pre-trained Transformer-based name encoder, (2) an RGB encoder, and (3) an RGB generator. The framework utilizes negative sampling to construct positive-negative pairs, contrasting RGB encoder outputs with positive and negative name embeddings. We adopt a multi-objective optimization strategy incorporating binary cross-entropy loss for neural collaborative filtering, and mean squared error loss for name-to-RGB mapping. Experimental results demonstrate substantial improvements over baseline methods, achieving 71.26% Top-10 accuracy in color-to-name recommendation and reducing CIELAB distance error to 26.61 in name-to-color generation.

The out-of-distribution (OOD) generalization of graph neural networks poses significant challenges in Web applications, where data resides in complex heterogeneous information networks. Such networks exhibit not only structural heterogeneity but also distribution shifts arising from evolving user behaviors and data collection biases. Conventional GNNs often struggle to identify stable patterns in such heterogeneous graphs, particularly in the absence of explicit environment labels. The core issue is that latent environments can create spurious correlations between node features, local topology, and labels. Models may then rely on these environment-specific shortcuts for predictions, failing to learn the invariant mechanisms that generalize under distribution shifts. To address these limitations, we propose InvHG (Invariant Learning on Heterogeneous Graphs via Subgraph Environment Inference), a causality-inspired framework that infers latent environments at the subgraph level, disentangles type-specific confounding effects, and leverages regularized expert fusion to learn invariant representations. Extensive experiments on heterogeneous graph OOD benchmarks demonstrate that InvHG consistently outperforms state-of-the-art methods, offering a robust solution for complex Web graph learning. The source code is available at https://github.com/mok630/InvHG.

Current object detectors often suffer significant performance degradation in real-world applications when encountering distributional shifts. Consequently, the out-of-distribution (OOD) generalization capability of object detectors has garnered increasing attention from researchers. Despite this growing interest, there remains a lack of a large-scale, comprehensive dataset and evaluation benchmark with fine-grained annotations tailored to assess the OOD generalization on more intricate tasks like object detection and grounding. To address this gap, we introduce COUNTS, a large-scale OOD dataset with object-level annotations. COUNTS encompasses 14 natural distributional shifts, over 222K samples, and more than 1,196K labeled bounding boxes. Leveraging COUNTS, we introduce two novel benchmarks: O(OD)1and OODG. O(OD)1is designed to comprehensively evaluate the OOD generalization capabilities of object detectors by utilizing controlled distribution shifts between training and testing data. OODG, on the other hand, aims to assess the OOD generalization of grounding abilities in multimodal large language models (MLLMs). Our findings reveal that, while large models and extensive pre-training data substantially enhance performance in in-distribution (IID) scenarios, significant limitations and opportunities for improvement persist in OOD contexts for both object detectors and MLLMs. In visual grounding tasks, even the advanced GPT-4o and Gemini-1.5 only achieve 56.7% and 28.0% accuracy, respectively. We hope COUNTS facilitates advancements in the development and assessment of robust object detectors and MLLMs capable of maintaining high performance under distributional shifts.

Recently, there has been gradually more attention paid to Out-of-Distribution (OOD) performance prediction, whose goal is to predict the performance of trained models on unlabeled OOD test datasets, so that we could better leverage and deploy off-the-shelf trained models in risk-sensitive scenarios. Although progress has been made in this area, evaluation protocols in previous literature are inconsistent, and most works cover only a limited number of real-world OOD datasets and types of distribution shifts. To provide convenient and fair comparisons for various algorithms, we propose Out-of-Distribution Performance Prediction Benchmark (ODP-Bench), a comprehensive benchmark that includes most commonly used OOD datasets and existing practical performance prediction algorithms. We provide our trained models as a testbench for future researchers, thus guaranteeing the consistency of comparison and avoiding the burden of repeating the model training process. Furthermore, we also conduct in-depth experimental analyses to better understand their capability boundary.

To provide actionable insights and interpretations for individuals affected by algorithmic decisions, algorithmic recourse-demonstrating how outcomes change with modifications to input features-is introduced to facilitate outcome adjustment.However, existing studies often focus on different notions of feasibility and impose complex optimization constraints, relying on strong assumptions and expert knowledge that may be impractical or not widely applicable.In this paper, we propose leveraging adherence to the perturbation data manifold to model typical feasibility challenges, providing both a theoretical clarification and a practical framework.We design optimization constraints based on this model and introduce our method, the Perturbation Data Manifold Constraint (PDMC), to ensure the feasibility of generated algorithmic recourses.Through extensive experiments on both simulated and real clinical data, we validate the rationale and effectiveness of PDMC.

Data-driven methods offer efficient and robust solutions for analyzing complex dynamical systems but rely on the assumption of I.I.D. data, driving the development of generalization techniques for handling environmental differences. These techniques, however, are limited by their dependence on environment labels, which are often unavailable during training due to data acquisition challenges, privacy concerns, and environmental variability, particularly in large public datasets and privacy-sensitive domains. In response, we propose DynaInfer, a novel method that infers environment specifications by analyzing prediction errors from fixed neural networks within each training round, enabling environment assignments directly from data. We prove our algorithm effectively solves the alternating optimization problem in unlabeled scenarios and validate it through extensive experiments across diverse dynamical systems. Results show that DynaInfer outperforms existing environment assignment techniques, converges rapidly to true labels, and even achieves superior performance when environment labels are available.

Heterogeneous Information Networks (HINs) encapsulate diverse entity and relation types, with meta-paths providing essential meta-level semantics for knowledge reasoning, although their utility is constrained by discovery challenges. While Large Language Models (LLMs) offer new prospects for meta-path discovery due to their extensive knowledge encoding and efficiency, their adaptation faces challenges such as corpora bias, lexical discrepancies, and hallucination . This paper pioneers the mitigation of these challenges by presenting EvoPath, an innovative framework that leverages LLMs to efficiently identify high-quality meta-paths. EvoPath is carefully designed, with each component aimed at addressing issues that could lead to potential knowledge conflicts. With a minimal subset of HIN facts, EvoPath iteratively generates and evolves meta-paths by dynamically replaying meta-paths in the buffer with prioritization based on their scores. Comprehensive experiments on three large, complex HINs with hundreds of relations demonstrate that our framework, EvoPath, enables LLMs to generate high-quality meta-paths through effective prompting, confirming its superior performance in HIN reasoning tasks. Further ablation studies validate the effectiveness of each module within the framework.

Massive amounts of data are the foundation of data-driven recommendation models. As an inherent nature of big data, data heterogeneity widely exists in real-world recommendation systems. It reflects the differences in the properties among sub-populations. Ignoring the heterogeneity in recommendation data could mislead the models, hurt the sub-populational robustness, and finally limit the performance of recommendation models. However, data heterogeneity has not received substantial attention within the recommendation community, prompting us to adequately explore and exploit data heterogeneity to solve these challenges and enhance data analysis. In this study, we specifically focus on two representative categories of heterogeneity in recommendation data: heterogeneity of prediction mechanism and covariate distribution. To explore the data heterogeneity, we propose an algorithm based on bilevel clustering. Additionally, we demonstrate how the explored data heterogeneity can be exploited for prediction and debias in recommendation scenarios, specifically by building models using multiple sub-models and augmenting the propensity score estimation. Extensive experiments conducted on real-world data substantiate the existence of heterogeneity in recommendation data and validate the effectiveness of exploring and exploiting data heterogeneity in improving recommendation performance.

Recommendation systems are frequently challenged by pervasive biases in the training set that can compromise model effectiveness. To address this issue, various debiasing techniques have been developed to eliminate biases and produce debiased models. However, when encountering varying test environments, some data patterns manifested by the training data could be beneficial to the model’s performance. Completely removing biases may overlook the beneficial data patterns and consequently diminish recommendation accuracy. Thus, it is crucial to carefully integrate certain biases to optimize performance, while the ideal level of bias integration is highly dependent on the test environment. Moreover, these systems operate in dynamic scenarios where the test environments could vary, necessitating an adaptive integration strategy customized to the environment. Our research establishes that discrepancies in predictions of models can guide the selection of the most fitting model for specific situations. Building on this understanding, we present AdaptSel, a pioneering method for the adaptive selection of the superior model during the testing phase. Empirical evaluations substantiate the foundational assumptions of AdaptSel, accentuating its effectiveness in adaptively selecting the most suitable model for varying test environments.

Domain generalization aims to solve the challenge of Out-of-Distribution (OOD) generalization by leveraging common knowledge learnedfrom multiple training domains to generalize to unseen test domains. To accurately evaluate the OOD generalization ability, it is required that test data information is unavailable. However, the current domain generalization protocol may still have potential test data information leakage. This paper examines the risks of test data information leakage from two aspects of the current evaluation protocol: supervised pretraining on ImageNet and oracle model selection. We propose modifications to the current protocol that we should employ self-supervised pretraining or train from scratch instead of employing the current supervised pretraining, and we should use multiple test domains. These would result in a more precise eval-uation of OOD generalization ability. We also rerun the algorithms with the modified protocol and introduce new leaderboards to encourage future research in domain gen-eralization with a fairer comparison.

Despite notable progress in enhancing the capability of machine learning against distribution shifts, training data quality remains a bottleneck for cross-distribution generalization. Recently, from a data-centric perspective, there have been considerable efforts to improve model performance through refining the preparation of training data. Inspired by realistic scenarios, this paper addresses a practical requirement of acquiring training samples from various domains on a limited budget to facilitate model generalization to target test domain with distribution shift. Our empirical evidence indicates that the advance in data acquisition can significantly benefit the model performance on shifted data. Additionally, by leveraging unlabeled test domain data, we introduce a Domain-wise Active Acquisition framework. This framework iteratively optimizes the data acquisition strategy as training samples are accumulated, theoretically ensuring the effective approximation of test distribution. Extensive real-world experiments demonstrate our proposal’s advantages in machine learning applications. The code is available at https://github.com/dongbaili/DAA.

Despite the encouraging successes in numerous applications, machine learning methods grounded on the i.i.d. assumption often experience performance deterioration when confronted with the distribution shift between training and test data. This challenge has instigated recent research endeavors focusing on out-of-distribution (OOD) generalization. A particularly pervasive and intricate OOD problem is to enhance the model's generalization ability by training it on samples drawn from a single environment. In response to the problem, we propose a simple model-agnostic method tailored for a practical OOD scenario in this paper. Our approach centers on pursuing robust weighted empirical risks, utilizing randomly shifted training distributions derived through a specific sample-based weighting strategy. Furthermore, we theoretically establish that the expected risk of the shifted training distribution can bound the expected risk of the test distribution. This theoretical foundation ensures the improved prediction performance of our method when employed in uncertain test distributions. Extensive experiments conducted on diverse real-world datasets affirm the effectiveness of our method, highlighting its potential to address the distribution shifts in machine learning applications.

Shifts in the marginal distribution of covariates from training to the test phase, named covariate-shifts, often lead to unstable prediction performance across agnostic testing data, especially under model misspecification. Recent literature on invariant learning attempts to learn an invariant predictor from heterogeneous environments. However, the performance of the learned predictor depends heavily on the availability and quality of provided environments. In this paper, we propose a simple and effective non-parametric method for generating heterogeneous environments via Random Sample Weighting (RSW). Given the training dataset from a single source environment, we randomly generate a set of covariate-determining sample weights and use each weighted training distribution to simulate an environment. We theoretically show that under appropriate conditions, such random sample weighting can produce sufficient heterogeneity to be exploited by common invariance constraints to find the invariant variables for stable prediction under covariate shifts. Extensive experiments on both simulated and real-world datasets clearly validate the effectiveness of our method.

The problem of covariate-shift generalization has attracted intensive research attention. Previous stable learning algorithms employ sample reweighting schemes to decorrelate the covariates when there is no explicit domain information about training data. However, with finite samples, it is difficult to achieve the desirable weights that ensure perfect independence to get rid of the unstable variables. Besides, decorrelating within stable variables may bring about high variance of learned models because of the over-reduced effective sample size. A tremendous sample size is required for these algorithms to work. In this paper, with theoretical justification, we propose SVI (Sparse Variable Independence) for the covariate-shift generalization problem. We introduce sparsity constraint to compensate for the imperfectness of sample reweighting under the finite-sample setting in previous methods. Furthermore, we organically combine independence-based sample reweighting and sparsity-based variable selection in an iterative way to avoid decorrelating within stable variables, increasing the effective sample size to alleviate variance inflation. Experiments on both synthetic and real-world datasets demonstrate the improvement of covariate-shift generalization performance brought by SVI.

Despite the remarkable performance that modern deep neural networks have achieved on independent and identically distributed (I.I.D.) data, they can crash under distribution shifts. Most current evaluation methods for do-main generalization (DG) adopt the leave-one-out strategy as a compromise on the limited number of domains. We propose a large-scale benchmark with extensive labeled domains named$NICO^{++}$along with more rational evaluation methods for comprehensively evaluating DC algorithms. To evaluate DG datasets, we propose two metrics to quantify covariate shift and concept shift, respectively. Two novel generalization bounds from the perspective of data construction are proposed to prove that limited concept shift and significant covariate shift favor the evaluation capability for generalization. Through extensive experiments,$NlCO^{++}$shows its superior evaluation capability compared with current DG datasets and its contribution in alleviating unfairness caused by the leak of oracle knowledge in model selection. The data and code for the benchmark based on$NICO^{++}$are available at https://github.com/xxgege/NICO-plus.

Deep learning has achieved tremendous success in recent years, but most of these successes are built on an independent and identically distributed (IID) assumption. This somewhat hinders the application of deep learning to the more challenging out-of-distribution (OOD) scenarios. Although many OOD methods have been proposed to address this problem and have obtained good performance on testing data that is of major shifts with training distributions, interestingly, we experimentally find that these methods achieve excellent OOD performance by making a great sacrifice of the IID performance. We call this finding the IID-OOD dilemma. Clearly, in real-world applications, distribution shifts between training and testing data are often uncertain, where shifts could be minor, and even close to the IID scenario, and thus it is truly important to design a deep model with the balanced generalization ability between IID and OOD. To this end, in this paper, we investigate an intriguing problem of balancing IID and OOD generalizations and propose a novel Model Agnostic adaPters (MAP) method, which is more reliable and effective for distribution-shift-agnostic real-world data. Our key technical contribution is to use auxiliary adapter layers to incorporate the inductive bias of IID into OOD methods. To achieve this goal, we apply a bilevel optimization to explicitly model and optimize the coupling relationship between the OOD model and auxiliary adapter layers. We also theoretically give a first-order approximation to save computational time. Experimental results on six datasets successfully demonstrate that MAP can greatly improve the performance of IID while achieving good OOD performance.

Current recommender systems have achieved great successes in online services, such as E-commerce and social media. However, they still suffer from the performance degradation in real scenarios, because various biases always occur in the generation process of user behaviors. Despite the recent development of addressing some specific type of bias, a variety of data bias, some of which are even unknown, are often mixed up in real applications. Although the uniform (or unbiased) data may help for the purpose of general debiasing, such data can either be hardly available or induce high experimental cost. In this paper, we consider a more practical setting where we aim to conduct general debiasing with the biased observational data alone. We assume that the observational user behaviors are determined by invariant preference (i.e. a user's true preference) and the variant preference (affected by some unobserved confounders). We propose a novel recommendation framework called InvPref which iteratively decomposes the invariant preference and variant preference from biased observational user behaviors by estimating heterogeneous environments corresponding to different types of latent bias. Extensive experiments, including the settings of general debiasing and specific debiasing, verify the advantages of our method.

In spite of the tremendous development of recommender system owing to the progressive capability of machine learning recently, the current recommender system is still vulnerable to the distribution shift of users and items in realistic scenarios, leading to the sharp decline of performance in testing environments. It is even more severe in many common applications where only the implicit feedback from sparse data is available. Hence, it is crucial to promote the performance stability of recommendation method in different environments. In this work, we first make a thorough analysis of implicit recommendation problem from the viewpoint of out-of-distribution (OOD) generalization. Then under the guidance of our theoretical analysis, we propose to incorporate the recommendation-specific DAG learner into a novel causal preference-based recommendation framework named CausPref, mainly consisting of causal learning of invariant user preference and anti-preference negative sampling to deal with implicit feedback. Extensive experimental results from real-world datasets clearly demonstrate that our approach surpasses the benchmark models significantly under types of out-of-distribution settings, and show its impressive interpretability.

Approaches based on deep neural networks have achieved striking performance when testing data and training data share similar distribution, but can significantly fail otherwise. Therefore, eliminating the impact of distribution shifts between training and testing data is crucial for building performance-promising deep models. Conventional methods assume either the known heterogeneity of training data (e.g. domain labels) or the approximately equal capacities of different domains. In this paper, we consider a more challenging case where neither of the above assumptions holds. We propose to address this problem by removing the dependencies between features via learning weights for training samples, which helps deep models get rid of spurious correlations and, in turn, concentrate more on the true connection between discriminative features and labels. Extensive experiments clearly demonstrate the effectiveness of our method on multiple distribution generalization benchmarks compared with state-of-the-art counterparts. Through extensive experiments on distribution generalization benchmarks including PACS, VLCS, MNIST-M, and NICO, we show the effectiveness of our method compared with state-of-the-art counterparts.

Discovering causal structure among a set of variables is a crucial task in various scientific and industrial scenarios. Given finite i.i.d. samples from a joint distribution, causal discovery is a challenging combinatorial problem in nature. The recent development in functional causal models, especially the NOTEARS provides a differentiable optimization framework for causal discovery. They formulate the structure learning problem as a task of maximum likelihood estimation over observational data (i.e., variable reconstruction) with specified structural constraints such as acyclicity and sparsity. Despite its success in terms of scalability, we find that optimizing the objectives of these differentiable methods is not always consistent with the correctness of learned causal graph especially when the variables carry heterogeneous noises (i.e., different noise types and noise variances) in real data from wild environments. In this paper, we provide the justification that their proneness to erroneous structures is mainly caused by the over-reconstruction problem, i.e., the noises of variables are absorbed into the variable reconstruction process, leading to the dependency among variable reconstruction residuals, and thus raise structure identifiability problems according to FCM theories. To remedy this, we propose a novel differentiable method DARING by imposing explicit residual independence constraint in an adversarial way. Extensive experimental results on both simulation and real data show that our proposed method is insensitive to the heterogeneity of external noise, and thus can significantly improve the causal discovery performances.

Matching is almost the first and most fundamental step in recommender systems, that is to quickly select hundreds or thousands of related entities from the whole commodity pool. Among all the matching methods, item-to-item (I2I) graph based matching is a handy and highly effective approach and is widely used in most applications, owing to the essential relationships of entities described in a powerful I2I graph. Yet, the I2I graph is not a ready-made product in a data source. To obtain it from users' behaviors, a common practice in the industry is to construct the graph based on the similarity of item embeddings or co-occurrence frequency directly. However, these methods tend to lose the complicated correlations (high-ordered or nonlinear) inside decision-making actions and cannot achieve the global optimal solution. Moreover, the correlations between items are usually contained in users' short-term actions, which are full of noise information (e.g. spurious association, missing connection). It is vitally important to filter out noise while generating the graph. In this paper, we propose a novel framework called Purified Graph Generation (PGG) dedicated to learn faithful I2I graph from sparse and noisy behavior data. We capture the 'confidence value' between user and item to get rid of exception action during decision making, and leverage it to re-sample purified sets that are fed into an unsupervised I2I graph structure learning framework called GPBG. Extensive experimental results from both simulation and real data demonstrate that our method could significantly benefit the performance of I2I graph compared to the typical baselines.

Nowadays graph has become a general and powerful representation to describe the rich relationships among different kinds of entities via the underlying patterns encoded in its structure. The knowledge (more generally) accumulated in graph is expected to be able to cross populations from one to another and the past to future. However the data collection process of graph generation is full of known or unknown sample selection biases, leading to spurious correlations among entities, especially in the non-stationary and heterogeneous environments. In this paper, we target the problem of learning stable graphs from multiple environments with selection bias. We purpose a Stable Graph Learning (SGL) framework to learn a graph that can capture general relational patterns which are irrelevant with the selection bias in an unsupervised way. Extensive experimental results from both simulation and real data demonstrate that our method could significantly benefit the generalization capacity of graph structure.

Estimating counterfactual outcome of different treatments from observational data is an important problem to assist decision making in a variety of fields. Among the various forms of treatment specification, bundle treatment has been widely adopted in many scenarios, such as recommendation systems and online marketing. The bundle treatment usually can be abstracted as a high dimensional binary vector, which makes it more challenging for researchers to remove the confounding bias in observational data. In this work, we assume the existence of low dimensional latent structure underlying bundle treatment. Via the learned latent representations of treatments, we propose a novel variational sample re-weighting (VSR) method to eliminate confounding bias by decorrelating the treatments and confounders. Finally, we conduct extensive experiments to demonstrate that the predictive model trained on this re-weighted dataset can achieve more accurate counterfactual outcome prediction.

Recent years have witnessed the success of the emerging hashing techniques in large-scale image retrieval. Owing to the great learning capacity, deep hashing has become one of the most promising solutions, and achieved attractive performance in practice. However, without semantic label information, the unsupervised deep hashing still remains an open question. In this paper, we propose a novel progressive generative hashing (PGH) framework to help learn a discriminative hashing network in an unsupervised way. Very different from existing studies, it first treats the hash codes as a kind of semantic condition for the similar image generation, and simultaneously feeds the original image and its codes into the generative adversarial networks (GANs). The real images together with the synthetic ones can further help train a discriminative hashing network based on a triplet loss. By iteratively inputting the learnt codes into the hash conditioned GANs, we can progressively enable the hashing network to discover the semantic relations. Extensive experiments on the widely-used image datasets demonstrate that PGH can significantly outperforms state-of-the-art unsupervised hashing methods.