Huiqi Deng

Qihan Ren, Huiqi Deng, Yunuo Chen et al.

In this paper, we focus on mean-field variational Bayesian Neural Networks (BNNs) and explore the representation capacity of such BNNs by investigating which types of concepts are less likely to be encoded by the BNN. It has been observed and studied that a relatively small set of interactive concepts usually emerge in the knowledge representation of a sufficiently-trained neural network, and such concepts can faithfully explain the network output. Based on this, our study proves that compared to standard deep neural networks (DNNs), it is less likely for BNNs to encode complex concepts. Experiments verify our theoretical proofs. Note that the tendency to encode less complex concepts does not necessarily imply weak representation power, considering that complex concepts exhibit low generalization power and high adversarial vulnerability. The code is available at https://github.com/sjtu-xai-lab/BNN-concepts.

Huilin Zhou, Hao Zhang, Huiqi Deng et al.

This paper explains the generalization power of a deep neural network (DNN) from the perspective of interactions. Although there is no universally accepted definition of the concepts encoded by a DNN, the sparsity of interactions in a DNN has been proved, i.e., the output score of a DNN can be well explained by a small number of interactions between input variables. In this way, to some extent, we can consider such interactions as interactive concepts encoded by the DNN. Therefore, in this paper, we derive an analytic explanation of inconsistency of concepts of different complexities. This may shed new lights on using the generalization power of concepts to explain the generalization power of the entire DNN. Besides, we discover that the DNN with stronger generalization power usually learns simple concepts more quickly and encodes fewer complex concepts. We also discover the detouring dynamics of learning complex concepts, which explains both the high learning difficulty and the low generalization power of complex concepts. The code will be released when the paper is accepted.

Huiqi Deng, Na Zou, Mengnan Du et al.

Various attribution methods have been developed to explain deep neural networks (DNNs) by inferring the attribution/importance/contribution score of each input variable to the final output. However, existing attribution methods are often built upon different heuristics. There remains a lack of a unified theoretical understanding of why these methods are effective and how they are related. To this end, for the first time, we formulate core mechanisms of fourteen attribution methods, which were designed on different heuristics, into the same mathematical system, i.e., the system of Taylor interactions. Specifically, we prove that attribution scores estimated by fourteen attribution methods can all be reformulated as the weighted sum of two types of effects, i.e., independent effects of each individual input variable and interaction effects between input variables. The essential difference among the fourteen attribution methods mainly lies in the weights of allocating different effects. Based on the above findings, we propose three principles for a fair allocation of effects to evaluate the faithfulness of the fourteen attribution methods.

Zirui He, Huiqi Deng, Haiyan Zhao et al.

Recent research has shown that large language models rely on spurious correlations in the data for natural language understanding (NLU) tasks. In this work, we aim to answer the following research question: Can we reduce spurious correlations by modifying the ground truth labels of the training data? Specifically, we propose a simple yet effective debiasing framework, named Soft Label Encoding (SoftLE). We first train a teacher model with hard labels to determine each sample's degree of relying on shortcuts. We then add one dummy class to encode the shortcut degree, which is used to smooth other dimensions in the ground truth label to generate soft labels. This new ground truth label is used to train a more robust student model. Extensive experiments on two NLU benchmark tasks demonstrate that SoftLE significantly improves out-of-distribution generalization while maintaining satisfactory in-distribution accuracy.

Xinhao Zheng, Huiqi Deng, Quanshi Zhang

This paper focuses on the fundamental challenge of partitioning input variables in attribution methods for Explainable AI, particularly in Shapley value-based approaches. Previous methods always compute attributions given a predefined partition but lack theoretical guidance on how to form meaningful variable partitions. We identify that attribution conflicts arise when the attribution of a coalition differs from the sum of its individual variables' attributions. To address this, we analyze the numerical effects of AND-OR interactions in AI models and extend the Shapley value to a new attribution metric for variable coalitions. Our theoretical findings reveal that specific interactions cause attribution conflicts, and we propose three metrics to evaluate coalition faithfulness. Experiments on synthetic data, NLP, image classification, and the game of Go validate our approach, demonstrating consistency with human intuition and practical applicability.

Dongrui Liu, Qihan Ren, Chen Qian et al.

The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.

Huiqi Deng, Qihan Ren, Zhuofan Chen et al.

Understanding what kinds of cooperative structures deep neural networks (DNNs) can represent remains a fundamental yet insufficiently understood problem. In this work, we treat interactions as the fundamental units of such structure and investigate a largely unexplored question: how DNNs encode interactions under different levels of contextual complexity, and how these microscopic interaction patterns shape macroscopic representation capacity. To quantify this complexity, we use multi-order interactions [57], where each order reflects the amount of contextual information required to evaluate the joint interaction utility of a variable pair. This formulation enables a stratified analysis of cooperative patterns learned by DNNs. Building on this formulation, we develop a comprehensive study of interaction structure in DNNs. (i) We empirically discover a universal interaction bottleneck: across architectures and tasks, DNNs easily learn low-order and high-order interactions but consistently under-represent mid-order ones. (ii) We theoretically explain this bottleneck by proving that mid-order interactions incur the highest contextual variability, yielding large gradient variance and making them intrinsically difficult to learn. (iii) We further modulate the bottleneck by introducing losses that steer models toward emphasizing interactions of selected orders. Finally, we connect microscopic interaction structures with macroscopic representational behavior: low-order-emphasized models exhibit stronger generalization and robustness, whereas high-order-emphasized models demonstrate greater structural modeling and fitting capability. Together, these results uncover an inherent representational bias in modern DNNs and establish interaction order as a powerful lens for interpreting and guiding deep representations.

Huiqi Deng, Yibo Li, Quanshi Zhang et al.

Understanding why deep neural networks (DNNs) fail to generalize to unseen samples remains a long-standing challenge. Existing studies mainly examine changes in externally observable factors such as data, representations, or outputs, yet offer limited insight into how a model's internal decision mechanism evolves from training to test. To address this gap, we introduce Decision Pattern Shift (DPS), a new perspective that defines generalization through the stability of internal decision patterns and quantifies failure as their deviation from those learned during training. Specifically, we represent each sample's decision pattern as a GradCAM-based channel-contribution vector, which captures how feature channels collectively support a prediction, and we propose the DPS metric to measure its discrepancy from the class-average pattern. Empirical analyses across multiple datasets and architectures show that, (i) decision patterns form a highly structured, class-consistent space with strong intra-class cohesion and low inter-class confusion, enabling direct analysis of a model's decision logic; (ii) the DPS magnitude correlates linearly with the generalization gap (nearly all Pearson r > 0.8), revealing generalization as a systematic drift in the model's internal decision mechanism; (iii) the DPS spectrum organizes diverse generalization degradation scenarios (covering ideal generalization, in-distribution degradation, domain shift, out-of-distribution, and shortcut learning) into a continuous trajectory, providing a unified explanation of their failure modes. These findings open up new possibilities for early generalization-risk detection, failure-mode diagnosis, and channel-level defect localization.

Huiqi Deng, Hongbin Pei, Quanshi Zhang et al.

Attribution explanation is a typical approach for explaining deep neural networks (DNNs), inferring an importance or contribution score for each input variable to the final output. In recent years, numerous attribution methods have been developed to explain DNNs. However, a persistent concern remains unresolved, i.e., whether and which attribution methods faithfully reflect the actual contribution of input variables to the decision-making process. The faithfulness issue undermines the reliability and practical utility of attribution explanations. We argue that these concerns stem from three core challenges. First, difficulties arise in comparing attribution methods due to their unstructured heterogeneity, differences in heuristics, formulations, and implementations that lack a unified organization. Second, most methods lack solid theoretical underpinnings, with their rationales remaining absent, ambiguous, or unverified. Third, empirically evaluating faithfulness is challenging without ground truth. Recent theoretical advances provide a promising way to tackle these challenges, attracting increasing attention. We summarize these developments, with emphasis on three key directions: (i) Theoretical unification, which uncovers commonalities and differences among methods, enabling systematic comparisons; (ii) Theoretical rationale, clarifying the foundations of existing methods; (iii) Theoretical evaluation, rigorously proving whether methods satisfy faithfulness principles. Beyond a comprehensive review, we provide insights into how these studies help deepen theoretical understanding, inform method selection, and inspire new attribution methods. We conclude with a discussion of promising open problems for further work.

Haiyan Zhao, Hanjie Chen, Fan Yang et al.

Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this paper, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations, and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional machine learning models.

Dongrui Liu, Shaobo Wang, Jie Ren et al.

In this paper, we focus on a typical two-phase phenomenon in the learning of multi-layer perceptrons (MLPs), and we aim to explain the reason for the decrease of feature diversity in the first phase. Specifically, people find that, in the training of MLPs, the training loss does not decrease significantly until the second phase. To this end, we further explore the reason why the diversity of features over different samples keeps decreasing in the first phase, which hurts the optimization of MLPs. We explain such a phenomenon in terms of the learning dynamics of MLPs. Furthermore, we theoretically explain why four typical operations can alleviate the decrease of the feature diversity.

Huiqi Deng, Qihan Ren, Hao Zhang et al.

This paper explores the bottleneck of feature representations of deep neural networks (DNNs), from the perspective of the complexity of interactions between input variables encoded in DNNs. To this end, we focus on the multi-order interaction between input variables, where the order represents the complexity of interactions. We discover that a DNN is more likely to encode both too simple interactions and too complex interactions, but usually fails to learn interactions of intermediate complexity. Such a phenomenon is widely shared by different DNNs for different tasks. This phenomenon indicates a cognition gap between DNNs and human beings, and we call it a representation bottleneck. We theoretically prove the underlying reason for the representation bottleneck. Furthermore, we propose a loss to encourage/penalize the learning of interactions of specific complexities, and analyze the representation capacities of interactions of different complexities.

Jie Ren, Mingjie Li, Qirui Chen et al.

This paper aims to illustrate the concept-emerging phenomenon in a trained DNN. Specifically, we find that the inference score of a DNN can be disentangled into the effects of a few interactive concepts. These concepts can be understood as causal patterns in a sparse, symbolic causal graph, which explains the DNN. The faithfulness of using such a causal graph to explain the DNN is theoretically guaranteed, because we prove that the causal graph can well mimic the DNN's outputs on an exponential number of different masked samples. Besides, such a causal graph can be further simplified and re-written as an And-Or graph (AOG), without losing much explanation accuracy.

Huiqi Deng, Na Zou, Mengnan Du et al.

Attribution methods provide an insight into the decision-making process of machine learning models, especially deep neural networks, by assigning contribution scores to each individual feature. However, the attribution problem has not been well-defined, which lacks a unified guideline to the contribution assignment process. Furthermore, existing attribution methods often built upon various empirical intuitions and heuristics. There still lacks a general theoretical framework that not only can offer a good description of the attribution problem, but also can be applied to unifying and revisiting existing attribution methods. To bridge the gap, in this paper, we propose a Taylor attribution framework, which models the attribution problem as how to decide individual payoffs in a coalition. Then, we reformulate fourteen mainstream attribution methods into the Taylor framework and analyze these attribution methods in terms of rationale, fidelity, and limitation in the framework. Moreover, we establish three principles for a good attribution in the Taylor attribution framework, i.e., low approximation error, correct Taylor contribution assignment, and unbiased baseline selection. Finally, we empirically validate the Taylor reformulations and reveal a positive correlation between the attribution performance and the number of principles followed by the attribution method via benchmarking on real-world datasets.

Huiqi Deng, Na Zou, Weifu Chen et al.

Back propagation based visualizations have been proposed to interpret deep neural networks (DNNs), some of which produce interpretations with good visual quality. However, there exist doubts about whether these intuitive visualizations are related to the network decisions. Recent studies have confirmed this suspicion by verifying that almost all these modified back-propagation visualizations are not faithful to the model's decision-making process. Besides, these visualizations produce vague "relative importance scores", among which low values can't guarantee to be independent of the final prediction. Hence, it's highly desirable to develop a novel back-propagation framework that guarantees theoretical faithfulness and produces a quantitative attribution score with a clear understanding. To achieve the goal, we resort to mutual information theory to generate the interpretations, studying how much information of output is encoded in each input neuron. The basic idea is to learn a source signal by back-propagation such that the mutual information between input and output should be as much as possible preserved in the mutual information between input and the source signal. In addition, we propose a Mutual Information Preserving Inverse Network, termed MIP-IN, in which the parameters of each layer are recursively trained to learn how to invert. During the inversion, forward Relu operation is adopted to adapt the general interpretations to the specific input. We then empirically demonstrate that the inverted source signal satisfies completeness and minimality property, which are crucial for a faithful interpretation. Furthermore, the empirical study validates the effectiveness of interpretations generated by MIP-IN.

Huiqi Deng, Na Zou, Mengnan Du et al.

Attribution methods have been developed to understand the decision-making process of machine learning models, especially deep neural networks, by assigning importance scores to individual features. Existing attribution methods often built upon empirical intuitions and heuristics. There still lacks a general and theoretical framework that not only can unify these attribution methods, but also theoretically reveal their rationales, fidelity, and limitations. To bridge the gap, in this paper, we propose a Taylor attribution framework and reformulate seven mainstream attribution methods into the framework. Based on reformulations, we analyze the attribution methods in terms of rationale, fidelity, and limitation. Moreover, We establish three principles for a good attribution in the Taylor attribution framework, i.e., low approximation error, correct contribution assignment, and unbiased baseline selection. Finally, we empirically validate the Taylor reformulations and reveal a positive correlation between the attribution performance and the number of principles followed by the attribution method via benchmarking on real-world datasets.