论文检索

General Machine Learning · Representation Learning

Yao Huang, Huanran Chen, Shouwei Ruan, Yichi Zhang, Xingxing Wei, Yinpeng Dong

Recent advances in Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex tasks such as mathematics and coding. However, these models frequently exhibit a phenomenon known as *overthinking* during inference, characterized by excessive validation loops and redundant deliberation, leading to substantial computational overheads. In this paper, we aim to mitigate overthinking by investigating the underlying mechanisms from the perspective of mechanistic interpretability. We first showcase that the tendency of overthinking can be effectively captured by a single direction in the model's activation space and the issue can be eased by intervening the activations along this direction. However, this efficacy soon reaches a plateau and even deteriorates as the intervention strength increases. We therefore systematically explore the activation space and find that the overthinking phenomenon is actually tied to a low-dimensional manifold, which indicates that the limited effect stems from the noises introduced by the high-dimensional steering direction. Based on this insight, we propose **Manifold Steering**, a novel approach that elegantly projects the steering direction onto the low-dimensional activation manifold given the theoretical approximation of the interference noise. Extensive experiments on DeepSeek-R1 distilled models validate that our method reduces output tokens by up to 71\% while maintaining and even improving the accuracy on several mathematical benchmarks. Our method also exhibits robust cross-domain transferability, delivering consistent token reduction performance in code generation and knowledge-based QA tasks. Code is available at: https://github.com/Aries-iai/Manifold_Steering.

Social Aspects · Accountability, Transparency and Interpretability

Sheng-Yu Wang, Aaron Hertzmann, Alexei Efros, Richard Zhang, Jun-Yan Zhu

Data attribution for text-to-image models aims to identify the training images that most significantly influenced a generated output. Existing attribution methods involve considerable computational resources for each query, making them impractical for real-world applications. We propose a novel approach for scalable and efficient data attribution. Our key idea is to distill a slow, unlearning-based attribution method to a feature embedding space for efficient retrieval of highly influential training images. During deployment, combined with efficient indexing and search methods, our method successfully finds highly influential images without running expensive attribution algorithms. We show extensive results on both medium-scale models trained on MSCOCO and large-scale Stable Diffusion models trained on LAION, demonstrating that our method can achieve better or competitive performance in a few seconds, faster than existing methods by 2,500x - 400,000x. Our work represents a meaningful step towards the large-scale application of data attribution methods on real-world models such as Stable Diffusion.

Applications · Time Series

Yutong Feng, Xu Liu, Yutong Xia, Yuxuan Liang

Accurately modeling complex dynamic spatio-temporal systems requires capturing flow-mediated interdependencies and context-sensitive interaction dynamics. Existing methods, predominantly graph-based or attention-driven, rely on similarity-driven connectivity assumptions, neglecting asymmetric flow exchanges that govern system evolution. We propose Spatio-Temporal Flow, a physics-inspired paradigm that explicitly models dynamic node couplings through quantifiable flow transfers governed by conservation principles. Building on this, we design FlowNet, a novel architecture leveraging flow tokens as information carriers to simulate source-to-destination transfers via Flow Allocation Modules, ensuring state redistribution aligns with physical laws. FlowNet dynamically adjusts the interaction radius through an Adaptive Spatial Masking module, suppressing irrelevant noise while enabling context-aware propagation. A cascaded architecture enhances scalability and nonlinear representation capacity. Experiments demonstrate that FlowNet significantly outperforms existing SOTA approaches on seven metrics in the modeling of three real-world systems, validating its efficiency and physical interpretability. We establish a principled methodology for modeling complex systems through spatio-temporal flow interactions.

Deep Learning · Algorithms

Chang Nie

Originating in quantum physics, tensor networks (TNs) have been widely adopted as exponential machines and parametric decomposers for recognition tasks. Typical TN models, such as Matrix Product States (MPS), have not yet achieved successful application in natural image recognition. When employed, they primarily serve to compress parameters within pre-existing networks, thereby losing their distinctive capability to enhance exponential-order feature interactions. This paper introduces a novel architecture named Deep Tree Tensor Network (DTTN), which captures -order multiplicative interactions across features through multilinear operations, while essentially unfolding into a tree-like TN topology with the parameter-sharing property. DTTN is stacked with multiple antisymmetric interacting modules (AIMs), and this design facilitates efficient implementation. Furthermore, our theoretical analysis demonstrates the equivalence between quantum-inspired TN models and polynomial/multilinear networks under specific conditions. We posit that the DTTN could catalyze more interpretable research within this field. The proposed model is evaluated across multiple benchmarks and application domains, demonstrating superior performance compared to both peer methods and state-of-the-art architectures. Our code will be made publicly available upon publication.

Social Aspects · Fairness, Equity, Justice and Safety

Zuhair Hasan Shaik, Abdullah Mazhar, Aseem Srivastava, Md Shad Akhtar

Large Language Models have demonstrated impressive fluency across diverse tasks, yet their tendency to produce toxic content remains a critical challenge for AI safety and public trust. Existing toxicity mitigation approaches primarily manipulate individual neuron activations, but these methods suffer from instability, context dependence, and often compromise the model’s core language abilities. To address these shortcomings, we investigate three key questions: the stability of neuron-level toxicity indicators, the advantages of structural (layer-wise) representations, and the interpretability of mechanisms driving toxic generation. Through extensive experiments on Jigsaw and ToxiCN datasets, we show that aggregated layer-wise features provide more robust signals than single neurons. Moreover, we observe conceptual limitations in prior works that conflate toxicity detection experts and generation experts within neuron-based interventions. To mitigate this, we propose a novel principled intervention technique, EigenShift, based on eigen-decomposition of the language model’s final output layer. This method selectively targets generation-aligned components, enabling precise toxicity suppression without impairing linguistic competence. Our method requires no additional training or fine-tuning, incurs minimal computational cost, and is grounded in rigorous theoretical analysis.

Yassine ABBAHADDOU, Amine Aboussalah

Generative models such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and diffusion models often suffer from mode collapse, failing to reproduce the full diversity of their training data. While this problem has been extensively studied in image generation, it remains largely unaddressed for time series. We introduce a formal definition of mode collapse for time series and propose DMD-GEN, a geometry-aware metric that quantifies its severity. DMD-GEN leverages Dynamic Mode Decomposition (DMD) to extract coherent temporal structures and uses Optimal Transport between DMD eigenvectors to measure discrepancies in underlying dynamics. By representing the subspaces spanned by the DMD eigenvectors as point structures on a Grassmann manifold, and comparing them via Wasserstein distances computed from principal angles, DMD-GEN enables a principled geometric comparison between real and generated sequences. The metric is efficient, requiring no additional training, supports mini-batch evaluation, and is easily parallelizable. Beyond quantification, DMD-GEN offers interpretability by revealing which dynamical modes are distorted or missing in the generated data. Experiments on synthetic and real-world datasets using TimeGAN, TimeVAE, and DiffusionTS show that DMD-GEN aligns with existing metrics while providing the first principled framework for detecting and interpreting mode collapse in time series.

Social Aspects · Accountability, Transparency and Interpretability

Landon Butler, Abhineet Agarwal, Justin Kang, Yigit Efe Erginbas, Bin Yu, Kannan Ramchandran

Large Language Models (LLMs) have achieved remarkable performance by capturing complex interactions between input features. To identify these interactions, most existing approaches require enumerating all possible combinations of features up to a given order, causing them to scale poorly with the number of inputs $n$. Recently, Kang et al. (2025) proposed SPEX, an information-theoretic approach that uses interaction sparsity to scale to $n \approx 10^3$ features. SPEX greatly improves upon prior methods but requires tens of thousands of model inferences, which can be prohibitive for large models. In this paper, we observe that LLM feature interactions are often *hierarchical*—higher-order interactions are accompanied by their lower-order subsets—which enables more efficient discovery. To exploit this hierarchy, we propose ProxySPEX, an interaction attribution algorithm that first fits gradient boosted trees to masked LLM outputs and then extracts the important interactions. Experiments across four challenging high-dimensional datasets show that ProxySPEX more faithfully reconstructs LLM outputs by 20\% over marginal attribution approaches while using *$10\times$ fewer inferences* than SPEX. By accounting for interactions, ProxySPEX efficiently identifies the most influential features, providing a scalable approximation of their Shapley values. Further, we apply ProxySPEX to two interpretability tasks. *Data attribution*, where we identify interactions among CIFAR-10 training samples that influence test predictions, and *mechanistic interpretability*, where we uncover interactions between attention heads, both within and across layers, on a question-answering task. The ProxySPEX algorithm is available at .

Computer Vision · Vision Models & Multimodal

Jonggwon Park, Byungmu Yoon, Soobum Kim, Kyoyun Choi

Recent advancements in multimodal models have significantly improved vision-language (VL) alignment in radiology. However, existing approaches struggle to effectively utilize complex radiology reports for learning and offer limited interpretability through attention probability visualizations. To address these challenges, we introduce $\textbf{RadZero}$, a novel framework for VL alignment in chest X-ray with zero-shot multi-task capability. A key component of our approach is $\textbf{VL-CABS}$ ($\textbf{V}$ision-$\textbf{L}$anguage $\textbf{C}$ross-$\textbf{A}$ttention $\textbf{B}$ased on $\textbf{S}$imilarity), which aligns text embeddings with local image features for interpretable, fine-grained VL reasoning. RadZero leverages large language models to extract concise semantic sentences from radiology reports and employs multi-positive contrastive training to effectively capture relationships between images and multiple relevant textual descriptions. It uses a pre-trained vision encoder with additional trainable Transformer layers, allowing efficient high-resolution image processing. By computing similarity between text embeddings and local image patch features, VL-CABS enables zero-shot inference with similarity probability for classification, and pixel-level VL similarity maps for grounding and segmentation. Experimental results on public chest radiograph benchmarks show that RadZero outperforms state-of-the-art methods in zero-shot classification, grounding, and segmentation. Furthermore, VL similarity map analysis highlights the potential of VL-CABS for improving explainability in VL alignment. Additionally, qualitative evaluation demonstrates RadZero's capability for open-vocabulary semantic segmentation, further validating its effectiveness in medical imaging. Code is available at https://github.com/deepnoid-ai/RadZero.

Social Aspects · Accountability, Transparency and Interpretability

Darin Tsui, Aryan Musharaf, Yigit Efe Erginbas, Justin Kang, Amirali Aghazadeh

The growing adoption of machine learning models for biological sequences has intensified the need for interpretable predictions, with Shapley values emerging as a theoretically grounded standard for model explanation. While effective for local explanations of individual input sequences, scaling Shapley-based interpretability to extract global biological insights requires evaluating thousands of sequences—incurring exponential computational cost per query. We introduce SHAP zero, a novel algorithm that amortizes the cost of Shapley value computation across large-scale biological datasets. After a one-time model sketching step, SHAP zero enables near-zero marginal cost for future queries by uncovering an underexplored connection between Shapley values, high-order feature interactions, and the sparse Fourier transform of the model. Applied to models of guide RNA efficacy, DNA repair outcomes, and protein fitness, SHAP zero explains predictions orders of magnitude faster than existing methods, recovering rich combinatorial interactions previously inaccessible at scale. This work opens the door to principled, efficient, and scalable interpretability for black-box sequence models in biology.

Deep Learning · Generative Models and Autoencoders

Qingyang Zhang, Haitao Wu, Changqing Zhang, Peilin Zhao, Yatao Bian

Existing methods to enhance the reasoning capability of large language models predominantly rely on supervised fine-tuning (SFT) followed by reinforcement learning (RL) on reasoning-specific data. These approaches critically depend on external supervisions--such as labeled reasoning traces, verified golden answers, or pre-trained reward models. In this work, we propose Entropy Minimized Policy Optimization (EMPO), which makes an early attempt at fully unsupervised LLM reasoning incentivization. By minimizing the semantic entropy of LLMs on unlabeled questions, EMPO achieves competitive performance compared to supervised counterparts. Specifically, without any supervised signals, EMPO boosts the accuracy of Qwen2.5-Math-7B Base from 33.7\% to 51.6\% on math benchmarks and improves the accuracy of Qwen2.5-7B Base from 32.1\% to 50.1\% on MMLU-Pro. Primary analysis are also provided to interpret the effectiveness of EMPO. Code is available at https://github.com/QingyangZhang/EMPO.

Social Aspects · Accountability, Transparency and Interpretability

Lin Zhang, Wenshuo Dong, Zhuoran Zhang, Shu Yang, Lijie Hu, Ninghao Liu, Pan Zhou, Di Wang

Understanding the internal mechanisms of transformer-based language models remains challenging. Mechanistic interpretability based on circuit discovery aims to reverse engineer neural networks by analyzing their internal processes at the level of computational subgraphs. In this paper, we revisit existing gradient-based circuit identification methods and find that their performance is either affected by the zero-gradient problem or saturation effects, where edge attribution scores become insensitive to input changes, resulting in noisy and unreliable attribution evaluations for circuit components. To address the saturation effect, we propose Edge Attribution Patching with GradPath (EAP-GP), EAP-GP introduces an integration path, starting from the input and adaptively following the direction of the difference between the gradients of corrupted and clean inputs to avoid the saturated region. This approach enhances attribution reliability and improves the faithfulness of circuit identification. We evaluate EAP-GP on 6 datasets using GPT-2 Small, GPT-2 Medium, and GPT-2 XL. Experimental results demonstrate that EAP-GP outperforms existing methods in circuit faithfulness, achieving improvements up to 17.7\%. Comparisons with manually annotated ground-truth circuits demonstrate that EAP-GP achieves precision and recall comparable to or better than previous approaches, highlighting its effectiveness in identifying accurate circuits.

Deep Learning · Algorithms

Gustavo Perez, Stella X. Yu

Classical image filters, such as those for averaging or differencing, are carefully normalized to ensure consistency, interpretability, and to avoid artifacts like intensity shifts, halos, or ringing. In contrast, convolutional filters learned end-to-end in deep networks lack such constraints. Although they may resemble wavelets and blob/edge detectors, they are not normalized in the same or any way. Consequently, when images undergo atmospheric transfer, their responses become distorted, leading to incorrect outcomes. We address this limitation by proposing filter normalization, followed by learnable scaling and shifting, akin to batch normalization. This simple yet effective modification ensures that the filters are atmosphere-equivariant, enabling co-domain symmetry. By integrating classical filtering principles into deep learning (applicable to both convolutional neural networks and convolution-dependent vision transformers), our method achieves significant improvements on artificial and natural intensity variation benchmarks. Our ResNet34 could even outperform CLIP by a large margin. Our analysis reveals that unnormalized filters degrade performance, whereas filter normalization regularizes learning, promotes diversity, and improves robustness and generalization.

Deep Learning · Attention Mechanisms

Qishuai Wen, Zhiyuan Huang, Chun-Guang Li

Attention mechanisms have achieved significant empirical success in multiple fields, but their underlying optimization objectives remain unclear yet. Moreover, the quadratic complexity of self-attention has become increasingly prohibitive. Although interpretability and efficiency are two mutually reinforcing pursuits, prior work typically investigates them separately. In this paper, we propose a unified optimization objective that derives inherently interpretable and efficient attention mechanisms through algorithm unrolling. Precisely, we construct a gradient step of the proposed objective with a set of forward-pass operations of our \emph{Contract-and-Broadcast Self-Attention} (CBSA), which compresses input tokens towards low-dimensional structures by contracting a few representatives of them. This novel mechanism can not only scale linearly by fixing the number of representatives, but also covers the instantiations of varied attention mechanisms when using different sets of representatives. We conduct extensive experiments to demonstrate comparable performance and superior advantages over black-box attention mechanisms on visual tasks. Our work sheds light on the integration of interpretability and efficiency, as well as the unified formula of attention mechanisms. Code is available at \href{https://github.com/QishuaiWen/CBSA}{this https URL}.

Social Aspects · Accountability, Transparency and Interpretability

Hang Chen, Jiaying Zhu, Xinyu Yang, Wenya Wang

Circuit discovery has gradually become one of the prominent methods for mechanistic interpretability, and research on circuit completeness has also garnered increasing attention. Methods of circuit discovery that do not guarantee completeness not only result in circuits that are not fixed across different runs but also cause key mechanisms to be omitted. The nature of incompleteness arises from the presence of OR gates within the circuit, which are often only partially detected in standard circuit discovery methods. To this end, we systematically introduce three types of logic gates: AND, OR, and ADDER gates, and decompose the circuit into combinations of these logical gates. Through the concept of these gates, we derive the minimum requirements necessary to achieve faithfulness and completeness. Furthermore, we propose a framework that combines noising-based and denoising-based interventions, which can be easily integrated into existing circuit discovery methods without significantly increasing computational complexity. This framework is capable of fully identifying the logic gates and distinguishing them within the circuit. In addition to the extensive experimental validation of the framework's ability to restore the faithfulness, completeness, and sparsity of circuits, using this framework, we uncover fundamental properties of the three logic gates, such as their proportions and contributions to the output, and explore how they behave among the functionalities of language models.

Computer Vision · Vision Models & Multimodal

Shufan Shen, Junshu Sun, Qingming Huang, Shuhui Wang

The alignment of vision-language representations endows current Vision-Language Models (VLMs) with strong multi-modal reasoning capabilities. However, the interpretability of the alignment component remains uninvestigated due to the difficulty in mapping the semantics of multi-modal representations into a unified concept set. To address this problem, we propose VL-SAE, a sparse autoencoder that encodes vision-language representations into its hidden activations. Each neuron in the hidden layer correlates to a concept represented by semantically similar images and texts, thereby interpreting these representations with a unified concept set. To establish the neuron-concept correlation, we encourage semantically similar representations to exhibit consistent neuron activations during self-supervised training. First, to measure the semantic similarity of multi-modal representations, we perform their alignment in an explicit form based on cosine similarity. Second, we construct the VL-SAE with a distance-based encoder and two modality-specific decoders to ensure the activation consistency of semantically similar representations. Experiments across multiple VLMs (e.g., CLIP, LLaVA) demonstrate the superior capability of VL-SAE in interpreting and enhancing the vision-language alignment. For interpretation, the alignment between vision and language representations can be understood by comparing their semantics with concepts. For enhancement, the alignment can be strengthened by aligning vision-language representations at the concept level, contributing to performance improvements in downstream tasks, including zero-shot image classification and hallucination elimination. Codes are provided in the supplementary and will be released to GitHub.

Applications · Neuroscience, Cognitive Science

Anushri Arora, Jonathan Pillow

Low-rank recurrent neural networks (RNNs) provide a powerful framework for characterizing how neural systems solve complex cognitive tasks. However, fitting and interpreting these networks remains an important open problem. In this paper, we develop new methods for efficiently fitting low-rank RNNs in ''teacher-training'' settings. In particular, we build upon the neural engineering framework (NEF), in which RNNs are viewed as approximating an ordinary differential equation (ODE) of interest using a set of random nonlinear basis functions. This view provides geometric insight into how the choice of neural nonlinearity (e.g. tanh, ReLU) and the distribution of model parameters affects an RNN's representational capacity. We adapt this framework for online training and demonstrate better performance with significantly smaller networks compared to FORCE. Additionally, we outperform backpropagation-trained networks of similar size, while requiring substantially less training time. Next, we ask: how many neurons---and what distribution over their parameters---are needed to approximate a given dynamical system? To address this, we introduce methods for finding the smallest low-rank RNN to approximate a given dynamical system using an extension of orthogonal matching pursuit (OMP). We then consider infinite unit low-rank RNNs, which converge to a Gaussian Process (GP) over ODEs. In particular, we show that we can optimize the distribution over RNN parameters using the marginal likelihood under the equivalent GP covariance function---which can be computed in closed form for particular choices of nonlinearity. This results in substantially better training performance even for finite low-rank RNNs. Finally, we describe active learning methods for low-rank RNNs, which speed up training through the selection of maximally informative activity patterns.

Social Aspects · Accountability, Transparency and Interpretability

Jiaran Ye, Zijun Yao, Zhidian Huang, Liangming Pan, Jinxin Liu, Yushi Bai, Amy Xin, Liu Weichuan, Xiaoyin Che, Lei Hou 等

Recent work suggests that large language models (LLMs) can perform multi-hop reasoning implicitly---producing correct answers without explicitly verbalizing intermediate steps---but the underlying mechanisms remain poorly understood. In this paper, we study how such implicit reasoning emerges by training transformers from scratch in a controlled symbolic environment. Our analysis reveals a three-stage developmental trajectory: early memorization, followed by in-distribution generalization, and eventually cross-distribution generalization. We find that training with atomic triples is not necessary but accelerates learning, and that second-hop generalization relies on query-level exposure to specific compositional structures. To interpret these behaviors, we introduce two diagnostic tools: cross-query semantic patching, which identifies semantically reusable intermediate representations, and a cosine-based representational lens, which reveals that successful reasoning correlates with the cosine-base clustering in hidden space. This clustering phenomenon in turn provides a coherent explanation for the behavioral dynamics observed across training, linking representational structure to reasoning capability. These findings provide new insights into the interpretability of implicit multi-hop reasoning in LLMs, helping to clarify how complex reasoning processes unfold internally and offering pathways to enhance the transparency of such models.

Computer Vision · Image and Video Generation

Hanru Bai, Weiyang Ding, Difan Zou

Diffusion models have achieved impressive success in high-fidelity image generation but suffer from slow sampling due to their inherently iterative denoising process. While recent one-step methods accelerate inference by learning direct noise-to-image mappings, they sacrifice the interpretability and fine-grained control intrinsic to diffusion dynamics, key advantages that enable applications like editable generation. To resolve this dichotomy, we introduce **Hierarchical Koopman Diffusion**, a novel framework that achieves both one-step sampling and interpretable generative trajectories. Grounded in Koopman operator theory, our method lifts the nonlinear diffusion dynamics into a latent space where evolution is governed by globally linear operators, enabling closed-form trajectory solutions. This formulation not only eliminates iterative sampling but also provides full access to intermediate states, allowing manual intervention during generation. To model the multi-scale nature of images, we design a hierarchical architecture that disentangles generative dynamics across spatial resolutions via scale-specific Koopman subspaces, capturing coarse-to-fine details systematically. We empirically show that the Hierarchical Koopman Diffusion not only achieves competitive one-step generation performance but also provides a principled mechanism for interpreting and manipulating the generative process through spectral analysis. Our framework bridges the gap between fast sampling and interpretability in diffusion models, paving the way for explainable image synthesis in generative modeling.

Applications · Robotics

Ruizhe Liu, Pei Zhou, Qian Luo, Li Sun, Jun CEN, Yibing Song, Yanchao Yang

Effective generalization in robotic manipulation requires representations that capture invariant patterns of interaction across environments and tasks. We present a self-supervised framework for learning hierarchical manipulation concepts that encode these invariant patterns through cross-modal sensory correlations and multi-level temporal abstractions without requiring human annotation. Our approach combines a cross-modal correlation network that identifies persistent patterns across sensory modalities with a multi-horizon predictor that organizes representations hierarchically across temporal scales. Manipulation concepts learned through this dual structure enable policies to focus on transferable relational patterns while maintaining awareness of both immediate actions and longer-term goals. Empirical evaluation across simulated benchmarks and real-world deployments demonstrates significant performance improvements with our concept-enhanced policies. Analysis reveals that the learned concepts resemble human-interpretable manipulation primitives despite receiving no semantic supervision. This work advances both the understanding of representation learning for manipulation and provides a practical approach to enhancing robotic performance in complex scenarios.

Deep Learning · Theory

Alessandro Benfenati, Alfio Ferrara, Alessio Marta, Davide Riva, Elisabetta Rocchetti

This paper introduces a general method for the exploration of equivalence classes in the input space of Transformer models. The proposed approach is based on sound mathematical theory which describes the internal layers of a Transformer architecture as sequential deformations of the input manifold. Using eigendecomposition of the pullback of the distance metric defined on the output space through the Jacobian of the model, we are able to reconstruct equivalence classes in the input space and navigate across them. Our method enables two complementary exploration procedures: the first retrieves input instances that produce the same class probability distribution as the original instance—thus identifying elements within the same equivalence class—while the second discovers instances that yield a different class probability distribution, effectively navigating toward distinct equivalence classes. Finally, we demonstrate how the retrieved instances can be meaningfully interpreted by projecting their embeddings back into a human-readable format.