论文检索

Theory · Interpretability and Visualization

Shawn Im, Changdae Oh, Zhen Fang, Sharon Li

Semantic associations such as the link between "bird" and "flew" are foundational for language modeling as they enable models to go beyond memorization and instead generalize and generate coherent text. Understanding how these associations are learned and represented in language models is essential for connecting deep learning with linguistic theory and developing a mechanistic foundation for large language models. In this work, we analyze how these associations emerge from natural language data in attention-based language models through the lens of training dynamics. By leveraging a leading-term approximation of the gradients, we develop closed-form expressions for the weights at early stages of training that explain how semantic associations first take shape. Through our analysis, we reveal that each set of weights of the transformer has closed-form expressions as simple compositions of three basis functions--bigram, token-interchangeability, and context mappings--reflecting the statistics in the text corpus and uncover how each component of the transformer captures the semantic association based on these compositions. Experiments on real-world LLMs demonstrate that our theoretical weight characterizations closely match the learned weights, and qualitative analyses further guide us on how our theorem shines light on interpreting the learned association in transformers.

Deep Learning · Generative Models and Autoencoders

María Martínez-García, Ricardo Vazquez Alvarez, Alejandro Lancho, Pablo Olmos, Isabel Valera

Concept Bottleneck Generative Models (CBGMs) incorporate a human-interpretable concept bottleneck layer, which makes them interpretable and steerable. However, designing such a layer for generative models poses the same challenges as for concept bottleneck models in a supervised context, if not greater ones. Deterministic mappings from the model inner representations to soft concepts in existing CBGMs: (i) limit steerable generation to modifying concepts in existing inputs; and, more importantly, (ii) are susceptible to *concept leakage*, which hinders their steerability. To address these limitations, we first introduce the Variational Hard Concept Bottleneck (VHCB) layer. The VHCB maps probabilistic estimates of binary latent variables to hard concepts, which have been shown to mitigate leakage. Remarkably, its probabilistic formulation enables direct generation from a specified set of concepts. Second, we propose a systematic evaluation framework for assessing the steerability of CBGMs across various tasks (e.g., activating and deactivating concepts). Our framework which allows us to empirically demonstrate that the VHCB layer consistently improves steerability.

Xiaoyuan Cheng, Wenxuan Yuan, Yiming Yang, Yuanzhao Zhang, Sibo Cheng, Yi He, Zhuo Sun

The Koopman operator provides a powerful framework for modeling dynamical systems and has attracted growing interest from the machine learning community. However, its infinite-dimensional nature makes identifying suitable finite-dimensional subspaces challenging, especially for deep architectures. We argue that these difficulties come from suboptimal representation learning, where latent variables fail to balance expressivity and simplicity. This tension is closely related to the information bottleneck (IB) dilemma: constructing compressed representations that are both compact and predictive. Rethinking Koopman learning through this lens, we demonstrate that latent mutual information promotes simplicity, yet an overemphasis on simplicity may cause latent space to collapse onto a few dominant modes. In contrast, expressiveness is sustained by the von Neumann entropy, which prevents such collapse and encourages mode diversity. This insight leads us to propose an information-theoretic Lagrangian formulation that explicitly balances this tradeoff. Furthermore, we propose a new algorithm based on the Lagrangian formulation that encourages both simplicity and expressiveness, leading to a stable and interpretable Koopman representation. Beyond quantitative evaluations, we further visualize the learned manifolds under our representations, observing empirical results consistent with our theoretical predictions. Finally, we validate our approach across a diverse range of dynamical systems, demonstrating improved performance over existing Koopman learning methods.

Usha Bhalla, Alex Oesterling, Claudio Mayrink Verdun, Hima Lakkaraju, Flavio Calmon

Translating the internal representations and computations of models into concepts that humans can understand is a key goal of interpretability. While recent dictionary learning methods such as Sparse Autoencoders (SAEs) provide a promising route to discover human-interpretable features, they often only recover token-specific, noisy, or highly local concepts. We argue that this limitation stems from neglecting the temporal structure of language, where semantic content typically evolves smoothly over sequences. Building on this insight, we introduce Temporal Sparse Autoencoders (T-SAEs), which incorporate a novel contrastive loss encouraging consistent activations of high-level features over adjacent tokens. This simple yet powerful modification enables SAEs to disentangle semantic from syntactic features in a self-supervised manner. Across multiple datasets and models, T-SAEs recover smoother, more coherent semantic concepts without sacrificing reconstruction quality. Strikingly, they exhibit clear semantic structure despite being trained without explicit semantic signal, offering a new pathway for unsupervised interpretability in language models.

General Machine Learning · Everything Else

Satchel Grant, Simon Jerome Han, Alexa Tartaglini, Christopher Potts

A common approach to mechanistic interpretability is to causally manipulate model representations via targeted interventions in order to understand what those representations encode. Here we ask whether such interventions create out-of-distribution (divergent) representations, and whether this raises concerns about how faithful their resulting explanations are to the target model in its natural state. First, we demonstrate theoretically and empirically that common causal intervention techniques often do shift internal representations away from the natural distribution of the target model. Then, we provide a theoretical analysis of two cases of such divergences: "harmless" divergences that occur in the behavioral null-space of the layer(s) of interest, and "pernicious" divergences that activate hidden network pathways and cause dormant behavioral changes. Finally, in an effort to mitigate the pernicious cases, we apply and modify the Counterfactual Latent (CL) loss from Grant (2025) allowing representations from causal interventions to remain closer to the natural distribution, reducing the likelihood of harmful divergences while preserving the interpretive power of the interventions. Together, these results highlight a path towards more reliable interpretability methods.

Optimization · Learning for Optimization

Zhenxing Xu, Yizhe Zhang, Weidong Bao, Hao Wang, Ming Chen, Haoran Ye, Wenzheng Jiang, Hui Yan, Ji Wang

Dynamically configuring algorithm hyperparameters is a fundamental challenge in computational intelligence. While learning-based methods offer automation, they suffer from prohibitive sample complexity and poor generalization. We introduce AutoEP, a novel framework that bypasses training entirely by leveraging Large Language Models (LLMs) as zero-shot reasoning engines for algorithm control. AutoEP's core innovation lies in a tight synergy between two components: (1) an online Exploratory Landscape Analysis (ELA) module that provides real-time, quantitative feedback on the search dynamics, and (2) a multi-LLM reasoning chain that interprets this feedback to generate adaptive hyperparameter strategies. This approach grounds high-level reasoning in empirical data, mitigating hallucination. Evaluated on three distinct metaheuristics across diverse combinatorial optimization benchmarks, AutoEP consistently outperforms state-of-the-art tuners, including neural evolution and other LLM-based methods. Notably, our framework enables open-source models like Qwen3-30B to match the performance of GPT-4, demonstrating a powerful and accessible new paradigm for automated hyperparameter design.Our code is available at https://anonymous.4open.science/r/AutoEP-3E11.

Theory · Interpretability and Visualization

Marco Bondaschi, Nived Rajaraman, Xiuying Wei, Razvan Pascanu, Caglar Gulcehre, Michael Gastpar, Ashok Makkuva

While transformer-based language models have driven the AI revolution thus far, their computational complexity has spurred growing interest in viable alternatives, such as structured state space sequence models (SSMs) and Selective SSMs. Among these, Mamba (S6) and its variant Mamba-2 have shown remarkable inference speed-ups over transformers while achieving comparable or superior performance on complex language modeling tasks. However, despite these architectural innovations and empirical successes, the fundamental learning capabilities of Mamba remain poorly understood. In this paper, we address this gap by studying in-context learning (ICL) on Markov chains and uncovering an interesting phenomenon: even a single-layer Mamba efficiently learns the in-context Laplacian smoothing estimator, which is both Bayes and minimax optimal. To explain this, we theoretically characterize the representation capacity of Mamba and reveal the fundamental role of convolution in enabling it to represent the optimal Laplacian smoothing. These theoretical insights align strongly with empirical results and, to the best of our knowledge, represent the first formal connection between Mamba and optimal statistical estimators. Finally, we outline promising research directions inspired by these findings.

Applications · Health

Hussain Alasmawi, Numan Saeed, Mohammad Yaqub

The growing demand for prenatal ultrasound imaging has intensified a global shortage of trained sonographers, creating barriers to essential fetal health monitoring. Deep learning has the potential to enhance sonographers' efficiency and support the training of new practitioners. Vision-Language Models (VLMs) are particularly promising for ultrasound interpretation, as they can jointly process images and text to perform multiple clinical tasks within a single framework. However, despite the expansion of VLMs, no standardized benchmark exists to evaluate their performance in fetal ultrasound imaging. This gap is primarily due to the modality’s challenging nature, operator dependency, and the limited public availability of datasets. To address this gap, we present Fetal-Gauge, the first and largest visual question answering benchmark specifically designed to evaluate VLMs across various fetal ultrasound tasks. Our benchmark comprises over 42,000 images and 93,000 question-answer pairs, spanning anatomical plane identification, visual grounding of anatomical structures, fetal orientation assessment, clinical view conformity, and clinical diagnosis. We systematically evaluate several state-of-the-art VLMs, including general-purpose and medical-specific models, and reveal a substantial performance gap: the best-performing model achieves only 55\% accuracy, far below clinical requirements. Our analysis identifies critical limitations of current VLMs in fetal ultrasound interpretation, highlighting the urgent need for domain-adapted architectures and specialized training approaches. Fetal-Gauge establishes a rigorous foundation for advancing multimodal deep learning in prenatal care and provides a pathway toward addressing global healthcare accessibility challenges. Our benchmark is publicly available at https://github.com/BioMedIA-MBZUAI/FETAL-GAUGE

Computer Vision · Vision Models & Multimodal

Xiaofeng Zhang, Yuanchao Zhu, Chaochen Gu, Xiaosong Yuan, Qiyan Zhao, Jiawei Cao, Barrett Tang, Sinan Fan, Yaomin Shen, Chen Shen 等

Recent studies have investigated attention dynamics in large vision language models (LVLMs), yet existing methods remain limited in reliably distinguishing hallucinated from correct outputs — primarily because they rely solely on forward-pass attention, ignoring gradient-based signals that reveal how token influence propagates through the model. To bridge this gap, we introduce \textbf{LVLMs-Saliency}, an \textit{gradient-aware diagnostic tool} that quantifies the grounding strength of each output token by fusing attention weights with their gradients. Through analysis, we identify a decisive pattern: \textit{Hallucinations occur when prior output tokens shows low saliency to the next token prediction}, indicating a failure of contextual memory. Building on this insight, we propose a dual-mechanism inference-time framework: (1) Saliency-Guided Rejection Sampling (SGRS), which dynamically filters candidate tokens during decoding by rejecting those with saliency below a context-adaptive threshold, thereby preventing coherence-breaking tokens from entering the sequence; and (2) Local Coherence Reinforcement (LocoRE), a lightweight plug-and-play module that strengthens attention from the current token to its most recent outputs, actively counteracting the “forgetting” behavior identified by LVLMs-Saliency. Experimental results demonstrate that our method significantly reduces hallucinations across multiple LVLMs, offering a robust and interpretable solution to improve model reliability.

Social Aspects · Accountability, Transparency and Interpretability

Thibaud Gloaguen, Robin Staab, Nikola Jovanović, Martin Vechev

Most LLM fingerprinting methods teach the model to respond to a few fixed queries with predefined atypical responses (keys). This memorization often does not survive common deployment steps such as finetuning or quantization, and such keys can be easily detected and filtered from LLM responses, ultimately breaking the fingerprint. To overcome these limitations we introduce *LLM fingerprinting via semantically conditioned watermarks*, replacing fixed query sets with a broad semantic domain, and replacing brittle atypical keys with a statistical watermarking signal diffused throughout each response. After teaching the model to watermark its responses only to prompts from a predetermined domain e.g., French language, the model owner can use queries from that domain to reliably detect the fingerprint and verify ownership. As we confirm in our thorough experimental evaluation, our fingerprint is both stealthy and robust to all common deployment scenarios.

Theory · Interpretability and Visualization

Jingzhe Shi, Qinwei (Martin) Ma, Hongyi Liu, Hang Zhao, Jenq-Neng Hwang, Lei Li

Long Context Language Models have drawn great attention in the past few years. There has been work discussing the impact of long context on Language Model performance: some find that long irrelevant context could harm performance, while some experimentally summarize loss reduction by relevant long context as Scaling Laws. This calls for a more thorough understanding of how long context impacts Language Modeling. In this work, we (1) propose to use `Intrinsic Entropy' for explaining the impact of context length on language modeling; and (2) conduct experiments on natural language and synthetic data, validating our proposed theoretical assumptions and deductions. Our theoretical framework can provide practical insights such as establishing that training dataset size dictates an optimal context length and bounds context length scaling for certain cases. We hope our work may inspire new long context Language Models, as well as future work studying the physics of Language Models.

General Machine Learning · Everything Else

Chen Feng, Minghe Shen, Ananth Balashankar, Carsten Gerner-Beuerle, Miguel Rodrigues

Reliable certification of Large Language Models (LLMs)—verifying that failure rates are below a safety threshold—is critical yet challenging. While "LLM-as-a-Judge" offers scalability, judge imperfections, noise, and bias can invalidate statistical guarantees. We introduce a "Noisy but Valid" hypothesis testing framework to address this. By leveraging a small human-labelled calibration set to estimate the judge's True Positive and False Positive Rates (TPR/FPR), we derive a variance-corrected critical threshold applied to a large judge-labelled dataset. Crucially, our framework theoretically guarantees finite-sample Type-I error control (validity) despite calibration uncertainty. This distinguishes our work from Prediction-Powered Inference (PPI), positioning our method as a diagnostic tool that explicitly models judge behavior rather than a black-box estimator. Our contributions include: (1) Theoretical Guarantees: We derive the exact conditions under which noisy testing yields higher statistical power than direct evaluation; (2) Empirical Validation: Experiments on Jigsaw Comment, Hate Speech and SafeRLHF confirm our theory; (3) The Oracle Gap: We reveal a significant performance gap between practical methods and the theoretical "Oracle" (perfectly known judge parameters), quantifying the cost of estimation. Specifically, we provide the first systematic treatment of the imperfect-judge setting, yielding interpretable diagnostics of judge reliability and clarifying how evaluation power depends on judge quality, dataset size, and certification levels. Together, these results sharpen understanding of statistical evaluation with LLM judges, and highlight trade-offs among competing inferential tools.

Applications · Genetics, Cell Biology, Health, etc

Jiarui Li, Zixiang Yin, Haley Smith, Zhengming Ding, Samuel Landry, Ramgopal Mettu

CD8+ “killer” T cells and CD4+ “helper” T cells play a central role in the adaptive immune system by recognizing antigens presented by Major Histocompatibility Complex (pMHC) molecules via T Cell Receptors (TCRs). Modeling binding between T cells and the pMHC complex is fundamental to understanding basic mechanisms of human immune response as well as in developing therapies. While transformer-based models such as TULIP have achieved impressive performance in this domain, their black-box nature precludes interpretability and thus limits a deeper mechanistic understanding of T cell response. Most existing post-hoc explainable AI (xAI) methods are confined to encoder-only, co-attention, or model-specific architectures and cannot handle encoder-decoder transformers used in TCR-pMHC modeling. To address this gap, we propose Quantifying Cross-Attention Interaction (QCAI), a new post-hoc method designed to interpret the cross-attention mechanisms in transformer decoders. Quantitative evaluation is a challenge for XAI methods; we have compiled TCR-XAI, a benchmark consisting of 274 experimentally determined TCR-pMHC structures to serve as ground truth for binding. Using these structures we compute physical distances between relevant amino acid residues in the TCR-pMHC interaction region and evaluate how well our method and others estimate the importance of residues in this region across the dataset. We show that QCAI achieves state-of-the-art performance on both interpretability and prediction accuracy under the TCR-XAI benchmark.

Social Aspects · Fairness, Equity, Justice and Safety

Abhipsa Basu, Mohana Singh, Shashank Agnihotri, Margret Keuper, Venkatesh Babu Radhakrishnan

Text-to-image (T2I) models are rapidly gaining popularity, yet their outputs often lack geographical diversity, reinforce stereotypes, and misrepresent regions. Given their broad reach, it is critical to rigorously evaluate how these models portray the world. Existing diversity metrics either rely on curated datasets or focus on surface-level visual similarity, limiting interpretability. We introduce *GeoDiv*, a framework leveraging large language and vision-language models to assess geographical diversity along two complementary axes: the Socio-Economic Visual Index (SEVI), capturing economic and condition-related cues, and the Visual Diversity Index (VDI), measuring variation in primary entities and backgrounds. Applied to images generated by models such as Stable Diffusion and FLUX.1-dev across $10$ entities and $16$ countries, *GeoDiv* reveals a consistent lack of diversity and identifies fine-grained attributes where models default to biased portrayals. Strikingly, depictions of countries like India, Nigeria, and Colombia are disproportionately impoverished and worn, reflecting underlying socio-economic biases. These results highlight the need for greater geographical nuance in generative models. *GeoDiv* provides the first systematic, interpretable framework for measuring such biases, marking a step toward fairer and more inclusive generative systems. Project page: https://abhipsabasu.github.io/geodiv

Applications · Health

Yilan Zhang, Li Nanbo, Changchun Yang, Jürgen Schmidhuber, Xin Gao

The integration of histology images and gene profiles has shown great promise for improving survival prediction in cancer. However, current approaches often struggle to model intra- and inter-modal interactions efficiently and effectively due to the high dimensionality and complexity of the inputs. A major challenge is capturing critical prognostic events that, though few, underlie the complexity of the observed inputs and largely determine patient outcomes. These events---manifested as high-level structural signals such as spatial histologic patterns or pathway co-activations---are typically sparse, patient-specific, and unannotated, making them inherently difficult to uncover. To address this, we propose SlotSPE, a slot-based framework for structural prognostic event modeling. Specifically, inspired by the principle of factorial coding, we compress each patient’s multimodal inputs into compact, modality-specific sets of mutually distinctive slots using slot attention. By leveraging these slot representations as encodings for prognostic events, our framework enables both efficient and effective modeling of complex intra- and inter-modal interactions, while also facilitating seamless incorporation of biological priors that enhance prognostic relevance. Extensive experiments on ten cancer benchmarks show that SlotSPE outperforms existing methods in 8 out of 10 cohorts, achieving an overall improvement of 2.9%. It remains robust under missing genomic data and delivers markedly improved interpretability through structured event decomposition.

General Machine Learning · Representation Learning

Lei Wang, Xi Ding, Yongsheng Gao, Piotr Koniusz

Learning from structured multi-way data, represented as higher-order tensors, requires capturing complex interactions across tensor modes while remaining computationally efficient. We introduce Uncertainty-driven Kernel Tensor Learning (UKTL), a novel kernel framework for $M$-mode tensors that compares mode-wise subspaces derived from tensor unfoldings, enabling expressive and robust similarity measures. To handle large-scale tensor data, we propose a scalable Nystr\"{o}m kernel linearization with dynamically learned pivot tensors obtained via soft $k$-means clustering. A key innovation of UKTL is its uncertainty-aware subspace weighting, which adaptively down-weights unreliable mode components based on estimated confidence, improving robustness and interpretability in comparisons between input and pivot tensors. Our framework is fully end-to-end trainable and naturally incorporates both multi-way and multi-mode interactions through structured kernel compositions. Extensive evaluations on action recognition benchmarks (NTU-60, NTU-120, Kinetics-Skeleton) show that UKTL achieves state-of-the-art performance, superior generalization, and meaningful mode-wise insights. This work establishes a principled, scalable, and interpretable kernel learning paradigm for structured multi-way and multi-modal tensor sequences.

Social Aspects · Accountability, Transparency and Interpretability

Hakaze Cho, Haolin Yang, Gouki Minegishi, Naoya Inoue

In-context Learning (ICL) is an emerging few-shot learning paradigm based on modern Language Models (LMs), yet its inner mechanism remains unclear. In this paper, we investigate the mechanism through a novel perspective of information removal. Specifically, we demonstrate that in the zero-shot scenario, LMs encode queries into non-selective representations in hidden states containing information for all possible tasks, leading to arbitrary outputs without focusing on the intended task, resulting in near-zero accuracy. Meanwhile, we find that selectively removing specific information from hidden states by a low-rank filter effectively steers LMs toward the intended task. Building on these findings, by measuring the hidden states on carefully designed metrics, we observe that few-shot ICL effectively simulates such task-oriented information removal processes, selectively removing the redundant information from entangled non-selective representations, and improving the output based on the demonstrations, which constitutes a key mechanism underlying ICL. Moreover, we identify essential attention heads inducing the removal operation, termed Denoising Heads, which enables the ablation experiments blocking the information removal operation from the inference, where the ICL accuracy significantly degrades, especially when the correct label is absent from the few-shot demonstrations, confirming both the critical role of the information removal mechanism and denoising heads.

Applications · Physics

Youngwoo Cho, Seunghoon Yi, Wooil Yang, Sungmo Kang, Young-Woo Son, Jaegul Choo, Joonseok Lee, Soo Kyung Kim, Hongkee Yoon

Pre-trained materials foundation models, or machine learning interatomic potentials, leverage general physicochemical knowledge to effectively approximate potential energy surfaces. However, they often require domain-specific calibration due to physicochemical diversity as well as mismatches between practical computational settings and those used in constructing the pre-training data. To address this, we propose a sparsity-promoting fine-tuning method that selectively updates model parameters by exploiting the structural properties of E(3)-equivariant materials foundation models. On energy and force prediction tasks across molecular and crystalline benchmarks, our method matches or surpasses full fine-tuning and equivariant low-rank adaptation while updating only ~3 \% of parameters, and in some cases as little as \~0.5 \%. Beyond energy and force calibration, we further demonstrate task generalizability by applying our method to magnetic moment prediction and magnetism-aware total energy modeling. Finally, analysis of sparsity patterns reveals physically interpretable signatures, such as enhanced $d$-orbital contributions in transition metal systems. Overall, our results establish sparsity-promoting fine-tuning as a flexible and interpretable method for domain specialization of equivariant materials foundation models.

Social Aspects · Accountability, Transparency and Interpretability

Shawn Im, Changdae Oh, Zhen Fang, Sharon Li

Semantic associations such as the link between "bird" and "flew" are foundational for language modeling as they enable models to go beyond memorization and instead generalize and generate coherent text. Understanding how these associations are learned and represented in language models is essential for connecting deep learning with linguistic theory and developing a mechanistic foundation for large language models. In this work, we analyze how these associations emerge from natural language data in attention-based language models through the lens of training dynamics. By leveraging a leading-term approximation of the gradients, we develop closed-form expressions for the weights at early stages of training that explain how semantic associations first take shape. Through our analysis, we reveal that each set of weights of the transformer has closed-form expressions as simple compositions of three basis functions--bigram, token-interchangeability, and context mappings--reflecting the statistics in the text corpus and uncover how each component of the transformer captures the semantic association based on these compositions. Experiments on real-world LLMs demonstrate that our theoretical weight characterizations closely match the learned weights, and qualitative analyses further guide us on how our theorem shines light on interpreting the learned association in transformers.

Computer Vision · Vision Models & Multimodal

Yilun Hao, Yongchao Chen, Chuchu Fan, Yang Zhang

Vision Language Models (VLMs) show strong potential for visual planning but struggle with precise spatial and long-horizon reasoning, while Planning Domain Definition Language (PDDL) planners excel at formal long-horizon planning but cannot interpret visual inputs. Recent works combine these complementary advantages by translating visual problems into PDDL. However, while VLMs can generate PDDL problem files satisfactorily, accurately generating PDDL domain files, which encode planning rules, remains challenging and typically requires human expertise or environment interaction. We propose VLMFP, a Dual-VLM-guided framework that autonomously generates both PDDL problem and domain files for formal visual planning. VLMFP combines a SimVLM that simulates action consequences with a GenVLM that generates and iteratively refines PDDL files by aligning symbolic execution with simulated outcomes, enabling multiple levels of generalization across unseen instances, visual appearances, and game rules. We evaluate VLMFP on 6 grid-world domains and demonstrate its generalization capability. On average, SimVLM achieves 87.3\% and 86.0\% scenario understanding and action simulation for seen and unseen appearances, respectively. With the guidance of SimVLM, VLMFP attains 70.0\%, 54.1\% planning success on unseen instances in seen and unseen appearances, respectively. We further demonstrate that VLMFP scales to complex long-horizon 3D planning tasks, including multi-robot collaboration and assembly scenarios with partial observability and diverse visual variations. Project page: https://sites.google.com/view/vlmfp.