论文检索

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.

Theory · Interpretability and Visualization

Yufeng Zhang, Xuepeng Wang, Lingxiang Wu, Jinqiao Wang

While chain-of-thought prompting enables powerful multi-step reasoning in Large Language Models (LLMs), the stochastic nature of the generation process undermines its reliability. In this work, we first analyze thousands of reasoning paths to identify Reasoning Drift, a key failure mode where models get locked into flawed reasoning patterns. We reveal that the manifestation of drift is a complex interplay between universal functional tendencies and unique, model-specific signatures. Based on the diagnosis, we propose Reasoning-Aware Activation Steering, a novel inference-time intervention method to gently nudge the model's activations away from pathological patterns. We pre-compute a library of vectors from contrastive functional transitions and apply them dynamically. Experiments show that our method effectively mitigates the drift problem and boosts accuracy. Additionally, it generalizes to out-of-distribution tasks, demonstrating a deeper capture of valid reasoning principles.

Theory · Interpretability and Visualization

Hongxu Liu, Jing Ma, Xiaojie Wang, Caixia YUAN, Fangxiang Feng

Figuring out how neural language models comprehend syntax acts as a key to revealing how they understand languages. We systematically analyzed methods for finding syntax structures in models, namely _probing_, and found limitations yet widely exist in previous probing practice. We proposed a method capable of estimating mutual information (MI) and extracting dependency trees from attention scores in a mathematical-rigorous way, requiring no additional network training effort. Compared with previous approaches, it has a much simpler model, while being able to probe more complex dependency trees, also transparent for fine-grained explanation. We tested our method on several open-source LLMs and demonstrated its effectiveness by systematically comparing it with a great many competitive baselines. Several informative conclusions can be drawn by further analysis of the results, shedding light on our method’s explanatory potential. Our code is released at https://github.com/ChristLBUPT/IPBP.

Social Aspects · Accountability, Transparency and Interpretability

Jingtong Su, Julia Kempe, Karen Ullrich

Transformers have achieved state-of-the-art performance across diverse language and vision tasks. This success drives the imperative to interpret their internal mechanisms with the dual goals of enhancing performance and improving behavioral control. Attribution methods help advance interpretability by assigning model outputs associated with a target concept to specific model components. Current attribution research primarily studies multi-layer perceptron (MLP) neurons and addresses relatively simple concepts such as factual associations (e.g., Paris is located in France). This focus tends to overlook the impact of the attention mechanism and lacks a unified approach for analyzing more complex concepts. To fill these gaps, we introduce Scalable Attention Module Discovery (SAMD), a concept-agnostic method for mapping arbitrary, complex concepts to specific attention heads of general transformer models. We accomplish this by representing each concept as a vector, calculating its cosine similarity with each attention head, and selecting the TopK-scoring heads to construct the concept-associated attention module. We then propose Scalar Attention Module Intervention (SAMI), a simple strategy to diminish or amplify the effects of a concept by adjusting the attention module using only a single scalar parameter. Empirically, we demonstrate SAMD on concepts of varying complexity, and visualize the locations of their corresponding modules. Our results demonstrate that module locations remain stable before and after LLM post-training, and confirm prior work on the mechanics of LLM multi-lingualism. Through SAMI, we facilitate jailbreaking on HarmBench (+72.7%) by diminishing “safety” and improve performance on the GSM8K benchmark (+1.6%) by amplifying “reasoning”. Lastly, we highlight the domain-agnostic nature of our approach by suppressing the image classification accuracy of vision transformers on ImageNet.

Theory · Interpretability and Visualization

Giordano Ramos-Traslosheros, Carlos Ponce

Signed connectivity is fundamental to neural computation in both brains (excitatory/inhibitory) and machines (positive/negative). Yet the role of signed weights in shaping visual representations in object recognition remains unclear. Dale's Law, the biological principle that neurons send exclusively excitatory or inhibitory outputs, is typically not enforced in artificial neural networks (ANNs). Here, we find that accuracy in ImageNet-trained ANNs correlates with the spontaneous emergence of sign-specific "Dale-like" segregation in their output layers. Ablation and feature visualization reveal a functional segregation in ANNs: removing positive inputs primarily disrupts localized, object-related structure, while removing negative inputs alters mainly dispersed background textures. This segregation is more pronounced in adversarially robust models, persists with unsupervised learning, and vanishes with non-rectified activation functions. We validate these observations in the macaque ventral visual cortex (V1, V4, and IT) using encoding models and in vivo feature visualization. The features recovered by encoding models qualitatively matched those identified in vivo. Model representations changed more upon positive than negative input ablations. We analyzed the most Dale-like units across neuron models, positive units showed localized features, while negative units showed larger, more dispersed features. Consistent with this, experimentally clearing the background around a neuron's preferred feature enhanced its response, likely by reducing inhibitory drive. Our results suggest that both artificial and biological vision systems segregate features by weight sign: positive weights emphasize object-related features, while negative weights refine context. This highlights a convergent representational strategy in brains and machines, yielding predictions for visual neuroscience.

Computer Vision · Image and Video Generation

Cong Wei, Quande Liu, Zixuan Ye, Qiulin Wang, Xintao WANG, Pengfei Wan, Kun Gai, Wenhu Chen

Unified multimodal models have shown promising results in multimodal content generation and editing, but remain largely limited to the image domain. In this work, we present UniVideo, a versatile framework that extends unified modelling to the video domain. UniVideo adopts a dual-stream design, combining a Multimodal Large Language Model (MLLM) for instruction understanding with a Multimodal DiT (MMDiT) for video generation. This design preserves the MLLM's original text generation capabilities, enables accurate interpretation of complex multimodal instructions, and maintains visual consistency in the generated content. Built on this architecture, UniVideo unifies diverse video generation and editing tasks under a single multimodal instruction paradigm and is jointly trained across them. Extensive experiments demonstrate that UniVideo matches or surpasses state-of-the-art task-specific baselines in text/image-to-video generation, in-context video generation and in-context video editing. Notably, the unified design of UniVideo enables two forms of generalization. First, UniVideo supports task composition, such as combining editing with style transfer, by integrating multiple capabilities within a single instruction. Second, even without explicit training on free-form video editing, UniVideo transfers its editing capability from large-scale image editing data to this setting, handling unseen instructions such as changing the environment or altering materials within a video. Beyond these core capabilities, UniVideo also supports generation with thinking, where the MLLM interprets complex prompts and guides the MMDiT during synthesis. To foster future research, we released our model and code.

Theory · Interpretability and Visualization

Mengyang Li, Pinlong Zhao, Zhong Zhang

The alignment of Large Language Models (LLMs) with human preferences is critically undermined by noisy labels in training datasets. Existing robust methods often prove insufficient, as they rely on single, narrow heuristics such as perplexity or loss, failing to address the diverse nature of real-world noise. We challenge this limited-scope approach by introducing a new paradigm where models learn to diagnose thyself, systematically fusing multiple streams of intrinsic feedback for a holistic reliability assessment of each preference pair. We instantiate this paradigm through a meta-learning methodology that learns to adaptively reweight samples based on a rich diagnostic vector. This vector captures three complementary perspectives: preference consistency, learning difficulty, and generation confidence. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods across various noise conditions. Crucially, our work provides the first quantitative analysis of these intrinsic diagnostics, revealing that their fusion is essential for overcoming the blind spots inherent in any single heuristic. This diagnostic-driven paradigm offers a principled path towards developing more robust and trustworthy LLMs.

Social Aspects · Everything Else

Xin Shen, Yan Ke, Xinyu Wang, Xin Yu

Fingerspelling plays a vital role in sign languages, particularly for conveying names, technical terms, and words not found in the standard lexicon. However, evaluation of two-handed fingerspelling detection and recognition is rarely addressed in existing sign language datasets—particularly for BANZSL (British, Australian, and New Zealand Sign Language), which share a common two-handed manual alphabet. To bridge this gap, we curate a large-scale dataset, dubbed BANZ-FS, focused on BANZSL fingerspelling in both controlled and real-world environments. Our dataset is compiled from three distinct sources: (1) live sign language interpretation in news broadcasts, (2) controlled laboratory recordings, and (3) diary vlogs from online platforms and social media. This composition enables BANZ-FS to capture variations in signing tempos and fluency across diverse signers and contents. Each instance in BANZ-FS is carefully annotated with multi-level alignment: video ↔ subtitles, video ↔ fingerspelled letters, and video ↔ target lexicons. In total, BANZ-FS includes over 35,000 video-aligned fingerspelling instances. Importantly, BANZ-FS highlights the unique linguistic and visual challenges posed by two-handed fingerspelling, including handshape coarticulation, self-occlusion, intra-letter variation, and rapid inter-letter transitions. We benchmark state-of-the-art models on the key tasks, including fingerspelling detection, isolated fingerspelling recognition, and fingerspelling recognition in context. Experimental results show that BANZ-FS presents substantial challenges while offering rich opportunities for BANZSL understanding and broader sign language technology. The dataset and benchmarks are available at BANZ-FS.

Theory · Interpretability and Visualization

Michael Freenor, Lauren Alvarez

Understanding how language and embedding models encode semantic relationships is fundamental to model interpretability. While early word embeddings exhibited intuitive vector arithmetic (''king'' - ''man'' + ''woman'' = ''queen''), modern high-dimensional text representations lack straightforward interpretable geometric properties. We introduce Rotor-Invariant Shift Estimation (RISE), a geometric approach that represents semantic-syntactic transformations as consistent rotational operations in embedding space, leveraging the manifold structure of modern language representations. RISE operations have the ability to operate across both languages and models without reducing performance, suggesting the existence of analogous cross-lingual geometric structure. We compare and evaluate RISE using two baseline methods, three embedding models, three datasets, and seven morphologically diverse languages in five major language groups. Our results demonstrate that RISE consistently maps discourse-level semantic-syntactic transformations with distinct grammatical features (e.g., negation and conditionality) across languages and models. This work provides the first demonstration that discourse-level semantic-syntactic transformations correspond to consistent geometric operations in multilingual embedding spaces, empirically supporting the linear representation hypothesis at the sentence level.

General Machine Learning · Representation Learning

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.

Computer Vision · Image and Video Generation

Kaiwen Shi, Sichen Liu, Ziyue Lin, Hangrui Guo, Gong Cheng

Flowcharts are widely used to represent processes and relationships through intuitive visual representations. However, accurately interpreting these diagrams remains challenging due to their structural complexity and high visual diversity. Existing flowchart datasets often lack fine-grained control over key properties such as graph complexity and rendering style, limiting their utility for training and testing of multimodal large language models (MLLMs) on visual reasoning tasks. To address these limitations, we introduce FlowGen, a controllable synthesizer that generates flowcharts that have customizable structural features and supports multiple renderer backends. FlowGen enables fine-grained control over graph properties such as graph order and size, branched arrows, and nested subgraphs, facilitating systematic evaluation of MLLMs' capabilities. Extensive experiments on open-source and proprietary MLLMs show that training on FlowGen substantially improves flowchart parsing and question answering (QA), while also enhancing generalization to other public datasets. Furthermore, FlowGen provides challenging test datasets that expose consistent weaknesses in current MLLMs, particularly related to high structural complexity and varied rendering styles. Our code and data are publicly available at https://github.com/nju-websoft/FlowGen.

Computer Vision · Classification and Understanding

Tsung-Ming Tai, Sofia Casarin, Andrea Pilzer, Werner Nutt, Oswald Lanz

Anticipating future actions in videos is challenging, as the observed frames provide only evidence of past activities, requiring the inference of latent intentions to predict upcoming actions. Existing transformer-based approaches, which rely on dot-product attention over pixel representations, often lack the high-level semantics necessary to model video sequences for effective action anticipation. As a result, these methods tend to overfit to explicit visual cues present in the past frames, limiting their ability to capture underlying intentions and degrading generalization to unseen samples. To address this, we propose Action-Guided Attention (AGA), an attention mechanism that explicitly leverages predicted action sequences as queries and keys to guide sequence modeling. Our approach fosters the attention module to emphasize relevant moments from the past based on the upcoming activity and combine this information with the current frame embedding via a dedicated gating function. The design of AGA enables post-training analysis of the knowledge discovered from the training set. Experiments on the widely adopted EPIC-Kitchens-100 benchmark demonstrate that AGA generalizes well from validation to unseen test sets. Post-training analysis can further examine the action dependencies captured by the model and the counterfactual evidence it has internalized, offering transparent and interpretable insights into its anticipative predictions.

Theory · Interpretability and Visualization

Kavi Gupta, Osbert Bastani, Armando Solar-Lezama

Real-world processes often contain intermediate state that can be modeled as an extremely sparse activation tensor. In this work, we analyze the identifiability of such sparse and local latent intermediate variables, which we call motifs. We prove our Motif Identifiability Theorem, stating that under certain assumptions it is possible to precisely identify these motifs exclusively by reducing end-to-end error. Notably, we do not assume identifiability of parameters, but rather of a latent intermediate representation output by a local model, thus allowing these representations to be arbitrarily complex functions of the input. Additionally, we provide the Sparling algorithm, which uses a new kind of informational bottleneck that enforces levels of activation sparsity unachievable using other techniques. We confirm empirically that extreme sparsity is necessary to achieve good intermediate state modeling. On synthetic domains, we are able to precisely localize the intermediate states up to feature permutation with $>90\%$ accuracy, even though we only train end-to-end.

Theory · Interpretability and Visualization

Muhammed Ustaomeroglu, Baris Askin, Gauri Joshi, Carlee Joe-Wong, Guannan Qu

The extent to which decoder-only language models (LMs) engage in planning, that is, organizing intermediate computations to support coherent long-range generation, remains an important question, with implications for interpretability, reliability, and principled model design. Planning involves structuring computations over long horizons, and considering multiple possible continuations, but how far transformer-based LMs exhibit them without external scaffolds, e.g., chain-of-thought prompting, is unclear. We address these questions by analyzing the hidden states at the core of transformer computations, which capture intermediate results and act as carriers of information. Since these hidden representations are redundant and encumbered with fine-grained details, we develop a pipeline based on vector-quantized variational autoencoders that compresses them into compact summary codes. These codes enable measuring mutual information and analyzing the computational structure of the underlying model behavior. Using this framework, we study planning in LMs across synthetic grammar, path-finding tasks, and natural language datasets, focusing on two planning properties: (i) the planning horizon of pre-output computations, and (ii) the extent to which the model considers alternative valid continuations. As a separate downstream use of the same pipeline, we also analyze how decision-relevant information is distributed across layers and earlier prefix blocks when producing next-token predictions. Together, these analyses advance our understanding of planning in LMs and provide a general-purpose pipeline for inspecting internal model dynamics. Our results reveal that the effective planning horizon is task-dependent, that models implicitly preserve information about unused correct continuations, and that predictions draw most on recent computations, though earlier blocks remain informative.

Computer Vision · Vision Models & Multimodal

Hanoona Rasheed, Abdelrahman Shaker, Anqi Tang, Muhammad Maaz, Ming-Hsuan Yang, Salman Khan, Fahad Khan

Mathematical reasoning in real-world video presents a fundamentally different challenge than static images or text. It requires interpreting fine-grained visual information, accurately reading handwritten or digital text, and integrating spoken cues, often dispersed non-linearly over time. In such multimodal contexts, success hinges not just on perception, but on selectively identifying and integrating the right details from a rich and noisy stream of content. To this end, we introduce VideoMathQA, a benchmark designed to evaluate whether models can perform such temporally extended cross-modal reasoning on videos. The benchmark spans 10 diverse mathematical domains, covering videos from 10 seconds to over 1 hour. We employ graduate-level experts to ensure high quality, for over 920 man-hours of annotation. To reflect real-world scenarios, questions are designed around three core reasoning challenges: direct problem solving, conceptual transfer, which requires applying learned methods to new problems; and deep instructional comprehension, involving multi-step reasoning over extended explanations and partially worked-out solutions. Each question includes multi-step reasoning annotations, enabling fine-grained diagnosis of model capabilities. Through this benchmark, we establish an evaluation framework for models that must reason, rather than merely perceive, jointly ground concepts across visual, audio, and textual modalities, across temporally extended mathematical problem settings.

Reinforcement Learning · Deep RL

Sungbin Mun, Younghwan Lee, Cheolhui MIn, Mineui Hong, Young Min Kim

Vision-based reinforcement learning can benefit from object-centric scene representation, which factorizes the visual observation into individual objects and their attributes, such as color, shape, size, and position. While such object-centric representations can extract components that generalize well for various multi-object manipulation tasks, they are prone to issues with occlusions and 3D ambiguity of object properties due to their reliance on single-view 2D image features. Furthermore, the entanglement between object configurations and camera poses complicates the object-centric disentanglement in 3D, leading to poor 3D reasoning by the agent in vision-based reinforcement learning applications. To address the lack of 3D awareness and the object-camera entanglement problem, we propose an enhanced 3D object-centric representation that utilizes multi-view 3D features and enforces more explicit 3D-aware disentanglement. The enhancement is based on the integration of the recent success of multi-view Transformer and the prototypical representation learning among the object-centric representations. The representation, therefore, can stably identify proxies of 3D positions of individual objects along with their semantic and physical properties, exhibiting excellent interpretability and controllability. Then, our proposed block transformer policy effectively performs novel tasks by assembling desired properties adaptive to the new goal states, even when provided with unseen viewpoints at test time. We demonstrate that our 3D-aware block representation is scalable to compose diverse novel scenes and enjoys superior performance in out-of-distribution tasks with multi-object manipulations under both seen and unseen viewpoints compared to existing methods.

Applications · Neuroscience, Cognitive Science

Franz Rieger, Jan-Matthis Lueckmann, Viren Jain, Michal Januszewski

Biological neurons come in many shapes. High-fidelity generative modeling of their varied morphologies is challenging yet underexplored in neuroscience, and crucial for the subfield of connectomics. We introduce MoGen (Neuronal Morphology Generation), a flow matching model to generate high-resolution 3D point clouds of mouse cortex axon and dendrite fragments. This is enabled by an adaptation that injects local geometric context into a scalable latent transformer backbone, allowing for the generation of high-fidelity, realistic samples. To assess MoGen's generation quality, we propose a dedicated evaluation suite with interpretable geometric and topological features tailored to neuronal structures that we validate in a user study. MoGen's practical utility is showcased through controllable generation for visualization via smooth interpolation and a direct downstream application: we augment the training set of a shape plausibility classifier from a production connectomics neuron reconstruction pipeline with millions of generated samples, thereby improving classifier accuracy and reducing the number of remaining split and merge errors by 4.4%. We estimate this can reduce manual proofreading labor by over 157 person-years for reconstruction of a full mouse brain.

Theory · Interpretability and Visualization

Karim Galliamov, Syed Muhammad Ahsan Raza Kazmi, Adil Khan, Adín Ramírez Rivera

Concept Bottleneck Models (CBMs) aim to deliver interpretable predictions by routing decisions through a human-understandable concept layer, yet they often suffer reduced accuracy and concept leakage that undermines faithfulness. We introduce an explicit Information Bottleneck regularizer on the concept layer that penalizes $I(X;C)$ while preserving task-relevant information in $I(C;Y)$, encouraging minimal-sufficient concept representations. We derive two practical variants (a variational objective and an entropy-based surrogate) and integrate them into standard CBM training without architectural changes or additional supervision. Evaluated across six CBM families and three benchmarks, the IB-regularized models consistently outperform their vanilla counterparts. Information-plane analyses further corroborate the intended behavior. These results indicate that enforcing a minimal-sufficient concept bottleneck improves both predictive performance and the reliability of concept-level interventions. The proposed regularizer offers a theoretic-grounded, architecture-agnostic path to more faithful and intervenable CBMs, resolving prior evaluation inconsistencies by aligning training protocols and demonstrating robust gains across model families and datasets.

Theory · Interpretability and Visualization

Zhihao Zhang, Qiaole Dong, Qi Zhang, Enyu Zhou, Jun Zhao, Zhiheng Xi, Senjie Jin, Xiaoran Fan, Yuhao Zhou, Mingqi Wu 等

Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt (multimodal) large language models to downstream tasks. While effective at task adaptation, their impact on retaining prior knowledge remains unclear. In this paper, we introduce jigsaw puzzles as a novel task absent from existing pretraining corpora and systematically study the behavior of SFT and RFT on the open-source Qwen2.5-VL series. Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly but better maintains prior knowledge. We study this phenomenon through learning dynamics by examining both the magnitude and direction of how training data influence prior knowledge. Our analysis shows that RFT mainly reinforces correct samples naturally aligned with the base model’s probability landscape, leading to weaker interference with prior knowledge. Moreover, training on RFT-simulated rollouts, which exert a smaller magnitude of influence and are better aligned in direction to prior knowledge, allows SFT to preserve prior knowledge better while rapidly learning new tasks. We further validate our framework on Qwen2.5 post-training in math and scientific QA, observing consistent forgetting and learning-dynamics trends. These findings suggest that the distribution of post-training data, rather than algorithmic differences alone, plays a central role in forgetting, and highlight RFT as a promising ingredient for stable continual post-training.

Theory · Interpretability and Visualization

Alex Ning, Vainateya Rangaraju, Yen-Ling Kuo

In this blog post, we extract, process, and visualize latent state geometries in Transformer-based language models through dimensionality reduction to build a better intuition of their internal dynamics. We demonstrate experiments with GPT-2 and LLaMa models, uncovering interesting geometric patterns in their latent spaces. Notably, we identify a clear separation between attention and MLP component outputs across intermediate layers, a pattern not documented in prior work to our knowledge.