Videos are unique in their integration of temporal elements, including camera, scene, action, and attribute, along with their dynamic relationships over time. However, existing benchmarks for video understanding often treat these properties separately or narrowly focus on specific aspects, overlooking the holistic nature of video content. To address this, we introduce TUNA, a temporal-oriented benchmark for fine-grained understanding on dense dynamic videos, with two complementary tasks: captioning and QA. Our TUNA features diverse video scenarios and dynamics, assisted by interpretable and robust evaluation criteria. We evaluate several leading models on our benchmark, providing fine-grained performance assessments across various dimensions. This evaluation reveals key challenges in video temporal understanding, such as limited action description, inadequate multi-subject understanding, and insensitivity to camera motion, offering valuable insights for improving video understanding models.
论文检索
输入标题、作者或关键词，从 7,537 篇学术成果中精准定位
Normalized AOPC: Fixing Misleading Faithfulness Metrics for Feature Attributions Explainability
PDF ↗Deep neural network predictions are notoriously difficult to interpret. Feature attribution methods aim to explain these predictions by identifying the contribution of each input feature. Faithfulness, often evaluated using the area over the perturbation curve (AOPC), reflects feature attributions’ accuracy in describing the internal mechanisms of deep neural networks. However, many studies rely on AOPC to compare faithfulness across different models, which we show can lead to false conclusions about models’ faithfulness. Specifically, we find that AOPC is sensitive to variations in the model, resulting in unreliable cross-model comparisons. Moreover, AOPC scores are difficult to interpret in isolation without knowing the model-specific lower and upper limits. To address these issues, we propose a normalization approach, Normalized AOPC (NAOPC), enabling consistent cross-model evaluations and more meaningful interpretation of individual scores. Our experiments demonstrate that this normalization can radically change AOPC results, questioning the conclusions of earlier studies and offering a more robust framework for assessing feature attribution faithfulness. Our code is available at https://github.com/JoakimEdin/naopc.
Large Language Models (LLMs) have demonstrated strong performance in handling complex tasks that require both extensive knowledge and reasoning abilities. However, the existing LLM inference pipeline operates as an opaque process without explicit separation between knowledge retrieval and reasoning steps, making the model’s decision-making process unclear and disorganized. Recent research has shown that this ambiguity will lead to issues such as knowledge forgetting, which significantly impact the reliability of LLMs. In this paper, we propose a novel language model inference paradigm that decomposes the complex inference process into two distinct and clear actions: (1) memory recall: which retrieves relevant knowledge in LLM, and (2) reasoning: which performs reasoning steps based on the recalled knowledge. To facilitate this decomposition, we introduce two special tokens memory and reason, guiding the model to distinguish between steps that require knowledge retrieval and those that involve reasoning. Our experiment results show that this decomposition not only improves LLMs’ performance among utility benchmarks but also enhances interpretability during the inference process, enabling users to identify sources of error and refine model responses effectively. The code is available at: https://github.com/MingyuJ666/Disentangling-Memory-and-Reasoning.
Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes often relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers leads to poor generalization on reasoning tasks that require more detailed explanations. To address this limitation, we propose a two-stage post-training strategy that extends the usage of short answer data for enhanced CoT reasoning. First, we augment short answers with CoT reasoning generated by GPT-4o, enhancing the VLM’s CoT capabilities through fine-tuning. Second, we leverage short answers as outcome rewards for reinforcement learning. Specifically, short answers are used as correctness indicators to construct positive (correct) and negative (incorrect) pairs from model-generated reasoning chains. These pairs are then used to calibrate the model’s reasoning via Direct Preference Optimization. Our experiments show significant improvements in CoT reasoning on benchmark datasets, along with enhanced generalization to direct answer prediction. This work provides a critical data resource for VLM CoT training and demonstrates the effectiveness of outcome rewards for multimodal models post-training.
Topic modeling is a fundamental task in natural language processing, allowing the discovery of latent thematic structures in text corpora. While Large Language Models (LLMs) have demonstrated promising capabilities in topic discovery, their direct application to topic modeling suffers from issues such as incomplete topic coverage, misalignment of topics, and inefficiency. To address these limitations, we propose LLM-ITL, a novel LLM-in-the-loop framework that integrates LLMs with Neural Topic Models (NTMs). In LLM-ITL, global topics and document representations are learned through the NTM. Meanwhile, an LLM refines these topics using an Optimal Transport (OT)-based alignment objective, where the refinement is dynamically adjusted based on the LLM’s confidence in suggesting topical words for each set of input words. With the flexibility of being integrated into many existing NTMs, the proposed approach enhances the interpretability of topics while preserving the efficiency of NTMs in learning topics and document representations. Extensive experiments demonstrate that LLM-ITL helps NTMs significantly improve their topic interpretability while maintaining the quality of document representation. Our code and datasets are available athttps://github.com/Xiaohao-Yang/LLM-ITL
Composed Image Retrieval (CIR) enables users to search for images using multimodal queries that combine text and reference images. While metric learning methods have shown promise, they rely on deterministic point embeddings that fail to capture the inherent uncertainty in the input data, in which user intentions may be imprecisely specified or open to multiple interpretations. We address this challenge by reformulating CIR through our proposed Composed Probabilistic Embedding (CoPE) framework, which represents both queries and targets as Gaussian distributions in latent space rather than fixed points. Through careful design of probabilistic distance metrics and hierarchical learning objectives, CoPE explicitly captures uncertainty at both instance and feature levels, enabling more flexible, nuanced, and robust matching that can handle polysemy and ambiguity in search intentions. Extensive experiments across multiple benchmarks demonstrate that CoPE effectively quantifies both quality and semantic uncertainties within Composed Image Retrieval, achieving state-of-the-art performance on recall rate. Code: https://github.com/tanghme0w/ACL25-CoPE.
Given a semi-structured knowledge base (SKB), where text documents are interconnected by relations, how can we effectively retrieve relevant information to answer user questions?Retrieval-Augmented Generation (RAG) retrieves documents to assist large language models (LLMs) in question answering; while Graph RAG (GRAG) uses structured knowledge bases as its knowledge source.However, many questions require both textual and relational information from SKB — referred to as “hybrid” questions — which complicates the retrieval process and underscores the need for a hybrid retrieval method that leverages both information.In this paper, through our empirical analysis, we identify key insights that show why existing methods may struggle with hybrid question answering (HQA) over SKB. Based on these insights, we propose HybGRAG for HQA, consisting of a retriever bank and a critic module, with the following advantages:1. Agentic, it automatically refines the output by incorporating feedback from the critic module, 2. Adaptive, it solves hybrid questions requiring both textual and relational information with the retriever bank,3. Interpretable, it justifies decision making with intuitive refinement path, and4. Effective, it surpasses all baselines on HQA benchmarks.In experiments on the STaRK benchmark, HybGRAG achieves significant performance gains, with an average relative improvement in Hit@1 of 51%.
The effective utilization of structured data, integral to corporate data strategies, has been challenged by the rise of large language models (LLMs) capable of processing unstructured information. This shift prompts the question: can LLMs interpret structured data directly in its unstructured form? We propose an automatic evaluation data generation method for assessing LLMs’ reasoning capabilities on structure-rich text to explore this. Our approach supports 8 structured languages and 29 tasks, generating data with adjustable complexity through controllable nesting and structural width. We introduce StrucText-Eval, a benchmark containing 5,800 pre-generated and annotated samples designed to evaluate how well LLMs understand and reason through structured text. StrucText-Eval is divided into two suites: a regular Test suite (3,712 samples) and a Test-Hard suite (2,088 samples), the latter emphasizing the gap between human and model performance on more complex tasks. Experimental results show that while open-source LLMs achieve a maximum accuracy of 74.9% on the standard dataset, their performance drops significantly to 45.8% on the harder dataset. In contrast, human participants reach an accuracy of 92.6% on StrucText-Eval-Hard, highlighting LLMs’ current limitations in handling intricate structural information.
CLEME2.0: Towards Interpretable Evaluation by Disentangling Edits for Grammatical Error Correction
PDF ↗The paper focuses on the interpretability of Grammatical Error Correction (GEC) evaluation metrics, which received little attention in previous studies. To bridge the gap, we introduce **CLEME2.0**, a reference-based metric describing four fundamental aspects of GEC systems: hit-correction, wrong-correction, under-correction, and over-correction. They collectively contribute to exposing critical qualities and locating drawbacks of GEC systems. Evaluating systems by combining these aspects also leads to superior human consistency over other reference-based and reference-less metrics. Extensive experiments on two human judgment datasets and six reference datasets demonstrate the effectiveness and robustness of our method, achieving a new state-of-the-art result. Our codes are released at https://github.com/THUKElab/CLEME.
Supervised Fine-Tuning (SFT) and Preference Optimization (PO) are key processes for aligning Language Models (LMs) with human preferences post pre-training. While SFT excels in efficiency and PO in effectiveness, they are often combined sequentially without integrating their optimization objectives. This approach ignores the opportunities to bridge their paradigm gap and take the strengths from both. In this paper, we interpret SFT and PO with two sub-processes — *Preference Estimation* and *Transition Optimization* — defined at token level within the Markov Decision Process (MDP). This modeling shows that SFT is only a special case of PO with inferior estimation and optimization. PO estimates the model’s preference by its entire generation, while SFT only scores model’s subsequent predicted tokens based on prior tokens from ground truth answer. These priors deviates from model’s distribution, hindering the preference estimation and transition optimization. Building on this view, we introduce **Intuitive Fine-Tuning (IFT)** to integrate SFT and PO into a single process. Through a temporal residual connection, IFT brings better estimation and optimization by capturing LMs’ intuitive sense of its entire answers. But it solely relies on a single policy and the same volume of non-preference-labeled data as SFT. Our experiments show that IFT performs comparably or even superiorly to SFT and some typical PO methods across several tasks, particularly those requires generation, reasoning, and fact-following abilities. An explainable Frozen Lake game further validates the effectiveness of IFT for getting competitive policy.
Social Aspects · Accountability, Transparency, and Interpretability
Transformers excel at discovering patterns in sequential data, yet their fundamental limitations and learning mechanisms remain crucial topics of investigation. In this paper, we study the ability of Transformers to learn pseudo-random number sequences from linear congruential generators (LCGs), defined by the recurrence relation $x_{t+1} = a x_t + c \\;\mathrm{mod}\\; m$. We find that with sufficient architectural capacity and training data variety, Transformers can perform in-context prediction of LCG sequences with unseen moduli ($m$) and parameters ($a,c$). By analyzing the embedding layers and attention patterns, we uncover how Transformers develop algorithmic structures to learn these sequences in two scenarios of increasing complexity. First, we investigate how Transformers learn LCG sequences with unseen ($a, c$) but fixed modulus; and demonstrate successful learning up to $m = 2^{32}$. We find that models learn to factorize $m$ and utilize digit-wise number representations to make sequential predictions. In the second, more challenging scenario of unseen moduli, we show that Transformers can generalize to unseen moduli up to $m_{\text{test}} = 2^{16}$. In this case, the model employs a two-step strategy: first estimating the unknown modulus from the context, then utilizing prime factorizations to generate predictions. For this task, we observe a sharp transition in the accuracy at a critical depth $d= 3$. We also find that the number of in-context sequence elements needed to reach high accuracy scales sublinearly with the modulus.
Social Aspects · Accountability, Transparency, and Interpretability
Fine-grained steering of language model outputs is essential for safety and reliability. Prompting and finetuning are widely used to achieve these goals, but interpretability researchers have proposed a variety of representation-based techniques as well, including sparse autoencoders (SAEs), linear artificial tomography, supervised steering vectors, linear probes, and representation finetuning. At present, there is no benchmark for making direct comparisons between these proposals. Therefore, we introduce AxBench, a large-scale benchmark for steering and concept detection, and report experiments on Gemma-2-2B and 9B. For steering, we find that prompting outperforms all existing methods, followed by finetuning. For concept detection, representation-based methods such as difference-in-means, perform the best. On both evaluations, SAEs are not competitive. We introduce a novel weakly-supervised representational method (Rank-1 Representation Finetuning; ReFT-r1), which is competitive on both tasks while providing the interpretability advantages that prompting lacks. Along with AxBench, we train and publicly release SAE-scale feature dictionaries for ReFT-r1 and DiffMean.
Position · Understanding
In the rapidly growing literature on explanation algorithms, it often remains unclear what precisely these algorithms are for and how they should be used. In this position paper, we argue for a novel and pragmatic perspective: Explainable machine learning needs to recognize its parallels with applied statistics. Concretely, explanations are statistics of high-dimensional functions, and we should think about them analogously to traditional statistical quantities. Among others, this implies that we must think carefully about the matter of interpretation, or how the explanations relate to intuitive questions that humans have about the world. The fact that this is scarcely being discussed in research papers is one of the main drawbacks of the current literature. Moving forward, the analogy between explainable machine learning and applied statistics provides a fruitful way for how research practices can be improved.
Position · Social_ethical_env_impact
AI systems are now ubiquitous in real-world decision-making. However, their use is often invisible and almost always difficult to understand for the ordinary people who now come into contact with AI regularly. As these AI-driven decision-making systems increasingly replace human counterparts, our ability to understand the reasons behind a decision, and to contest that decision fairly, is quickly being eroded. In the United States legal system due process includes the right to understand the reasons for certain major decisions and the right to openly contest those decisions. Everyone is entitled to due process under the law, and human decision-makers have been required to adhere to due process when making many important decisions that are now slowly being relegated to AI systems. Using two recent court decisions as a foundation, this paper takes the position that AI in its current form cannot guarantee due process, and therefore cannot and (should not) be used to make decisions that should be subject to due process. The supporting legal analysis investigates how the current lack of technical answers about the interpretability and causality of AI decisions, coupled with extreme trade secret protections severely limiting any exercise of the small amount of technical knowledge we do have, serve as a fatal anti-due-process combination. Throughout the analysis, this paper explains why technical researchers' involvement is vital to informing the legal process and restoring due process protections.
Position · Methodology
Benchmarks and evaluations are central to machine learning methodology and direct research in the field. Current evaluations commonly test systems in the absence of humans. This position paper argues that the machine learning community should increasingly use _centaur evaluations_, in which humans and AI jointly solve tasks. Centaur Evaluations refocus machine learning development toward human augmentation instead of human replacement, they allow for direct evaluation of human-centered desiderata, such as interpretability and helpfulness, and they can be more challenging and realistic than existing evaluations. By shifting the focus from _automation_ toward _collaboration_ between humans and AI, centaur evaluations can drive progress toward more effective and human-augmenting machine learning systems.
Position · Social_ethical_env_impact
Patients have long sought health information online, and increasingly, they are turning to generative AI to answer their health-related queries. Given the high stakes of the medical domain, techniques like retrieval-augmented generation and citation grounding have been widely promoted as methods to reduce hallucinations and improve the accuracy of AI-generated responses and have been widely adopted into search engines. However, we argue that even when these methods produce literally accurate content drawn from source documents sans hallucinations, they can still be highly misleading. Patients may derive significantly different interpretations from AI-generated outputs than they would from reading the original source material, let alone consulting a knowledgeable clinician. Through a large-scale query analysis on topics including disputed diagnoses and procedure safety, we support our argument with quantitative and qualitative evidence of the suboptimal answers resulting from current systems. In particular, we highlight how these models tend to decontextualize facts, omit critical relevant sources, and reinforce patient misconceptions or biases. We propose a series of recommendations---such as the incorporation of communication pragmatics and enhanced comprehension of source documents---that could help mitigate these issues and extend beyond the medical domain.
Applications · Computer Vision
Video Question-Answering (VideoQA) remains challenging in achieving advanced cognitive reasoning due to the uncontrollable and opaque reasoning processes in existing Multimodal Large Language Models (MLLMs). To address this issue, we propose a novel Language-centric Tree Reasoning (LTR) framework that targets on enhancing the reasoning ability of models. In detail, it recursively divides the original question into logically manageable parts and conquers them piece by piece, enhancing the reasoning capabilities and interpretability of existing MLLMs. Specifically, in the first stage, the LTR focuses on language to recursively generate a language-centric logical tree, which gradually breaks down the complex cognitive question into simple perceptual ones and plans the reasoning path through a RAG-based few-shot approach. In the second stage, with the aid of video content, the LTR performs bottom-up logical reasoning within the tree to derive the final answer along with the traceable reasoning path. Experiments across 11 VideoQA benchmarks demonstrate that our LTR framework significantly improves both accuracy and interpretability compared to state-of-the-art MLLMs. To our knowledge, this is the first work to implement a language-centric logical tree to guide MLLM reasoning in VideoQA, paving the way for language-centric video understanding from perception to cognition.
Deep Learning · Attention Mechanisms
We present a nonparametric interpretation for deep learning compatible modern Hopfield models and utilize this new perspective to debut efficient variants. Our key contribution stems from interpreting the memory storage and retrieval processes in modern Hopfield models as a nonparametric regression problem subject to a set of query-memory pairs.Interestingly,our framework not only recovers the known results from the original dense modern Hopfield model but also fills the void in the literature regarding efficient modern Hopfield models, by introducing *sparse-structured* modern Hopfield models with sub-quadratic complexity.We establish that this sparse model inherits the appealing theoretical properties of its dense analogue --- connection with transformer attention, fixed point convergence and exponential memory capacity.Additionally, we showcase the versatility of our framework by constructing a family of modern Hopfield models as extensions, including linear, random masked, top-$K$ and positive random feature modern Hopfield models.Empirically, we validate our framework in both synthetic and realistic settings for memory retrieval and learning tasks.
Applications · Chemistry, Physics, and Earth Sciences
Deep learning has been extensively employed as a powerful function approximator for modeling physics-based problems described by partial differential equations (PDEs). Despite their popularity, standard deep learning models often demand prohibitively large computational resources and yield limited accuracy when scaling to large-scale, high-dimensional physical problems. Their black-box nature further hinders their application in industrial problems where interpretability and high precision are critical. To overcome these challenges, this paper introduces Interpolating Neural Network-Tensor Decomposition (INN-TD), a scalable and interpretable framework that has the merits of both machine learning and finite element methods for modeling large-scale physical systems. By integrating locally supported interpolation functions from finite element into the network architecture, INN-TD achieves a sparse learning structure with enhanced accuracy, faster training/solving speed, and reduced memory footprint. This makes it particularly effective for tackling large-scale high-dimensional parametric PDEs in training, solving, and inverse optimization tasks in physical problems where high precision is required.
Applications · Computer Vision
We propose **DocVXQA**, a novel framework for visually self-explainable document question answering, where the goal is not only to produce accurate answers to questions but also to learn visual heatmaps that highlight critical regions, offering interpretable justifications for the model decision. To integrate explanations into the learning process, we quantitatively formulate explainability principles as explicit learning criteria.Unlike conventional relevance map methods that solely emphasize regions relevant to the answer, our context-aware DocVXQA delivers explanations that are contextually sufficient yet representation-efficient. This fosters user trust while achieving a balance between predictive performance and interpretability in document visual question answering applications. Extensive experiments, including human evaluation, provide strong evidence supporting the effectiveness of our method.