Recent advances in speech foundation models are largely driven by scaling both model size and data, enabling them to perform a wide range of tasks, including speech recognition. Traditionally, ASR models are evaluated using metrics like Word Error Rate (WER) and Character Error Rate (CER), which depend on ground truth labels. As a result of limited labeled data from diverse domains and testing conditions, the true generalization capabilities of these models beyond standard benchmarks remain unclear. Moreover, labeling data is both costly and time-consuming. To address this, we propose a novel label-free approach for approximating ASR performance metrics, eliminating the need for ground truth labels. Our method utilizes multimodal embeddings in a unified space for speech and transcription representations, combined with a high-quality proxy model to compute proxy metrics. These features are used to train a regression model to predict key ASR metrics like Word Error Rate (WER) and Character Error Rate (CER). We experiment with over 40 models across 14 datasets representing both standard and in-the-wild testing conditions. Our results show that we approximate the metrics within a single-digit absolute difference across all experimental configurations, outperforming the most recent baseline by more than 50%.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
Existing LMs struggle with proof-oriented programming due to data scarcity, which manifest in two key ways: (1) a lack of sufficient corpora for proof-oriented programming languages such as F*, and (2) the absence of large-scale, project-level proof-oriented implementations that can teach the model the intricate reasoning process when performing proof-oriented programming. We present the first on synthetic data augmentation for project level proof oriented programming for both generation and repair. Our method addresses data scarcity by synthesizing basic proof-oriented programming problems for proficiency in that language; incorporating diverse coding data for reasoning capability elicitation and creating new proofs and repair data within existing repositories. This approach enables language models to both synthesize and repair proofs for function- and repository-level code. We show that our fine-tuned 14B parameter model, PoPilot, can exceed the performance of the models that outperforms GPT-4o in project-level proof-oriented programming by 64% relative margin, and can improve GPT-4o’s performance by 54% by repairing its outputs over GPT-4o’s self-repair.
Dynamic Personality in LLM Agents: A Framework for Evolutionary Modeling and Behavioral Analysis in the Prisoner’s Dilemma
PDF ↗Using Large Language Model agents to simulate human game behaviors offers valuable insights for human social psychology in anthropomorphic AI research. While current models rely on static personality traits, real-world evidence shows personality evolves through environmental feedback. Recent work introduced dynamic personality traits but lacked natural selection processes and direct psychological metrics, failing to accurately capture authentic dynamic personality variations. To address these limitations, we propose an enhanced framework within the Prisoner’s Dilemma, a socially significant scenario. By using game payoffs as environmental feedback, we drive adaptive personality evolution and analyze correlations between personality metrics and behavior. Our framework reveals new behavioral patterns of agents and evaluates personality-behavior relationships, advancing agent-based social simulations and human-AI symbiosis research.
Detecting toxicity in online multimodal environments, such as memes, remains a challenging task due to the complex contextual connections across modalities (e.g., text and visual), which demand both common-sense reasoning and contextual awareness. To bridge this gap, we propose a hybrid neurosymbolic framework that unifies (1) distillation of implicit contextual knowledge (e.g., sarcasm, cultural references) from Large Vision-Language Models (LVLMs) and (2) infusion of explicit relational semantics through sub-graphs from Knowledge Graphs (KGs). Experimental results on two benchmark datasets show the superior performance of our approach, Knowledge-Infused Distilled Vision-Language Model (KID-VLM), over the state-of-the-art baselines across AUC and F1, with improvements of 0.5%, and 10.6%, respectively, in HatefulMemes Benchmark across variants. Further, KID-VLM demonstrates better generalizability and achieves the best performance across all baselines in the HarMeme Dataset with a 6.3% and 3.2% in F1 and AUC.Given the contextual complexity of the toxicity detection, KID-VLM showcases the significance of learning compact models (~500M parameters) from both explicit (i.e., KG) and implicit (i.e., LVLMs) contextual cues incorporated through a hybrid neurosymbolic approach. Our codes and pretrained models are publicly available.
Triton, a high-level Python-like language designed for building efficient GPU kernels, is widely adopted in deep learning frameworks due to its portability, flexibility, and accessibility. However, programming and parallel optimization still require considerable trial and error from Triton developers. Despite advances in large language models (LLMs) for conventional code generation, these models struggle to generate accurate, performance-optimized Triton code, as they lack awareness of its specifications and the complexities of GPU programming. More critically, there is an urgent need for systematic evaluations tailored to Triton. In this work, we introduce TritonBench, the first comprehensive benchmark for Triton operator generation. TritonBench features two evaluation channels: a curated set of 184 real-world operators from GitHub and a collection of operators aligned with PyTorch interfaces. Unlike conventional code benchmarks prioritizing functional correctness, TritonBench also profiles efficiency performance on widely deployed GPUs aligned with industry applications. Our study reveals that current state-of-the-art code LLMs struggle to generate efficient Triton operators, highlighting a significant gap in high-performance code generation.
Your Language Model May Think Too Rigidly: Achieving Reasoning Consistency with Symmetry-Enhanced Training
PDF ↗Large Language Models (LLMs) have demonstrated strong reasoning capabilities across various tasks. However, even minor variations in query phrasing, despite preserving the underlying semantic meaning, can significantly affect their performance. To address this, we focus on enhancing LLMs’ awareness of symmetry in query variations and propose syMmetry-ENhanceD (MEND) data augmentation, a data-centric approach that improves the model’s ability to extract useful information from context. Unlike existing methods that emphasize reasoning chain augmentation, our approach improves model robustness at the knowledge extraction stage through query augmentation, enabling more data-efficient training and stronger generalization to Out-of-Distribution (OOD) settings. Extensive experiments on both logical and arithmetic reasoning tasks show that MEND enhances reasoning performance across diverse query variations, providing new insights into improving LLM robustness through structured dataset curation.
Automatic depression detection provides cues for early clinical intervention by clinicians. Clinical interviews for depression detection involve dialogues centered around multiple themes. Existing studies primarily design end-to-end neural network models to capture the hierarchical structure of clinical interview dialogues. However, these methods exhibit defects in modeling the thematic content of clinical interviews: 1) they fail to explicitly capture intra-theme and inter-theme correlation, and 2) they do not allow clinicians to intervene and focus on themes of interest. To address these issues, this paper introduces an interactive depression detection framework, namely Predicting Depression in Screening Interviews from Interactive Multi-Theme Collaboration (PDIMC). PDIMC leverages in-context learning techniques to identify themes in clinical interviews and then models both intra-theme and inter-theme correlation. Additionally, it employs AI-driven feedback to simulate the interests of clinicians, enabling interactive adjustment of theme importance. PDIMC achieves absolute improvements of 12% on Recall and 35% on F1-dep. metrics, compared to the previous state-of-the-art model on the depression detection dataset DAIC-WOZ, which demonstrates the effectiveness of capturing theme correlation and incorporating interactive external feedback.
Writing is inherently iterative, each revision enhancing information representation. One revision may contain many edits. Examination of the intentions behind edits provides valuable insights into an editor’s expertise, the dynamics of collaborative writing, and the evolution of a document. Current research on edit intentions lacks a comprehensive edit intention taxonomy (EIT) that spans multiple application domains. As a result, researchers often create new EITs tailored to specific needs, a process that is both time-consuming and costly. To address this gap, we propose UniT, a Unified edit intention Taxonomy that integrates existing EITs encompassing a wide range of edit intentions. We examine the lineage relationship and the construction of 24 EITs. They together have 232 categories across various domains. During the literature survey and integration process, we identify challenges such as one-to-many category matches, incomplete definitions, and varying hierarchical structures. We propose solutions for resolving these issues. Finally, our evaluation shows that our UniT achieves higher inter-annotator agreement scores compared to existing EITs and is applicable to a large set of application domains.
Evaluation of LLMs in Medical Text Summarization: The Role of Vocabulary Adaptation in High OOV Settings
PDF ↗Large Language Models (LLMs) recently achieved great success in medical text summarization by simply using in-context learning. However, these recent efforts do not perform fine-grained evaluations under difficult settings where LLMs might fail. They typically report performance scores over the entire dataset. Through our benchmarking study, we show that LLMs show a significant performance drop for data points with high concentration of out-of-vocabulary (OOV) words or with high novelty. Vocabulary adaptation is an intuitive solution to this vocabulary mismatch issue where the LLM vocabulary gets updated with certain expert domain (here, medical) words or subwords. An interesting finding from our study is that Llama-3.1, even with a vocabulary size of around 128K tokens, still faces _over-fragmentation_ issue with medical words. To that end, we show vocabulary adaptation helps improve the LLM summarization performance even in difficult settings. Through extensive experimentation of multiple vocabulary adaptation strategies, two continual pretraining strategies, and three benchmark medical summarization datasets, we gain valuable insights into the role of vocabulary adaptation strategies for customizing LLMs to the medical domain. We also performed a human evaluation study with medical experts where they found that vocabulary adaptation results in more relevant and faithful summaries. Our codebase is made publicly available at https://github.com/gb-kgp/LLM-MedicalSummarization-Benchmark.
GSQ-Tuning: Group-Shared Exponents Integer in Fully Quantized Training for LLMs On-Device Fine-tuning
PDF ↗Large Language Models (LLMs) fine-tuning technologies have achieved remarkable results. However, traditional LLM fine-tuning approaches face significant challenges: they require large Floating Point(FP) computation, raising privacy concerns when handling sensitive data, and are impractical for resource-constrained edge devices. While Parameter-Efficient Fine-Tuning (PEFT) techniques reduce trainable parameters, their reliance on floating-point arithmetic creates fundamental incompatibilities with edge hardware. In this work, we introduce a novel framework for on-device LLM fine-tuning that eliminates the need for floating-point operations in both inference and training, named GSQ-Tuning. At its core is the Group-Shared Exponents Integer format, which efficiently represents model parameters in integer format using shared exponents among parameter groups. When combined with LoRA-like adapters, this enables fully integer-based fine-tuning that is both memory and compute efficient. We demonstrate that our approach achieves accuracy comparable to FP16-based fine-tuning while significantly reducing memory usage ( 50%). Moreover, compared to FP8, at comparable performance levels, our method can reduce 5x power consumption and 11x chip area, making large-scale model adaptation feasible on edge devices.
Slice discovery refers to identifying systematic biases in the mistakes of pre-trained vision models. Current slice discovery methods in computer vision rely on converting input images into sets of attributes and then testing hypotheses about configurations of these pre-computed attributes associated with elevated error patterns. However, such methods face several limitations: 1) they are restricted by the predefined attribute bank; 2) they lack the common sense reasoning and domain-specific knowledge often required for specialized fields radiology; 3) at best, they can only identify biases in image attributes while overlooking those introduced during preprocessing or data preparation. We hypothesize that bias-inducing variables leave traces in the form of language (logs), which can be captured as unstructured text. Thus, we introduce ladder, which leverages the reasoning capabilities and latent domain knowledge of Large Language Models (LLMs) to generate hypotheses about these mistakes. Specifically, we project the internal activations of a pre-trained model into text using a retrieval approach and prompt the LLM to propose potential bias hypotheses. To detect biases from preprocessing pipelines, we convert the preprocessing data into text and prompt the LLM. Finally, ladder generates pseudo-labels for each identified bias, thereby mitigating all biases without requiring expensive attribute annotations.Rigorous evaluations on 3 natural and 3 medical imaging datasets, 200+ classifiers, and 4 LLMs with varied architectures and pretraining strategies – demonstrate that ladder consistently outperforms current methods. Code is available: https://github.com/batmanlab/Ladder.
Evaluating tables qualitatively and quantitatively poses a significant challenge, as standard metrics often overlook subtle structural and content-level discrepancies. To address this, we propose a rubric-based evaluation framework that integrates multi-level structural descriptors with fine-grained contextual signals, enabling more precise and consistent table comparison. Building on this, we introduce TabXEval, an eXhaustive and eXplainable two-phase evaluation framework. TabXEval first aligns reference and predicted tables structurally via TabAlign, then performs semantic and syntactic comparison using TabCompare, offering interpretable and granular feedback. We evaluate TabXEval on TabXBench, a diverse, multi-domain benchmark featuring realistic table perturbations and human annotations. A sensitivity-specificity analysis further demonstrates the robustness and explainability of TabXEval across varied table tasks. Code and data are available at https://corallab- asu.github.io/tabxeval/.
Improvements in large language models have led to increasing optimism that they can serve as reliable evaluators of natural language generation outputs. In this paper, we challenge this optimism in regards to factuality evaluation.We re-evaluate five state-of-the-art factuality metrics on a collection of 11 datasets for summarization, retrieval-augmented generation, and question answering.We find that these evaluators are inconsistent with each other and often misestimate the factual accuracy of NLG systems, both of which can lead to a variety of pitfalls.We further show that these metrics exhibit biases against highly paraphrased outputs and outputs that draw upon faraway parts of the source documents.We urge users of factuality metrics to proceed with caution and manually validate the reliability of these metrics in their domain of interest.
Tokens for Learning, Tokens for Unlearning: Mitigating Membership Inference Attacks in Large Language Models via Dual-Purpose Training
PDF ↗Large language models (LLMs) have become the backbone of modern natural language processing but pose privacy concerns about leaking sensitive training data. Membership inference attacks (MIAs), which aim to infer whether a sample is included in a model’s training dataset, can serve as a foundation for broader privacy threats. Existing defenses designed for traditional classification models do not account for the sequential nature of text data. As a result, they either require significant computational resources or fail to effectively mitigate privacy risks in LLMs. In this work, we propose DuoLearn, a lightweight yet effective empirical privacy defense for protecting training data of language models by leveraging token-specific characteristics. By analyzing token dynamics during training, we propose a token selection strategy that categorizes tokens into hard tokens for learning and memorized tokens for unlearning. Subsequently, our training-phase defense optimizes a novel dual-purpose token-level loss to achieve a Pareto-optimal balance between utility and privacy. Extensive experiments demonstrate that our approach not only provides strong protection against MIAs but also improves language modeling performance by around 10% across various LLM architectures and datasets compared to the baselines.
Exploiting Phonetics and Glyph Representation at Radical-level for Classical Chinese Understanding
PDF ↗The diachronic gap between classical and modern Chinese arises from century-scale language evolution through cumulative changes in phonological, syntactic, and lexical systems, resulting in substantial semantic variation that poses significant challenges for the computational modeling of historical texts. Current methods always enhance classical Chinese understanding of pre-trained language models through corpus pre-training or semantic integration. However, they overlook the synergistic relationship between phonetic and glyph features within Chinese characters, which is a critical factor in deciphering characters’ semantics. In this paper, we propose RPGCM, a radical-level phonetics and glyph representation enhanced Chinese model, with powerful fine-grained semantic modeling capabilities. Our model establishes robust contextualized representations through: (1) rules-based radical decomposition and bype pair encoder (BPE) based radical aggregated for structural pattern recognition, (2) phonetic-glyph semantic mapping, and (3) dynamic semantic fusion. Experimental results on CCMRC, WYWEB, and C³Bench benchmarks demonstrate the RPGCM’s superiority and validate that explicit radical-level modeling effectively mitigates semantic variations.
Recent advancements in Large Language Models (LLMs) have showcased striking results on existing logical reasoning benchmarks, with some models even surpassing human performance. However, the true depth of their competencies and robustness in reasoning tasks remains an open question. To this end, in this paper, we focus on two popular reasoning tasks: arithmetic reasoning and code generation. Particularly, we introduce (i) a general ontology of perturbations for math and coding questions, (ii) a semi-automatic method to apply these perturbations, and (iii) two datasets, GSMore and HumanEval-Core, respectively, of perturbed math and coding problems to probe LLM capabilities in numeric reasoning and coding tasks.Through comprehensive evaluations of both closed-source and open-source LLMs, we show a significant performance drop across all the models against the perturbed questions, suggesting that the current LLMs lack robust problem solving skills and structured reasoning abilities in many areas, as defined by our ontology.
Evaluating large language models (LLMs) on their linguistic reasoning capabilities is an important task to understand the gaps in their skills that may surface during large-scale adoption. In this work, we investigate the abilities of such models to perform abstract multilingual reasoning through the lens of linguistic puzzles on extremely low-resource languages. As these translation tasks involve inductive and deductive reasoning from reference instances, we examine whether diverse auxiliary demonstrations can be automatically induced from seed exemplars, through analogical prompting. We employ a two-stage procedure, first generating analogical exemplars with a language model, and then applying them in-context along with provided target language exemplars. Our results on the modeLing dataset show that analogical prompting is effective in eliciting models’ knowledge of language grammar similarities, boosting the performance of GPT-4o by as much as 8.1% and Llama-3.1-405B-Instruct by 5.9% over chain-of-thought approaches. Furthermore, we demonstrate that our method generalizes to other tasks present in Linguistics Olympiad competitions, achieving state-of-the-art results across nearly all problem types and difficulty levels in the LINGOLY dataset.
Query-driven recommendation with unknown items poses a challenge for users to understand why certain items are appropriate for their needs. Query-driven Contrastive Summarization (QCS) is a methodology designed to address this issue by leveraging language-based item descriptions to clarify contrasts between them. However, existing state-of-the-art contrastive summarization methods such as STRUM-LLM fall short of this goal. To overcome these limitations, we introduce Q-STRUM Debate, a novel extension of STRUM-LLM that employs debate-style prompting to generate focused and contrastive summarizations of item aspects relevant to a query. Leveraging modern large language models (LLMs) as powerful tools for generating debates, Q-STRUM Debate provides enhanced contrastive summaries. Experiments across three datasets demonstrate that Q-STRUM Debate yields significant performance improvements over existing methods on key contrastive summarization criteria, thus introducing a novel and performant debate prompting methodology for QCS.
The challenge of developing agents capable of open-world planning remains fundamental to artificial general intelligence (AGI). While large language models (LLMs) have made progress with their vast world knowledge, their limitations in perception, memory, and reliable reasoning still hinder LLM-based agents from achieving human-level performance in long-term tasks. Drawing inspiration from human cognitive-metacognitive collaboration, we propose Metagent-P, integrating the world knowledge of LLMs, the symbolic reasoning capabilities of cognitive architectures, and the self-reflection characteristic of metacognition to construct a “planning-verification-execution-reflection” framework. Metagent-P improves experience utilization through multimodal memory integration. It uses a neural-symbolic hierarchical representation structure to ensure the plan’s reasoning correctness in advance. Finally, it actively adapts the agent to dynamic environments through monitoring, evaluation, and regulation mechanisms. Experimental results show Metagent-P significantly outperforms current state-of-the-art methods in Minecraft. In long-term tasks, Metagent-P reduces the average replanning counts by 34% and exceeds the average human success rate by 18.96%. Additionally, Metagent-P also demonstrates self-evolution through step-by-step open-world exploration.
Beyond instruction-conditioning, MoTE: Mixture of Task Experts for Multi-task Embedding Models
PDF ↗Dense embeddings are fundamental to modern machine learning systems, powering Retrieval-Augmented Generation (RAG), information retrieval, and representation learning. While instruction-conditioning has become the dominant approach for embedding specialization, its direct application to low-capacity models imposes fundamental representational constraints that limit the performance gains derived from specialization. In this paper, we analyze these limitations and introduce the Mixture of Task Experts (MoTE) transformer block, which leverages task-specialized parameters trained with Task-Aware Contrastive Learning () to enhance the model’s ability to generate specialized embeddings. Empirical results show that MoTE achieves 64% higher performance gains in retrieval datasets (+3.27\rightarrow +5.21) and 43% higher performance gains across all datasets (+1.81\rightarrow 2.60). Critically, these gains are achieved without altering instructions, training data, inference time, or number of active parameters.