Just-in-Time software defect prediction (JIT-SDP) plays a critical role in prioritizing risky code changes during code review and continuous integration. However, existing datasets often suffer from noisy labels and low precision in identifying bug-inducing commits. To address this, we present ReDef (Revert-based Defect dataset), a high-confidence benchmark of function-level modifications curated from 22 large-scale C/C++ projects. Defective cases are anchored by revert commits, while clean cases are validated through post-hoc history checks. Ambiguous instances are conservatively filtered out via a GPT-assisted triage process involving multiple votes and audits. This pipeline yields 3,164 defective and 10,268 clean modifications, offering substantially more reliable labels than prior existing resources. Beyond dataset construction, we provide the first systematic evaluation of how pre-trained language models (PLMs) reason about code modifications -- specifically, which input encodings most effectively expose change information, and whether models genuinely capture edit semantics. We fine-tune CodeBERT, CodeT5+, and UniXcoder under five encoding strategies, and further probe their sensitivity through counterfactual perturbations that swap added/deleted blocks, invert diff polarity, or inject spurious markers. Our results show that compact diff-style encodings consistently outperform whole-function formats across all PLMs, with statistical tests confirming large, model-independent effects. However, under counterfactual tests, performance degrades little or not at all -- revealing that what appears to be robustness in fact reflects reliance on superficial cues rather than true semantic understanding. These findings indicate that, unlike in snapshot-based tasks, current PLMs remain limited in their ability to genuinely comprehend code modifications.

Speech-to-speech large language models (SLLMs) are attracting increasing attention. Derived from text-based large language models (LLMs), SLLMs often exhibit degradation in knowledge and reasoning capabilities. We hypothesize that this limitation arises because current training paradigms for SLLMs fail to bridge the acoustic-semantic gap in the feature representation space. To address this issue, we propose EchoX, which leverages semantic representations and dynamically generates speech training targets. This approach integrates both acoustic and semantic learning, enabling EchoX to preserve strong reasoning abilities as a speech LLM. Experimental results demonstrate that EchoX, with about six thousand hours of training data, achieves advanced performance on multiple knowledge-based question-answering benchmarks. The project is available at https://github.com/FreedomIntelligence/EchoX.

Knowledge-based visual question answering (KB-VQA) requires a model to understand images and utilize external knowledge to provide accurate answers. Existing approaches often directly augment models with retrieved information from knowledge sources while ignoring substantial knowledge redundancy, which introduces noise into the answering process. To address this, we propose a training-free framework with knowledge focusing for KB-VQA, that mitigates the impact of noise by enhancing knowledge relevance and reducing redundancy. First, for knowledge retrieval, our framework concludes essential parts from the image-question pairs, creating low-noise queries that enhance the retrieval of highly relevant knowledge. Considering that redundancy still persists in the retrieved knowledge, we then prompt large models to identify and extract answer-beneficial segments from knowledge. In addition, we introduce a selective knowledge integration strategy, allowing the model to incorporate knowledge only when it lacks confidence in answering the question, thereby mitigating the influence of redundant information. Our framework enables the acquisition of accurate and critical knowledge, and extensive experiments demonstrate that it outperforms state-of-the-art methods.

Video understanding has advanced rapidly, fueled by increasingly complex datasets and powerful architectures. Yet existing surveys largely classify models by task or family, overlooking the structural pressures through which datasets guide architectural evolution. This survey is the first to adopt a dataset-driven perspective, showing how motion complexity, temporal span, hierarchical composition, and multimodal richness impose inductive biases that models should encode. We reinterpret milestones, from two-stream and 3D CNNs to sequential, transformer, and multimodal foundation models, as concrete responses to these dataset-driven pressures. Building on this synthesis, we offer practical guidance for aligning model design with dataset invariances while balancing scalability and task demands. By unifying datasets, inductive biases, and architectures into a coherent framework, this survey provides both a comprehensive retrospective and a prescriptive roadmap for advancing general-purpose video understanding.

First submitted: 30 Oct 2023. The final version will be available open access via the journal. Abstract This study explores the use of generative AI for automating the classification of tutors' Dialogue Acts (DAs), aiming to reduce the time and effort required by traditional manual coding. This case study uses the open - source CIMA corpus, in which tutors' re sponses are pre - annotated into four DA categories. Both GPT - 3.5 - turbo and GPT - 4 models were tested using tailored prompts. Results show that GPT - 4 achieved 80% accuracy, a weighted F1 - score of 0.81, and a Cohen's Kappa of 0.74, surpassing baseline performa nce and indicating substantial agreement with human annotations. These findings suggest that generative AI has strong potential to provide an efficient and accessible approach to DA classification, with meaningful implications for educational dialogue analysis. The study also highlights the importance of task - specific label definitions and contextual information in enhanc ing the quality of automated annotation. Finally, it underscores the ethical considerations associated with the use of generative AI and the need for responsible and transparent research practices.

Large Language Models (LLMs) have transformed both everyday life and scientific research. However, adapting LLMs from general-purpose models to specialized tasks remains challenging, particularly in resource-constrained environments. Low-Rank Adaptation (LoRA), a prominent method within Parameter-Efficient Fine-Tuning (PEFT), has emerged as a promising approach to LLMs by approximating model weight updates using low-rank decomposition. However, LoRA is limited by its uniform rank ( r ) allocation to each incremental matrix, and existing rank allocation techniques aimed at addressing this issue remain computationally inefficient, complex, and unstable, hindering practical applications. To address these limitations, we propose Sensitivity-LoRA, an efficient fine-tuning method that dynamically allocates ranks to weight matrices based on both their global and local sensitivities. It leverages the second-order derivatives (Hessian Matrix) of the loss function to effectively capture weight sensitivity, enabling optimal rank allocation with minimal computational overhead. Our experimental results have demonstrated robust effectiveness, efficiency and stability of Sensitivity-LoRA across diverse tasks and benchmarks.

Despite being the 5th most spoken language, Bangla remains underrepresented in Large Language Models (LLMs), particularly for code generation. This primarily stems from the scarcity of high-quality data to pre-train and/or finetune such models. Hence, we introduce the first dedicated family of Code LLMs for Bangla (1B & 9B). We offer three major contributions: (1) a comprehensive Bangla code instruction datasets for programming domain adaptation; (2) MBPP-Bangla, an evaluation benchmark for Bangla code generation; and (3) the TigerCoder-family of Code LLMs, achieving significant ~11-18% performance gains at Pass@1 over existing multilingual and general-purpose Bangla LLMs. Our findings show that curated, high-quality datasets can overcome limitations of smaller models for low-resource languages. We open-source all resources to advance further Bangla LLM research.

CPU-based trusted execution environments (TEEs) and differential privacy (DP) have gained wide applications for private inference. Due to high inference latency in TEEs, researchers use partition-based approaches that offload linear model components to GPUs. However, dense nonlinear layers of large language models (LLMs) result in significant communication overhead between TEEs and GPUs. DP-based approaches apply random noise to protect data privacy, but this compromises LLM performance and semantic understanding. To overcome the above drawbacks, this paper proposes CMIF, a Confidential and efficient Model Inference Framework. CMIF confidentially deploys the embedding layer in the client-side TEE and subsequent layers on GPU servers. Meanwhile, it optimizes the Report-Noisy-Max mechanism to protect sensitive inputs with a slight decrease in model performance. Extensive experiments on Llama-series models demonstrate that CMIF reduces additional inference overhead in TEEs while preserving user data privacy.

Vision-Language-Action (VLA) models exhibit unprecedented capabilities for embodied intelligence. However, their extensive computational and memory costs hinder their practical deployment. Existing VLA compression and acceleration approaches conduct quantization or token pruning in an ad-hoc manner but fail to enable both for a holistic efficiency improvement due to an observed incompatibility. This work introduces SQAP-VLA, the first structured, training-free VLA inference acceleration framework that simultaneously enables state-of-the-art quantization and token pruning. We overcome the incompatibility by co-designing the quantization and token pruning pipeline, where we propose new quantization-aware token pruning criteria that work on an aggressively quantized model while improving the quantizer design to enhance pruning effectiveness. When applied to standard VLA models, SQAP-VLA yields significant gains in computational efficiency and inference speed while successfully preserving core model performance, achieving a $\times$1.93 speedup and up to a 4.5\% average success rate enhancement compared to the original model.

Information extraction (IE) is a fundamental task in natural language processing (NLP), which encompasses a wide range of subtasks such as Named Entity Recognition (NER), Relation Extraction (RE), and Event Extraction (EE) [1-4]. Traditionally, these tasks have been addressed by specialized models trained in task-specific datasets. However, the fragmentation of tasks and schemas has hindered the development of generalizable and scalable IE tasks. To address this limitation, recent research has focused on universal information extraction (UIE), which aims to model all IE tasks within a universal framework. A seminal work in this direction is proposed by Lu et al., which introduced a structured generation paradigm that encodes diverse IE tasks into a common semantic representation[5]. Building on this, InstructUIE[6] extended the idea by incorporating multi-task instruction tuning, enabling models to generalize across tasks via natural language instructions. With the emergence of powerful LLMs[7-11], significant advancements have been made across long-standing NLP tasks such as text classification[12-16], intent recognition[17, 18], entity linking[19-22], and beyond. Inspired by their robust performance and adaptability, researchers have explored their potential for information extraction through prompting and in-context learning learning[23, 24]. For example, CodeIE demonstrated that code generation models can serve as strong few-shot IE extractors by using structured code-like commands[25].