Jia Li
12 papers · Latest:
An Empirical Study of Proactive Coding Assistants in Real-World Software Development
This study reveals a significant simulation-to-reality gap in proactive coding assistant research, introducing ProCodeBench, a real-world benchmark for intent prediction.
To Fuse or to Drop? Dual-Path Learning for Resolving Modality Conflicts in Multimodal Emotion Recognition
DCR is a dual-path framework that intelligently fuses or drops modalities to resolve conflicts in multimodal emotion recognition, improving robustness.
MEMCoder: Multi-dimensional Evolving Memory for Private-Library-Oriented Code Generation
MEMCoder improves private library code generation by autonomously evolving usage guidelines from problem-solving, outperforming RAG.
RealBench: A Repo-Level Code Generation Benchmark Aligned with Real-World Software Development Practices
RealBench is a new benchmark for repo-level code generation, using structured designs (UML) to better align LLM evaluation with real-world software development.
MER 2026: From Discriminative Emotion Recognition to Generative Emotion Understanding
MER2026 expands emotion understanding challenges from discriminative recognition to generative analysis, introducing new tasks and multimodal approaches.
DebugRepair: Enhancing LLM-Based Automated Program Repair via Self-Directed Debugging
DebugRepair enhances LLM-based automated program repair by using self-directed debugging to collect intermediate runtime evidence, significantly improving bug-fixing performance.
Bridging the Gap between User Intent and LLM: A Requirement Alignment Approach for Code Generation
REA-Coder improves LLM code generation by iteratively aligning user requirements, addressing the common issue of LLMs misunderstanding prompts.
Trajectory Planning for a Multi-UAV Rigid-Payload Cascaded Transportation System Based on Enhanced Tube-RRT*
This paper introduces a two-stage trajectory planning framework for multi-UAV rigid-payload systems in cluttered environments, using Enhanced Tube-RRT*.
Seeing is Believing: Robust Vision-Guided Cross-Modal Prompt Learning under Label Noise
VisPrompt is a vision-guided cross-modal prompt learning framework that robustly learns prompts for VLMs even with noisy labels.
Evaluating the Formal Reasoning Capabilities of Large Language Models through Chomsky Hierarchy
ChomskyBench evaluates LLM formal reasoning across the Chomsky Hierarchy, revealing performance stratification and severe efficiency barriers for complex tasks.
StarCoder 2 and The Stack v2: The Next Generation
StarCoder2 is a next-generation open-source Code LLM trained on a vastly expanded and diverse dataset, achieving state-of-the-art performance on multiple code benchmarks while being more parameter-efficient than larger models.
StarCoder: may the source be with you!
StarCoder is a 15.5B parameter open-source code generation model trained on a trillion tokens that outperforms existing open Code LLMs across multiple languages and offers advanced safety and usability features.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.