论文
已发表与在投工作的本地 PDF 归档。
已接收
To Run or Not to Run: Analyzing the Cost-Effectiveness of Code Execution in LLM-Based Program Repair
PDFISSTA 2026
#Program Repair #Code Agents
How Much Static Structure Do Code Agents Need? A Study of Deterministic Anchoring
PDFISSTA 2026
#Static Analysis #Code Agents
Phantom Rendering Detection: Identifying and Analyzing Unnecessary UI Computations
PDFFSE 2026
#Performance
MazeBreaker: Multi-Agent Reinforcement Learning for Dynamic Jailbreaking of LLM Security Defenses
PDFICSE 2026
#LLM Security
HapRepair: Learn to Repair OpenHarmony Apps
PDFFSE Industry 2025
#Program Repair #Static Analysis
Effective Fine-tuning for Low-resource Languages: A Case Study of Cangjie
PDFEMSE 2026
#Misc
Open-Source AI-based SE Tools: Opportunities and Challenges of Collaborative Software Learning
PDFTOSEM 2024
#Survey
Exploring Code Analysis: Zero-Shot Insights on Syntax and Semantics with LLMs
PDFTOSEM 2026
#Static Analysis
在投与预印本
OpenHarmony Bench: Evaluating LLMs and Coding Agents on OpenHarmony App Development
预印本Technical Report 2026
An app-level benchmark for OpenHarmony ArkTS development that evaluates whether coding agents can turn feature requests, structured specifications, and bug reports into buildable applications with correct end-to-end behavior.
RepoRescue: An Empirical Study of LLM Agents on Whole-Repository Compatibility Rescue
预印本Studies whether deployed LLM agents can rescue old repositories broken by runtime and dependency drift, with source-only auditing, runtime blocking, and realistic-use validation.
AtomicCommitBench: Can Coding Agents Reconstruct Commit Histories from Squashed Patches?
预印本Benchmarks whether coding agents can reconstruct replayable, reviewable commit histories from squashed patches, measuring replay validity, human-aligned grouping, and selective-revert failure containment.
Executing as You Generate: Hiding Execution Latency in LLM Code Generation
预印本Proposes EAGER, a parallel execution paradigm for LLM code generation that overlaps code generation and execution, reducing end-to-end latency by up to 55%.