AEG-Edit: Bridging API Evolution and Code Semantics for Accurate API Usage in Code Generation
As APIs across standard libraries and software ecosystems evolve, the API knowledge frozen in pre-trained LLMs becomes outdated. Prompt-based methods supply updated documentation at inference time yet suffer from context neglect, often reverting to memorized deprecated usages. Existing parametric approaches, whether fine-tuning or model editing, face two common challenges: they treat API changes as textual fact updates, overlooking the changes in code semantics needed for executable correctness (C1), and lack a mechanism to concentrate updates on positions involved in the API evolution, causing parameter interference beyond the evolution boundary (C2). We propose AEG-Edit, which constructs a heterogeneous API evolution graph, encodes and aligns the graph signal with the LLM hidden space, and concentrates gradient effort on positions involved in the API evolution through API-weighted windowing and optimization. On RustEvo²+ (708 Rust tasks) and PyEvo (622 Python tasks) across two representative open-source LLMs at the 7–8B parameter scale, AEG-Edit achieves statistically significant Pass@1 gains on RustEvo²+, improving over the strongest baseline by up to 6.49% on Qwen2.5-7B-Instruct. Ablation and transferability analyses confirm that the heterogeneous API evolution graph drives the gains and transfers to alternative editing backbones.
Thu 15 OctDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
10:30 - 12:30 | |||
10:30 15mTalk | DepWareTrans: Dependency-Aware Incremental Repository Migration across Co-executable Languages Industry Showcase Sivajeet Chand Technical University of Munich, Alexander Pretschner TU Munich, Steve Haupt andrena objects, Derui Zhu Rochester Institute of Technology, Sushant Kumar Pandey University of Groningen, The Netherlands Pre-print | ||
10:45 15mTalk | Aligning User Requirement with Large Language Model in Code Generation Research Papers Jia Li Wuhan University, Ruiqi Bai Wuhan University, Yangkang Luo Wuhan University, Yiran Zhang Nanyang Technological University, Wentao Yang Wuhan University, Zeyu Sun Institute of Software, Chinese Academy of Sciences, Tiankuo Zhao Wuhan University, Dongming Jin Peking University, China, Lei Li University of Hong Kong, Zhi Jin Peking University, Wuhan University | ||
11:00 15mTalk | Pattern over Pixels: Measuring Pattern Completion Bias in Multimodal Code Generation Research Papers Khai Nguyen William & Mary, Oscar Chaparro William & Mary, Antonio Mastropaolo William and Mary, USA | ||
11:15 15mTalk | Customizing an LLM for Enterprise Software Engineering Industry Showcase Aditya Kini Google, Satish Chandra Meta Platforms, Inc., Milad Hashemi Google DeepMind, Saksham Thakur Google, Aditya Pandey Google, Vincent Nguyen Google, Marc Brockschmidt Google DeepMind, Franjo Ivančić Google, Danny Tarlow Google, Parthasarathy Ranganathan Google, Petros Maniatis Google DeepMind, Ahmed Omran Google, Zaheer Abbas Google DeepMind, Anita Gergely Google DeepMind, Martin Sevenich Google, Gufeng Zhang Google, Amy Hua Google, Alexander Frömmgen Google, Inc. | ||
11:30 15mTalk | From Vibe to Spec: An Empirical Study to Demystify Specifications in Code Generation Research Papers Ruirui He Nanjing University, Hanqing Lu Nanjing University, Lezhi Ma Nanjing University, Kai Ye University of Hong Kong, Qiong Wu Shandong Normal University, Shangqing Liu Nanjing University, Lei Bu Nanjing University, Xuandong Li Nanjing University | ||
11:45 15mTalk | AEG-Edit: Bridging API Evolution and Code Semantics for Accurate API Usage in Code Generation Research Papers Yitong Lin Zhejiang University, Haoye Tian Aalto University, Dehai Zhao Zhejiang University, Chao Ni Zhejiang University, Xiaohu Yang Zhejiang University, Xiaoxue Ren Zhejiang University | ||
12:00 15mTalk | Effective and Efficient Context Retrieval via Partial Dependency Graph for Repository-Level Code Generation Research Papers Zhongxin Liu Zhejiang University, Zhonghao Jiang Zhejiang University, Zhifan Ye Zhejiang University, Haoye Wang Hangzhou City University, Jiakun Liu Harbin Institute of Technology, Xiaoxue Ren Zhejiang University | ||
12:15 15mTalk | Towards Fully Automated Medical Imaging Code Generation via Validation-Based Context Engineering Research Papers Zixiao Zhao University of Auckland, Jing Sun School of Computer Science, University of Auckland, Zhe Hou Griffith University, Cheng-Hao Cai Monash University, Qian Liu University of Auckland, Mengze Li University of Auckland, Zijian Zhang Beijing Institute of Technology, Jin Song Dong National University of Singapore | ||