AutoAdapt: On the Application of AutoML for Parameter-Efficient Fine-Tuning of Pre-Trained Code Models
This program is tentative and subject to change.
Large Language Models (LLMs) have demonstrated their ability to solve tasks across various domains, including software engineering. However, their extensive number of parameters makes full fine-tuning computationally prohibitive. While Parameter-efficient fine-tuning (PEFT) methods, such as adapter fine-tuning, have been proposed to address this issue; yet, they typically employ default configurations that use the same adapter settings across all layers. Concurrently, Automated Machine Learning (AutoML) has demonstrated success in hyperparameter optimization, while Neural Architecture Search (NAS) has proven effective in optimizing neural network architectures. Building on these successes, we introduce AutoAdapt, a novel approach that leverages NAS to automatically discover task-specific, layer-wide adapter configurations, allowing each layer to adopt distinct adapter parameters. AutoAdapt defines a search space tailored for adapter-based fine-tuning and employs an evolutionary algorithm to explore a diverse range of configurations, thereby evaluating the benefits of customizing each layer individually. We evaluate AutoAdapt on well-established software engineering tasks, including vulnerability detection, code clone detection, and code search. Our empirical results demonstrate that AutoAdapt outperforms manually engineered adapter configurations, achieving up to a 5% improvement in F1-score for clone detection and defect detection, and up to a 25% improvement in MRR for code search. Additionally, it surpasses other PEFT techniques, such as Prefix Tuning and LoRA. Furthermore, AutoAdapt is capable of identifying configurations that outperform even full fine-tuning, while training less than 2.5% of the model parameters. A comprehensive analysis reveals that factors such as selective layer adaptation, module selection (e.g., attention versus feed-forward layers), normalization, and dropout significantly influence performance across different tasks. Additionally, our findings suggest the possibility of transferring adapter configurations to similar datasets and tasks, thus simplifying the search for optimal PEFT settings. Our code and data are available for access at: https://github.com/serval-uni-lu/AutoAdapt
This program is tentative and subject to change.
Mon 17 NovDisplayed time zone: Seoul change
11:00 - 12:30 | |||
11:00 10mTalk | TensorGuard: Gradient-Based Model Fingerprinting for LLM Similarity Detection and Family Classification Research Papers Zehao Wu Huazhong University of Science and Technology, Yanjie Zhao Huazhong University of Science and Technology, Haoyu Wang Huazhong University of Science and Technology | ||
11:10 10mTalk | Root Cause Analysis of RISC-V Build Failures via LLM and MCTS Reasoning Research Papers Weipeng Shuai Institute of Software, Chinese Academy of Sciences, Jie Liu Institute of Software, Chinese Academy of Sciences, Zhirou Ma Institute of Software, Chinese Academy of Sciences, Liangyi Kang Institute of Software, Chinese Academy of Sciences, Zehua Wang Institute of Software, Chinese Academy of Sciences, Shuai Wang Institute of Software, Chinese Academy of Sciences, Dan Ye Institute of Software at Chinese Academy of Sciences, Hui Li , Wei Wang Institute of Software at Chinese Academy of Sciences, Jiaxin Zhu Institute of Software at Chinese Academy of Sciences | ||
11:20 10mTalk | An Empirical Study of Knowledge Transfer in AI Pair Programming Research Papers Alisa Carla Welter Saarland University, Niklas Schneider Saarland University, Tobias Dick Saarland University, Kallistos Weis Saarland University, Christof Tinnes Saarland University, Marvin Wyrich Saarland University, Sven Apel Saarland University | ||
11:30 10mTalk | Efficient Understanding of Machine Learning Model Mispredictions Research Papers Martin Eberlein Humboldt-Universtität zu Berlin, Jürgen Cito TU Wien, Lars Grunske Humboldt-Universität zu Berlin | ||
11:40 10mTalk | Can Mamba Be Better? An Experimental Evaluation of Mamba in Code Intelligence Research Papers Shuo Liu City University of Hong Kong, Jacky Keung City University of Hong Kong, Zhen Yang Shandong University, Zhenyu Mao City University of Hong Kong, Yicheng Sun City University of Hong Kong | ||
11:50 10mTalk | "My productivity is boosted, but ..." Demystifying Users’ Perception on AI Coding Assistants Research Papers | ||
12:00 10mTalk | HFUZZER: Testing Large Language Models for Package Hallucinations via Phrase-based Fuzzing Research Papers Yukai Zhao , Menghan Wu Zhejiang University, Xing Hu Zhejiang University, Xin Xia Zhejiang University | ||
12:10 10mTalk | Provable Fairness Repair for Deep Neural Networks Research Papers Jianan Ma Hangzhou Dianzi University, China; Zhejiang University, Hangzhou, China, Jingyi Wang Zhejiang University, Qi Xuan Zhejiang University of Technology; Binjiang Institute of Artificial Intelligence, Zhen Wang Hangzhou Dianzi University, China | ||
12:20 10mTalk | AutoAdapt: On the Application of AutoML for Parameter-Efficient Fine-Tuning of Pre-Trained Code Models Journal-First Track Amal Akli University of Luxembourg, Maxime Cordy University of Luxembourg, Luxembourg, Mike Papadakis University of Luxembourg, Yves Le Traon University of Luxembourg, Luxembourg | ||