Enhancing Differential Testing With LLMs For Testing Deep Learning Libraries
This program is tentative and subject to change.
Differential testing offers a promising strategy to alleviate the test oracle problem by comparing the test results between alternative implementations. However, existing differential testing techniques for deep learning (DL) libraries are limited by the key challenges of finding alternative implementations (called counterparts) for a given API and subsequently generating diverse test inputs. To address the two challenges, this paper introduces DLLens, an LLM-enhanced differential testing technique for DL libraries. The first challenge is addressed by an observation that DL libraries are commonly designed to support the computation of a similar set of DL algorithms. Therefore, the counterpart of a given API’s computation could be successfully synthesized through certain composition and adaptation of the APIs from another DL library. DLLens incorporates a novel counterpart synthesis workflow, leveraging a large language model (LLM) to search for valid counterparts for differential testing. To address the second challenge, DLLens incorporates a static analysis technique that extracts the path constraints from the implementations of a given API and its counterpart to guide diverse test input generation. The extraction is facilitated by LLM’s knowledge of the concerned DL library and its upstream libraries. DLLens incorporates validation mechanisms to manage the LLM’s hallucinations in counterpart synthesis and path constraint extraction. We evaluate DLLens on two popular DL libraries, TensorFlow and PyTorch. Our evaluation shows that DLLens synthesizes counterparts for 1.84 times as many APIs as those found by state-of-the-art techniques on these libraries. Moreover, under the same time budget, DLLens covers 7.23% more branches and detects 1.88 times as many bugs as state-of-the-art techniques on 200 randomly sampled APIs. DLLens has successfully detected 71 bugs in recent TensorFlow and PyTorch libraries. Among them, 59 are confirmed by developers, including 46 confirmed as previously unknown bugs, and 10 of these previously unknown bugs have been fixed in the latest version of TensorFlow and PyTorch.
This program is tentative and subject to change.
Mon 17 NovDisplayed time zone: Seoul change
14:00 - 15:30 | |||
14:00 10mTalk | Mokav: Execution-driven Differential Testing with LLMs Journal-First Track Khashayar Etemadi ETH Zurich, Bardia Mohammadi Sharif University of Technology, Zhendong Su ETH Zurich, Martin Monperrus KTH Royal Institute of Technology | ||
14:10 10mTalk | Validity-Preserving Delta Debugging via Generator Trace Reduction Journal-First Track Luyao Ren Peking University, Xing Zhang Peking University, Ziyue Hua Peking University, Yanyan Jiang Nanjing University, Xiao He Bytedance, Yingfei Xiong Peking University, Tao Xie Peking University | ||
14:20 10mTalk | Execution-Aware Program Reduction for WebAssembly via Record and Replay Research Papers Doehyun Baek University of Stuttgart, Daniel Lehmann Google, Germany, Ben L. Titzer Carnegie Mellon University, Sukyoung Ryu KAIST, Michael Pradel CISPA Helmholtz Center for Information Security | ||
14:30 10mTalk | DebCovDiff: Differential Testing of Coverage Measurement Tools on Real-World Projects Research Papers Wentao Zhang University of Illinois Urbana-Champaign, Jinghao Jia University of Illinois Urbana-Champaign, Erkai Yu University of Illinois Urbana-Champaign, Darko Marinov University of Illinois at Urbana-Champaign, Tianyin Xu University of Illinois at Urbana-Champaign Media Attached | ||
14:40 10mTalk | DRIFT: Debug-based Trace Inference for Firmware Testing Research Papers Changming Liu Northeastern University, Alejandro Mera Northeastern University, Meng Xu University of Waterloo, Engin Kirda Northeastern University | ||
14:50 10mTalk | Enhancing Differential Testing With LLMs For Testing Deep Learning Libraries Journal-First Track Meiziniu LI The Hong Kong University of Science and Technology, Dongze Li The Hong Kong University of Science and Technology, Jianmeng Liu The Hong Kong University of Science and Technology, Jialun Cao Hong Kong University of Science and Technology, Yongqiang Tian Monash University, Shing-Chi Cheung Hong Kong University of Science and Technology | ||
15:00 10mTalk | Unit Test Update through LLM-Driven Context Collection and Error-Type-Aware Refinement Research Papers Yuanhe Zhang Zhejiang University, Zhiquan Yang Zhejiang University, Shengyi Pan Zhejiang University, Zhongxin Liu Zhejiang University | ||
15:10 10mTalk | Metamorphic Testing for Audio Content Moderation Software Research Papers Wenxuan Wang Hong Kong University of Science and Technology, Yongjiang Wu The Chinese University of Hong Kong, Junyuan Zhang The Chinese University of Hong Kong, Shuqing Li The Chinese University of Hong Kong, Yun Peng The Chinese University of Hong Kong, Wenting Chen City University of Hong Kong, Shuai Wang Hong Kong University of Science and Technology, Michael Lyu The Chinese University of Hong Kong | ||
15:20 10mTalk | Comprehend, Imitate, and then Update: Unleashing the Power of LLMs in Test Suite Evolution Research Papers Tangzhi Xu Nanjing University, Jianhan Liu Nanjing University, Yuan Yao Nanjing University, Cong Li ETH Zurich, Feng Xu Nanjing University, Xiaoxing Ma Nanjing University | ||