一、报告时间
2023年6月19日(周一)9:00-11:30
二、报告地点
电航楼219
三、主讲人
谢肖飞博士,新加坡管理大学助理教授。谢博士于2018年在天津大学获得博士学位,并获得了中国CCF优秀博士论文奖(2019)。他的研究主要集中在传统软件和AI软件的质量保证上,在软件工程、安全和AI领域顶级会议/期刊如ICSE、ESEC/FSE、ISSTA、ASE、TSE、TOSEM、ICLR、NeurIPS、ICML、TPAMI、Usenix Security和CCS 上发表多篇论文,并获得了三个ACM SIGSOFT杰出论文奖(FSE'16、ASE'19和ISSTA'22)。
四、内容简介
在过去数十年,基于学习的软件应用在人脸识别、自动驾驶和内容生成等多个领域已经展示了其巨大的潜力。软件的发展从传统的基于代码的程序扩展到了AI驱动的软件(也称为智能软件)。然而,与传统的软件一样,智能软件也可能表现出不正确的行为,从而导致严重的事故和损失。因此,智能软件的质量和安全性是非常重要的。相比传统软件,智能软件的“黑盒”特性使在分析和解释其行为时带来了重大挑战。本次演讲将从传统的软件分析到基于深度学习模型的软件分析展开系统的介绍,在此基础上为给定的软件(例如代码或深度神经网络)构建抽象模型。基于该模型,我们可以进行全面的分析、测试、故障定位和自动化修复,以提高软件的质量和安全性。
Abstract: Over the past decade, the application of learning-based software in various domains, such as face recognition, autonomous driving, and content generation, has shown tremendous potential. The evolution of software has led to a diverse landscape, ranging from traditional code-based programs to AI-driven software (a.k.a., intelligent software). However, like traditional software, intelligent software can exhibit incorrect behaviors, which may result in severe accidents and losses. Ensuring the quality and security of software, particularly in safety- and security-critical scenarios, is of utmost importance. However, the black-box nature of intelligent software poses significant challenges in analyzing and explaining its behaviors. In this talk, I will present the model-based analysis from traditional software to deep learning-based software. Our approach involves constructing an abstract model for a given software (e.g., code or a deep neural network). Based on this model, we can perform comprehensive analysis, testing, fault localization, and automated repair to enhance its quality and security.
信息科学技术学院
2023年6月9日