报告题目：Software Engineering for Machine Learning-based Systems

报告时间：2025年9月11日17:30

报告地点：计算机学院E202会议室

报告人：Luciano Baresi

报告人国籍：意大利

报告人单位：米兰理工大学

报告人简介：

Luciano Baresi is a full professor at the Politecnico di Milano, with the Dipartimento di Elettronica, Informazione e Bioingegneria, where he also earned his laurea (master) degree and PhD in computer science. He has held visiting positions at the University of Oregon (USA), Tongji University (China), and the University of Paderborn (Germany).

Luciano has served as program chair for ICECCS, FASE, ICWE, ICSOC, SEAMS, ESEC/FSE, and SCC, and was general chair for WICSA/CompArch and SEAMS. He is currently the editor-in-chief of Proceedings of the ACM on Software Engineering, senior associate editor for ACM Transactions on Autonomous and Adaptive Systems, and an editorial board member for Science of Computer Programming, IEEE Software, International Journal of Cooperative Information Systems, and Requirements Engineering.

Luciano has co-authored over 200 papers and a book (in italian). His research spans a wide range of topics in software engineering. Early in his career, he focused on formal modeling and specification languages, later transitioning to UML and the design of web applications. His current interests include self-adaptive systems, dynamic software architectures, edge computing, and AI/ML-based systems.

报告摘要：

This talk would like to discuss how software engineering can help realize better ML systems. These systems are imposing software engineers to rethink all the activities of the usual development process. The ML framework is often given, but one has first to reason on how to select and tailor the model behind the system and then the quality of the results it produces. This means that requirements must (also) focus on new aspects, and systems must be designed, implemented, deployed, tested, provisioned, and maintained in a peculiar way. Training and inherence require different frameworks, pose different challenges, and demand for appropriate solutions. For example, how can one generate appropriate test cases to test an autonomous driving system? It is also true that ML experts focus mainly on ML algorithms and tend to underestimate the problems of designing complete, scalable, and robust systems, and of deploying and operating them properly. The aim of this talk is to frame the different contributions, highlight the mutual benefits and possible problems, and focus on a couple of solutions we developed for a thorough presentation and discussion. The identification of some future directions and opportunities will conclude the talk.

邀请人：金芝、玄跻峰