Na próxima segunda-feira, 02, às 14 horas, na sala 2077 do Instituto de Ciências Exatas, o professor da Faculdade de Computação da UFMS, Ronaldo A. Ferreira, irá proferir a palestra “When It Looks Too Good to Be True, It Usually Is: Unlocking Black-Box Models for Networking Problems with Trustee”. O evento é aberto ao público e não há necessidade de inscrição prévia.
Abstract:
In recent years, machine learning has demonstrated strong potential for detecting complex patterns in network traffic across a range of security and performance applications. Despite these advances, network operators remain hesitant to deploy such models in production environments due to their black-box nature. A central concern is underspecification, i.e., the difficulty of determining whether a model’s apparent success stems from capturing the essential structure of the underlying system or merely reflects inductive biases embedded during training.
In this talk, I will present Trustee, an eXplainable-AI (XAI) framework designed to help detect underspecification in machine-learning models for networking tasks. Given a trained black-box model and its associated training dataset, Trustee constructs a transparent surrogate in the form of a high-quality, interpretable decision tree, complemented by a comprehensive trust report. Through case studies based on published machine-learning models, I will show how we used Trustee to uncover and diagnose instances of model underspecification in peer-reviewed research.
Ronaldo A. Ferreira is a Full Professor of Computer Science at the College of Computing (Faculdade de Computação) at UFMS. He received his B.Sc. from UFMS in 1992, his M.S. from the University of Campinas in 1998, and his Ph.D. from Purdue University in 2006, all in Computer Science. From 2014 to 2016, he was a Visiting Research Scholar and Visiting Associate Professor at Princeton University. He served as Chair of the Special Interest Group on Computer Networks and Distributed Systems of the Brazilian Computer Society (SBC) from 2011 to 2013. During the same period, he was a member of the Board of Administration of the Brazilian National Research and Educational Network (RNP). From 2014 to 2023, he served on the Board of Directors of the Brazilian National Laboratory of Computer Networks (LARC). He is currently the Director of Cooperation with Scientific Societies of the SBC. His research has been published in top conferences and journals, including ACM SIGCOMM, ACM SIGMETRICS, ACM CCS, and USENIX ATC, among others.
