当前位置：
[an error occurred while processing this directive]
报告主题：Out-of-Distribution Detection through Transformation Rectified Activation for Reliable AI
报 告 人：Maozhen Li教授
时 间：2026年7月9日15:00-17：00
地 点：工科E1716
报告摘要：
Deep learning models often assume that the data encountered during deployment follow the same distribution as the training data. In real-world applications, however, models frequently face out-of-distribution (OOD) samples that differ significantly from the training distribution, leading to unreliable and overconfident predictions. Detecting such samples is therefore essential for building trustworthy and reliable AI systems. This talk presents Transformation Rectified Activation (TRA), a novel approach for OOD detection that improves the separation between in-distribution and OOD samples by rectifying neural activations through input transformations. The method leverages the observation that meaningful transformations reveal distinct activation patterns for familiar and unfamiliar inputs, enabling more reliable OOD detection without requiring access to OOD data during training. The presentation will introduce the motivation behind TRA, describe its underlying methodology, and discuss its effectiveness across standard OOD benchmark datasets. Experimental results demonstrate that TRA consistently enhances detection performance while maintaining compatibility with existing deep neural networks.
报告人简介：
Maozhen Li is a Professor of the Department of Engineering, Brunel University of London, UK. He received the Ph.D. degree from the Institute of Software, Chinese Academy of Sciences, Beijing, China, in 1997. He did his Post-Doctoral research in the Department of Computer Science at Cardiff University UK in 1999-2002. His main research interests include high-performance computing, big data analytics, and intelligent systems with applications to smart grids, smart manufacturing and smart cities. He has about 240 research publications in these areas, including 4 books. His book entitled “The Grid: Core Technologies” published by Wiley in 2005 had been well received by the research community. He is a Fellow of the British Computer Society (BCS) and the Institute of Engineering and Technology (IET). He has served over 30 IEEE conferences and is on the editorial board for a number of journals. His research work on Big Data Modelling on Large Scale Road Networks was shortlisted by Computing UK in May 2018 for BIG DATA EXCELLENCE AWARDS in the category of Most Innovative Big Data Solution.
青岛软件学院、计算机与科学与技术学院