Toward Weakly-supervised Anomaly Detection

2024年6月3日 8:30 - 10:00

信管学院308会议室

Minqi Jiang (江敏祺) is currently a fourth-year PhD candidate at Shanghai University of Finance and Economics. He mainly focuses on Anomaly Detection, Time-series Forecasting and Quantitative Investment. His works are published in top-tier conferences like NeurIPS and KDD.

Toward Weakly-supervised Anomaly Detection

Anomaly Detection (AD) is the task of identifying unusual instances that deviate significantly from the majority of data. Recent studies propose to learn valuable distinguishing features from a few labeled anomalies that may be identified by domain experts in practice, which is termed as the weakly-supervised AD. This talk provides several interesting topics to introduce our recent works in weakly-supervised AD, from large-scale anomaly detection benchmark, to weakly-supervised AD algorithm improvement, and finally the automated design choices for deep anomaly detection. All of these works provide valuable research references for algorithm analysis, improvement, and automated model construction in weakly-supervised AD scenarios.