سخنرانی کلیدی مراسم افتتاحیه

Robust Anomaly Detection

دکتر محمدحسین رهبان

عضو هیئت علمی دانشکده مهندسی کامپیوتر دانشگاه صنعتی شریف

پنج‌شنبه، ۴ دی، ساعت ۲:۳۰ الی ۳ بعدازظهر

سالن آمفی‌تئاتر دانشکده مهندسی کامپیوتر (سالن دکتر ربیعی)

Abstract

Anomaly detection deals with identifying data samples that deviate from the normal class defined through samples in a given dataset. In this talk, I will go through several such techniques and discuss our recently proposed research. We are motivated by adversarial robustness of such methods. This relates to how vulnerable these methods are against noises that are designed to change the model output once added to the input. I will talk on how making anomaly detection robust is different from regular classification.