Semi-supervised out-of-distribution detection and fine-tuning method

Abstract

As the importance of data continues to grow, out of distribution detection is becoming increasingly crucial in various fields. In this thesis, we aim to address the escalating significance of anomaly detection within the context of realistic scenarios. Specifically, we focus on a methodology that utilizes a semi-supervised learning model for anomaly detection. Additionally, we explore approaches to enhance detection performance through fine-tuning when provided with a limited amount of out of distribution data.