Out of Distribution Detection in Deep Generative Models

As generative models become increasingly prominent in machine learning, the need for accurately detecting out-of-distribution data has become crucial. The primary objective of this research is to develop an approach that can identify when the program encounters data that is vastly different from what it was trained on. In machine learning, programs may make errors when they encounter data that is dissimilar to what they have learned. To tackle this issue, we will investigate various techniques utilizing deep generative models to help the program comprehend what types of data it should expect to encounter. However, even the most sophisticated deep generative models may occasionally mistake new data as similar to old data, leading to inaccurate predictions. Therefore, we aim to investigate the underlying reasons for this phenomenon and explore potential solutions to address this issue.

Faculty Supervisor:

Rahul G. Krishnan

Student:

Partner:

Layer 6 AI

Discipline:

Computer science

Sector:

Artificial Intelligence; Information and Communications Technology; Technology

University:

University of Toronto

Program: