Few-shot Classification with Novelty Detection using Meta-learning

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Few-shot Classification with Novelty Detection using Meta-learning"

by

WANG Haoqi

Abstract:

The current works of few-shot classification assume that a set of known 
classification categories are given. However, our visual world is 
obviously open and dynamic, and we could not ignore the possibility that 
novel classes could arise in the training and testing dataset. In this 
final year thesis, I present a modified MAML algorithm that achieves a 
higher precision score in detecting novel classes in few-shot 
classification tasks. The algorithm contains a higher dimension of the 
one-hot vector and can accommodate novel classes during training and 
testing.


Date            : 15 May 2020 (Friday)

Time            : 14:50 - 15:30

Zoom Meeting    : https://hkust.zoom.us/j/113843145

Advisor         : Prof. YEUNG Dit-Yan

2nd Reader      : Prof. CHEUNG Shing-Chi