SUPR
Robust training based on QP method
Dnr:

NAISS 2023/22-1330

Type:

NAISS Small Compute

Principal Investigator:

Shudian Zhao

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2023-12-11

End Date:

2025-01-01

Primary Classification:

10105: Computational Mathematics

Webpage:

Allocation

Abstract

This project focus on generating robust neural networks against adversarial attacks for image classifications. Instead of using large training data and heaving computing expense in classic training process. We introduce a optimisation based approach which only require a small size of adversarial dataset to improve the robustness a model trained with default settings.