Scaling laws under different task complexities
Title: Scaling laws under different task complexities
DNr: Berzelius-2024-173
Project Type: LiU Berzelius
Principal Investigator: Kevin Smith <ksmith@kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2024-05-01 – 2024-11-01
Classification: 10207
Keywords:

Abstract

Our aim in this project is to investigate how deep neural networks perform when faced with increasing problem complexities in image classification tasks. Our study will address an overlooked question in deep learning literature, that is, how does the scaling laws change under different task complexities?