Year: 2024
Type: preprint
Abstract: We explore the impact of training with more diverse datasets, characterized by the number of unique samples, on the performance of self-supervised learning (SSL) under a fixed computational budget. Our findings consistently demonstrate that increasing pretraining data diversity enhances SSL performance, albeit only when the distribution distance to the downstream data is minimal. Notably, even with an exceptionally large pretraining data diversity achieved through methods like web crawling or diffusion-generated data, among other ways, the distribution shift remains a challenge. Our experiments are comprehensive with seven SSL methods using large-scale datasets such as ImageNet and YFCC100M amounting to over 200 GPU days. Code and trained models are available at https://github.com/hammoudhasan/DiversitySSL
Source: arXiv (Cornell University)
Authors: Hammoud, Hasan Abed Al Kader, Das, Tuhin, Pizzati, Fabio, Torr, Philip, Bibi, Adel, Ghanem, Bernard
Institution: University of Oxford
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Awards: G1526486803, G6505763180
OpenAlex ID: W4393063889