Re-Simulation-based Self-Supervised Learning for Pre-Training Foundation Models

arXiv:2403.07066v1 Announce Type: cross
Abstract: Self-Supervised Learning (SSL) is at the core of training modern large machine learning models, providing a scheme for learning powerful representations that can be used in a variety of downstream tasks. However, SSL strategies must be adapted to the type of training data and downstream tasks required. We propose RS3L, a novel simulation-based SSL strategy that employs a method of re-simulation to drive data augmentation for contrastive learning. By intervening in the middle of the …

abstract arxiv core cs.lg data foundation hep-ex hep-ph however machine machine learning machine learning models modern pre-training self-supervised learning simulation ssl strategies supervised learning tasks training training data type