INDICO Agenda: https://indico.cern.ch/event/1345965/
In the last few years natural language processing and computer vision have experienced a fundamental shift in the way these fields use machine learning. Rather than training neural networks from a randomly initialized set of parameters researchers have often found superior performance can be achieved by fine-tuning a general pre-trained “foundation model” trained on vast amounts of diverse data – perhaps because this model comes with better “priors” than an untrained network. Polymathic AI[1] is a new research collaboration that aims to usher in the same shift in machine learning for scientific datasets. In this talk I will present the motivations behind the collaboration and describe the findings of our three new papers in this space which examine: better numerical encodings for large language models[2] contrastive embeddings for multi-modal scientific data[3] and building machine learning models that learn from multiple types of physics[4].
- https://polymathic-ai.org/
- https://arxiv.org/abs/2310.02989
- https://arxiv.org/abs/2310.03024
- https://arxiv.org/abs/2310.02994
Bio:
Miles Cranmer is Assistant Professor in Data Intensive Science DAMTP & Institute of Astronomy University of Cambridge.