Foundation models for Science: A paradigm shift in AI

Go back to programme

Foundation models for Science: A paradigm shift in AI

  • Watch

    * Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay

    • Watch

      Register

      * Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay

    Foundation models represent a paradigm shift in artificial intelligence (AI), offering large-scale, pre-trained machine learning models that serve as adaptable starting points for various specialized applications. Unlike traditional models focused on specific tasks, these models leverage extensive pre-training on diverse datasets to recognize patterns across data, significantly enhancing their flexibility and efficiency for different scientific purposes. This method reduces the need for large, task-specific labeled datasets and lengthy training times, allowing a single foundation model to excel in numerous scientific tasks, often surpassing traditional models, especially in scenarios with limited labeled data. Successful implementation and advancement of foundation models in science necessitate collaborative efforts across interdisciplinary teams, drawing from different research groups, academic and governmental institutions, and technology companies. Emphasizing open science principles, such collaboration promotes transparency, reproducibility, and shared knowledge, enabling broader access to datasets, models, and fine-tuning code. This collective approach not only fosters the development of more effective and versatile foundation models but also ensures their broader acceptance and application within the scientific community, paving the way for accelerated scientific discovery and innovation. This joint presentation will give an overview of the “playbook” to develop Foundation Models for Science grounded in the principles described above. Examples of Foundation models already released as well as the ones being developed will also be presented. These models include the Large Language Model focussed on NASA Science Mission Directorate corpus, Prithvi – Geospatial HLS model as well as new Weather and Climate (WxC) FM. 

    Share this session