Mathematical understanding of emergent behaviors within complex AI systems
Abstract
Research Summary / AbstractOur research plan for this proposed work is to apply the tools of Scientific Artificial Intelligence (SciAI) to enhance human understanding regarding the emergent behaviors that Transformer, large language models exhibit when displayingthe #sparks# of artificial general intelligence (AGI).We approach this problem from a deeply mathematical perspective, since humansunderstand mathematical formalism, and also since math is uniquely suited to characterizing complex systems (through operator learning, differential geometry, and topology, in the case of the proposed work.)The impact of this research will be novel insights into how, and why, Transformers are able to exhibit emergent behaviors, such as in-context learning: a vitally important, open question within the field of AI. Approved for Public Release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2023
- Source ID
- N000142312716
Entities
People
- Christopher Earls
Organizations
- Cornell University
- Office of Naval Research
- United States Navy