Large language models and audio transformers are known for mastering human speech and tackling complex tasks. But can they learn the grammar of spacetime (gravitational waves) and the elusive ragas of black holes?
I’m excited to share my lab’s latest breakthrough, now published in The Astrophysical Journal Letters (
@AAS_Office), where we adapted OpenAI's audio transformer to analyze gravitational-wave data from a rare black hole and learn signal structure directly from spacetime itself.
By treating gravitational-wave data like an audio “language,” we applied our in-house machine-learning architecture to listen directly to
@LIGO data. We found that GW231123 is indeed a rare “Lite” intermediate-mass black hole merger whose signals show hint of some unaccounted physics.
More broadly, this work demonstrates how foundation models developed for human communication can be repurposed to probe the laws of nature. This marks the beginning of a new era of gaining deep insights into the General Theory of Relativity with AI.
Big kudos to my team led by Dr. Chayan Chatterjee, PhD students Kaylah McGowan, Suyash Deshmukh, and master's student Nicholas-Tyler Howard for this breakthrough work.