Talk at the 9th Machine Learning and AI in (Bio)Chemical Engineering Conference, Cambridge
Can Large Language Models accelerate scientific model discovery? 🤖
This is one of the questions we’re exploring in our latest work on LLM-guided symbolic regression for kinetic model discovery, with Roberto Aliaga Medina and Ehecatl Antonio del Rio Chanona.
The idea is to combine the strengths of both: symbolic regression recovers interpretable equations from data, while the LLM brings in scientific knowledge to steer the search toward models that actually make physical sense.
Across four case studies, our framework:
- cut the number of new experiments needed to find the ground-truth model by 42–79% vs. standard symbolic regression
- had the LLM directly propose the correct model structure in half of the guided runs
- matched baseline predictive accuracy → so fewer experiments, no loss in quality
I presented this (ongoing) work at the 9th Machine Learning and AI in (Bio)Chemical Engineering Conference in Cambridge — thanks so much to the organisers for such a great event! 😊
Huge kudos to our brilliant Roberto Aliaga Medina, who has done all of this while still finishing his MSc in Chile and working with us in parallel!
Dr Paulina Quintanilla
Principal Investigator
Assistant Professor in Process Systems Engineering at UCL, leading the LOOPS research group. Research interests include machine learning, optimisation and control for processes and physical systems.