Luke Mahoney (MLabs)
project owner
LLM + Genetic Optimizing Multi Objective Functions
MOSES can solve genetic programming problems as well as optimize the search towards a solution, being able to find patterns within search spaces, which it can then exploit. However, it requires a single-objective scoring function, even for problems that prefer, or indeed need, multiple objectives. Constructing such a function is a non-trivial task with unpredictable effects on the search.
We will use LLMs to combine multiple objective functions into a single objective function, using the circuit synthesis problem as a test bed. We will demonstrate the capabilities of this approach, first on general EAs, then using MOSES specifically, evaluating behavior on two different problem instances.
Join the Discussion (0)
Please create account or login to post comments.