Symbolic regression looks for concise, human readable mathematical expressions to describe functional dependences in data. Ultimately a very effective symbolic regression technique can be conceived of as a system for automatic discovery of physical laws, given suitable data. The current state of the art for symbolic regression is based on genetic algorithms, see e.g. PySR https://github.com/MilesCranmer/PySR
Will we see a major breakthrough in symbolic regression in terms of speed, ease of use, robustness to noise or other dimensions by Dec 31, 2025?
Resolution is subjective so I won't bet on this market.
Related questions
Maybe anchor to an AlphaFold-scale jump in competitions results? https://cavalab.org/srbench/competition-2022/
@AustinCary A grad thesis in symbolic regression that introduced a non trivial improvement would, as long as the result becomes well known enough that I take notice of it. I will be searching the arxiv for papers on symbolic regression at the end of 2025 to check whether any of them qualifies. Do you feel that more restrictive criteria would make the question more useful?