EVIL: Evolving Interpretable Algorithms for Zero-Shot Inference on Event Sequences and Time Series with LLMs

TL;DR

EVIL uses LLM-guided evolutionary search to discover simple, interpretable algorithms for zero-shot inference on event sequences and time series, outperforming some deep learning models in speed and interpretability.

Contribution

First to demonstrate LLM-guided program evolution can find a single, interpretable inference algorithm for multiple dynamical systems tasks without dataset training.

Findings

Discovered algorithms generalize across datasets without per-dataset training.

Algorithms are often competitive with or outperform state-of-the-art deep models.

Evolved algorithms are faster, interpretable, and effective across three domains.

Abstract

We introduce EVIL (\textbf{EV}olving \textbf{I}nterpretable algorithms with \textbf{L}LMs), an approach that uses LLM-guided evolutionary search to discover simple, interpretable algorithms for dynamical systems inference. Rather than training neural networks on large datasets, EVIL evolves pure Python/NumPy programs that perform zero-shot, in-context inference across datasets. We apply EVIL to three distinct tasks: next-event prediction in temporal point processes, rate matrix estimation for Markov jump processes, and time series imputation. In each case, a single evolved algorithm generalizes across all evaluation datasets without per-dataset training (analogous to an amortized inference model). To the best of our knowledge, this is the first work to show that LLM-guided program evolution can discover a single compact inference function for these dynamical-systems problems. Across the three domains, the discovered algorithms are often competitive with, and even outperform, state-of-the-art deep learning models while being orders of magnitudes faster, and remaining fully interpretable.