CliffSearch: Structured Agentic Evolution for Scientific Algorithm Discovery

Abstract

Scientific algorithm discovery is iterative: hypotheses are proposed, implemented, stress-tested, and revised. Current LLM-guided search systems accelerate proposal generation, but often under-represent scientific structure by optimizing code-only artifacts with weak correctness/originality gating. We present CliffSearch, an agentic evolutionary framework in which the core evolution operators (pair selection, crossover, mutation, and review) are implemented as LLM agents, and the loop is designed around three principles: (1) each node is a structured scientific artifact carried in either theory+code or code-only mode, (2) reviewer judgments of correctness and originality are first-class selection gates alongside optimization of the benchmark metric of interest, and (3) mutation is split into exploration and correction pathways with distinct objectives. Exploration mutation imports ideas from adjacent scientific domains to increase novelty, while correction mutation performs targeted evidence-guided repair using reviewer signals over theory, code, benchmark results, and runtime errors. We illustrate the framework on optimizer discovery over a fixed transformer training stack, transformer hyper-connection evolution, and native optimizer discovery on fixed classification/MLP benchmarks, where the same loop supports benchmark-grounded comparison with explicit metric direction and reproducible persistence. The result is a discovery workflow that prioritizes scientific interpretability and correctness while optimizing task metrics under controlled novelty constraints, rather than maximizing candidate throughput alone.

What Is On This Site

Preprint Preview (Full PDF)

Citation

@article{cliffsearch2026preprint,
  title   = {CliffSearch: Structured Agentic Co-Evolution over Theory and Code for Scientific Algorithm Discovery},
  author  = {Youssef Mroueh and Carlos Fonseca and Brian Belgodere and David Cox},
  journal = {arXiv preprint arXiv:XXXX.XXXXX},
  year    = {2026},
  url     = {https://arxiv.org/abs/XXXX.XXXXX}
}