ImprovEvolve: Ask AlphaEvolve to Improve the Input Solution and Then Improvise Conference: Under review for the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; Under review; Jeju, Korea CCS: Computing methodologies Neural networks CCS: Computing methodologies Evolutionary algorith
Alexey Kravatskiy, Valentin Khrulkov, Ivan Oseledets · 2026
Read ImprovEvolve: Ask AlphaEvolve to Improve the Input Solution and Then Improvise Conference: Under review for the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; Under review; Jeju, Korea CCS: Computing methodologies Neural networks CCS: Computing methodologies Evolutionary algorith by Alexey Kravatskiy, Valentin Khrulkov, Ivan Oseledets (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-10233.txt.
Description
AlphaEvolve-style LLM-guided evolution searches for novel mathematical constructions, but evolving one monolithic program that outputs a full solution strains the model. A 2026 paper led by Alexey Kravatskiy, with two co-authors, presents ImprovEvolve, which evolves three specialized operators instead: initialization, local improvement, and perturbation. Iterating local improvements with intensity-scheduled perturbations drives a basin-hopping search run by LLM-evolved routines.
In hexagon-in-hexagon packing, ImprovEvolve finds new state-of-the-art packings for eleven, twelve, fifteen, and sixteen hexagons, plus fourteen, seventeen, and twenty-three after minimal expert tuning. For the second autocorrelation inequality, the evolved program, scaled up by humans, slightly raises the lower bound. For spherical codes, it lowers the best-known maximum cosine for most of ninety diverse state-of-the-art codes, with relative gains up to about two and a half percent.
Catalog record
| CREATOR | Alexey Kravatskiy, Valentin Khrulkov, Ivan Oseledets |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10233