AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization
Mert Cemri, Shubham Agrawal, Akshat Gupta, Shu Liu, Audrey Cheng, Qiuyang Mang, Ashwin Naren, Lutfi Eren Erdogan, Koushik Sen, Matei Zaharia, Alex Dimakis, Ion Stoica · 2026
Read AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization by Mert Cemri, Shubham Agrawal, Akshat Gupta, Shu Liu, Audrey Cheng, Qiuyang Mang, Ashwin Naren, Lutfi Eren Erdogan, Koushik Sen, Matei Zaharia, Alex Dimakis, Ion Stoica (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-2602-20133.txt.
Description
Program generation is moving from one-shot output to search at inference time. Language models serve as semantic mutation tools inside evolution loops. Those loops still run on fixed schedules. Fixed schedules ignore how search dynamics change over time. Compute gets wasted on stalled populations while promising frontiers go neglected. A 2026 arXiv paper by Mert Cemri and eleven co-authors introduces AdaEvolve. It recasts language-model-driven evolution as layered tuning that adjusts.
A running gain signal unifies decisions across three levels. Local adaptation tunes search level inside each population of solution entries. Global routing sends the overall resource budget across populations through bandit-based planning. Meta-guidance generates novel solution tactics from earlier solutions. Together the three levels let the search follow where gains actually build, rather than where a fixed schedule sends it.
Catalog record
| CREATOR | Mert Cemri, Shubham Agrawal, Akshat Gupta, Shu Liu, Audrey Cheng, Qiuyang Mang, Ashwin Naren, Lutfi Eren Erdogan, Koushik Sen, Matei Zaharia, Alex Dimakis, Ion Stoica |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20133