# Online Surrogate Repair: Decoupling High-Fidelity Feedback from Search Length in Closed-Loop Discovery

arXiv: https://arxiv.org/abs/2609.07655 · HTML: https://mlsanity.com/paper/2609.07655/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 49 references that are in the map.
- First posted on arXiv 7 Sep 2026.

## Builds on (most cited first, 20 of 20)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery](https://mlsanity.com/paper/2408.06292/) (2024)
- [Crystal Diffusion Variational Autoencoder for Periodic Material Generation](https://mlsanity.com/paper/2110.06197/) (2021)
- [TabArena: A Living Benchmark for Machine Learning on Tabular Data](https://mlsanity.com/paper/2506.16791/) (2025)
- [TabICLv2: A better, faster, scalable, and open tabular foundation model](https://mlsanity.com/paper/2602.11139/) (2026)
- [Conservative Objective Models for Effective Offline Model-Based Optimization](https://mlsanity.com/paper/2107.06882/) (2021)
- [Multi-Information Source Optimization](https://mlsanity.com/paper/1603.00389/) (2016)
- [LLMatDesign: Autonomous Materials Discovery with Large Language Models](https://mlsanity.com/paper/2406.13163/) (2024)
- [Agents for self-driving laboratories applied to quantum computing](https://mlsanity.com/paper/2412.07978/) (2024)
- [System of Agentic AI for the Discovery of Metal-Organic Frameworks](https://mlsanity.com/paper/2504.14110/) (2025)
- [MatSciBench: Benchmarking the Reasoning Ability of Large Language Models in Materials Science](https://mlsanity.com/paper/2510.12171/) (2025)
- [LLEMA: Evolutionary Search with LLMs for Multi-Objective Materials Discovery](https://mlsanity.com/paper/2510.22503/) (2025)
- [Benchmarking Active Learning Strategies for Materials Optimization and Discovery](https://mlsanity.com/paper/2204.05838/) (2022)
- [MOFGPT: Generative Design of Metal-Organic Frameworks using Language Models](https://mlsanity.com/paper/2506.00198/) (2025)
- [Graph Neural Network Interatomic Potential Ensembles with Calibrated Aleatoric and Epistemic Uncertainty on Energy and Forces](https://mlsanity.com/paper/2305.16325/) (2023)
- [Closing the Loop to Discover Psychological Theories with an Automated Cognitive Scientist](https://mlsanity.com/paper/2606.26448/) (2026)
- [MADE: Benchmark Environments for Closed-Loop Materials Discovery](https://mlsanity.com/paper/2601.20996/) (2026)
- [BLIPs: Bayesian Learned Interatomic Potentials](https://mlsanity.com/paper/2508.14022/) (2025)
- [Gemini: Dynamic Bias Correction for Autonomous Experimentation and Molecular Simulation](https://mlsanity.com/paper/2103.03391/) (2021)
- [Active In-Context Learning for Tabular Foundation Models](https://mlsanity.com/paper/2603.27385/) (2026)

## Abstract (arXiv)

Closed-loop AI scientists can generate candidate designs at low marginal computational cost, whereas reliable feedback may require wet-lab synthesis, characterization, or high-fidelity computation. Addressing this imbalance through custom laboratory automation remains infrastructure-intensive and costly, while replacing new experiments with a fixed surrogate leaves persistent model errors that can be amplified by optimization. We propose online surrogate repair (OSR), a closed-loop algorithm that uses sparse high-fidelity evaluations to update the surrogate throughout a longer agent search conducted primarily with inexpensive surrogate feedback. An acquisition rule selects which designs from the agent's accumulated proposals receive high-fidelity evaluation, and the resulting labels update the surrogate used in subsequent episodes. Across controlled synthetic environments, we demonstrate that improving global surrogate fit does not necessarily reduce maximum regret, whereas Q90-UCB and expected improvement (EI) substantially reduce regret by directing evaluations toward regions that determine the optimizer's decisions. On MADE, controls receiving high-fidelity feedback after every episode require 6.36--7.23× more oracle queries to match Online EI under two LLM orchestrators and 10.27× more under the non-LLM Chemeleon+MLIP workflow. Online surrogate repair introduces a novel third feedback regime between fixed-surrogate operation and high-fidelity feedback after every episode, separating the frequency of high-fidelity evaluation from the duration of the agent's search.