Learning to Discover at Test Time

Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed Mccaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:152432-152496, 2026.

Abstract

How can we use AI to discover a new state of the art for a scientific problem? Prior work in test-time scaling, such as AlphaEvolve, performs search by prompting a frozen LLM. We perform reinforcement learning at test time, so the LLM can continue to train, but now with experience specific to the test problem. This form of continual learning is quite special, because its goal is to produce one great solution rather than many good ones on average, and to solve this very problem rather than generalize to other problems. Therefore, our learning objective and search subroutine are designed to prioritize the most promising solutions. We call this method Test-Time Training to Discover (TTT-Discover). Following prior work, we focus on problems with continuous rewards. We report results for every problem we attempted, across mathematics, GPU kernel engineering, algorithm design, and biology. TTT-Discover sets the new state of the art in almost all of them: (i) Erdős’ minimum overlap problem and an autocorrelation inequality; (ii) a GPUMode kernel competition (up to 2$\times$ faster than prior art); (iii) past AtCoder algorithm competitions; and (iv) denoising problem in single-cell analysis. Our solutions are reviewed by experts or the organizers. All our results are achieved with an open model, OpenAI gpt-oss-120b, and can be reproduced with our publicly available code, in contrast to previous best results that required closed frontier models. Our test-time training runs are performed using Tinker, an API by Thinking Machines, with a cost of only a few hundred dollars per problem.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-yuksekgonul26a, title = {Learning to Discover at Test Time}, author = {Yuksekgonul, Mert and Koceja, Daniel and Li, Xinhao and Bianchi, Federico and Mccaleb, Jed and Wang, Xiaolong and Kautz, Jan and Choi, Yejin and Zou, James and Guestrin, Carlos and Sun, Yu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {152432--152496}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/yuksekgonul26a/yuksekgonul26a.pdf}, url = {https://proceedings.mlr.press/v306/yuksekgonul26a.html}, abstract = {How can we use AI to discover a new state of the art for a scientific problem? Prior work in test-time scaling, such as AlphaEvolve, performs search by prompting a frozen LLM. We perform reinforcement learning at test time, so the LLM can continue to train, but now with experience specific to the test problem. This form of continual learning is quite special, because its goal is to produce one great solution rather than many good ones on average, and to solve this very problem rather than generalize to other problems. Therefore, our learning objective and search subroutine are designed to prioritize the most promising solutions. We call this method Test-Time Training to Discover (TTT-Discover). Following prior work, we focus on problems with continuous rewards. We report results for every problem we attempted, across mathematics, GPU kernel engineering, algorithm design, and biology. TTT-Discover sets the new state of the art in almost all of them: (i) Erdős’ minimum overlap problem and an autocorrelation inequality; (ii) a GPUMode kernel competition (up to 2$\times$ faster than prior art); (iii) past AtCoder algorithm competitions; and (iv) denoising problem in single-cell analysis. Our solutions are reviewed by experts or the organizers. All our results are achieved with an open model, OpenAI gpt-oss-120b, and can be reproduced with our publicly available code, in contrast to previous best results that required closed frontier models. Our test-time training runs are performed using Tinker, an API by Thinking Machines, with a cost of only a few hundred dollars per problem.} }
Endnote
%0 Conference Paper %T Learning to Discover at Test Time %A Mert Yuksekgonul %A Daniel Koceja %A Xinhao Li %A Federico Bianchi %A Jed Mccaleb %A Xiaolong Wang %A Jan Kautz %A Yejin Choi %A James Zou %A Carlos Guestrin %A Yu Sun %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-yuksekgonul26a %I PMLR %P 152432--152496 %U https://proceedings.mlr.press/v306/yuksekgonul26a.html %V 306 %X How can we use AI to discover a new state of the art for a scientific problem? Prior work in test-time scaling, such as AlphaEvolve, performs search by prompting a frozen LLM. We perform reinforcement learning at test time, so the LLM can continue to train, but now with experience specific to the test problem. This form of continual learning is quite special, because its goal is to produce one great solution rather than many good ones on average, and to solve this very problem rather than generalize to other problems. Therefore, our learning objective and search subroutine are designed to prioritize the most promising solutions. We call this method Test-Time Training to Discover (TTT-Discover). Following prior work, we focus on problems with continuous rewards. We report results for every problem we attempted, across mathematics, GPU kernel engineering, algorithm design, and biology. TTT-Discover sets the new state of the art in almost all of them: (i) Erdős’ minimum overlap problem and an autocorrelation inequality; (ii) a GPUMode kernel competition (up to 2$\times$ faster than prior art); (iii) past AtCoder algorithm competitions; and (iv) denoising problem in single-cell analysis. Our solutions are reviewed by experts or the organizers. All our results are achieved with an open model, OpenAI gpt-oss-120b, and can be reproduced with our publicly available code, in contrast to previous best results that required closed frontier models. Our test-time training runs are performed using Tinker, an API by Thinking Machines, with a cost of only a few hundred dollars per problem.
APA
Yuksekgonul, M., Koceja, D., Li, X., Bianchi, F., Mccaleb, J., Wang, X., Kautz, J., Choi, Y., Zou, J., Guestrin, C. & Sun, Y.. (2026). Learning to Discover at Test Time. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:152432-152496 Available from https://proceedings.mlr.press/v306/yuksekgonul26a.html.

Related Material