Test-time memory evolution improves scientific discovery efficiency and accuracy
Can Test-Time Memory Evolution Outperform Test-Time Training for Scientific Discovery?
Artificial IntelligenceMachine Learning
Summary
When trying to discover new things with large language models, updating the model during use can be slow and expensive. The authors introduce a new way called Test-Time Memory Evolution (TTME) that learns from experience without changing the model itself. TTME remembers what it learned and uses that memory to get better results faster. Their tests show TTME is more accurate and much quicker than previous methods on math challenges.
What this means in practice
- •For machine learning engineers: Improve online adaptation of discovery models to reduce runtime and computational costs without retraining model weights.
- •For software developers in scientific computing: Incorporate TTME to accelerate mathematical problem-solving tools while maintaining or improving accuracy on test-time tasks.
Authors
Yepeng Liu, Shengmai Chen, Xuandong Zhao, Chengzhi Liu, Dawn Song, Xin Eric Wang, Yuheng Bu
Abstract
Recent \emph{test-time training} (TTT) frameworks have enabled open-weight large language models (LLMs) to outperform inference-only frameworks powered by frontier models on scientific discovery tasks. However, TTT introduces additional computational cost for online weight updates during the discovery process, making computational efficiency an important consideration when scaling to larger models. This motivates us to explore an efficient \emph{test-time adaptation} method for scientific discovery. We introduce \emph{Test-Time Memory Evolution} (TTME), a plug-and-play test-time adaptation method, as an alternative to TTT. TTME continually updates the external memory by distilling the experience to guide subsequent discovery without updating model weights. To have a clean comparison, we evaluate TTME and TTT under the same discovery framework, replacing only the test-time adaptation method. Our evaluation shows that TTME outperforms TTT on all five mathematical tasks, while reducing computational cost by $82.8$-$84.6\%$ in FLOPs and $47.7$-$58.4\%$ in runtime. These results establish TTME as an effective and computationally efficient alternative to TTT in the evaluated discovery settings.