Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

AttriLens-Mol: Attribute Guided Reinforcement Learning for Molecular Property Prediction with Large Language Models

Created by
  • Haebom

Author

Xuan Lin, Long Chen, Yile Wang

Outline

AttriLens-Mol is an attribute-based reinforcement learning framework for molecular feature prediction using large-scale language models (LLMs). This framework guides model inference through a formal reward that encourages feature-based structured outputs, a count reward that prevents listing irrelevant attributes, and a rationality reward that verifies the relevance of generated attributes. AttriLens-Mol effectively leverages the model's inherent knowledge of relevant molecular attributes to improve the performance of molecular feature prediction. Training the 7B-sized R1-Distilled-Qwen2.5 and R1-Distilled-LLaMA3.1 models on 4,000 samples yielded comparable or better results than supervised fine-tuning and advanced models, and the extracted attributes also outperformed an interpretable decision tree model.

Takeaways, Limitations

•
Takeaways:
◦
Improving the molecular property prediction performance of LLM using attribute-based reinforcement learning.
◦
Achieve equivalent or better performance compared to supervised learning models and advanced models.
◦
Improve the performance and predictability of interpretable models using extracted attributes.
•
Limitations:
◦
Specific Limitations is not explicitly mentioned in the paper (limited number of samples, dependence on specific models, etc.).
◦
Future research is needed to verify scalability on other models and datasets.
◦
The complexity of reinforcement learning frameworks.
👍
Made with Slashpage