MatAgent: A human-in-the-loop multi-agent LLM framework for accelerating the material science discovery cycle
Submission Track: Findings & Open Challenges (Tiny Paper)
Submission Category: All of the above
Keywords: Multi-agent LLM, Material Property Prediction, Hypothesis Generation, Experimental Data Analysis, Alloy and Polymer Discovery, Data-driven Experimentation, Literature Review Automation
Abstract: The automation of materials science research through multi-agent large language models (LLMs) offers a transformative approach to accelerating discovery, optimizing experimentation, and enhancing data-driven decision-making. This study employs an LLM framework, called as MatAgent, across six key areas: material property prediction, hypothesis generation, experimental data analysis, high-performance alloy and polymer discovery, data-driven experimentation, and literature review automation. Machine learning models successfully predicted material properties, generated novel material hypotheses, analyzed experimental data, and optimized material compositions, significantly improving efficiency and accuracy. AI-driven methodologies enabled rapid screening of high-performance alloys and polymers, predictive modeling of concrete strength, and automated literature synthesis in perovskite solar cell research. The results demonstrate that MatAgent can revolutionize materials science by reducing research time, enhancing reproducibility, and paving the way for autonomous laboratories capable of AI-guided discovery and real-time adaptation. All corresponding codes and datasets related to this study are open-sourced in the GitHub repository available at https://github.com/adibgpt/MatAgent.
Submission Number: 53
Loading