Advanced Mechanisms AI Research

Independent Research Team

AMAIR is at the forefront of combining Artificial Intelligence with Physicochemical Research to revolutionize our understanding of natural mechanisms and principles.

Our mission is to accelerate scientific discovery by leveraging cutting-edge AI technologies to explore, predict, and explain complex physical and chemical phenomena. We bridge the gap between computer science and natural science to uncover the fundamental laws governing our world.

Research Pillars

Discovery

Identifying hidden patterns and symmetries in complex physicochemical systems using advanced machine learning.

Principles

Uncovering fundamental principles across diverse domains, including biology, chemistry, and physics.

Bridging

Synthesizing computational rigor with experimental insights to create a new paradigm of AI-driven science.

We are always looking for passionate collaborators who are excited about the intersection of AI and natural sciences. If you are interested in joining our mission, please feel free to reach out via email.

News

May 01, 2026	Our PiEvo has been accepted by ICML 2026!
Mar 16, 2025	Our recent work PriM has been accepted to AI4Mat at ICLR 2025!

Selected Publications

ICML 2026
Principle-Evolvable Scientific Discovery via Uncertainty Minimization

Yingming Pu, Tao Lin^†, and Hongyu Chen

In arXiv, Feb 2026

Abs Bib PDF Code

Large Language Model (LLM)-based scientific agents have accelerated scientific discovery, yet they often suffer from significant inefficiencies due to adherence to fixed initial priors. Existing approaches predominantly operate within a static hypothesis space, which restricts the discovery of novel phenomena, resulting in computational waste when baseline theories fail. To address this, we propose shifting the focus from searching hypotheses to evolving the underlying scientific principles. We present PiEvo, a principle-evolvable framework that treats scientific discovery as Bayesian optimization over an expanding principle space. By integrating Information-Directed Hypothesis Selection via Gaussian Process and an anomaly-driven augmentation mechanism, PiEvo enables agents to autonomously refine their theoretical worldview. Evaluation across four benchmarks demonstrates that PiEvo (1) achieves an average solution quality of up to 90.81% 93.15%, representing a 29.7% 31.1% improvement over the state-of-the-art, (2) attains an 83.3% speedup in convergence step via significantly reduced sample complexity by optimizing the compact principle space, and (3) maintains robust performance across diverse scientific domains and LLM backbones.
(Remark: Principle-evolvable scientific discovery paradigm. )
@inproceedings{pu2025PiEvo, title = {Principle-Evolvable Scientific Discovery via Uncertainty Minimization}, author = {Pu, Yingming and Lin, Tao and Chen, Hongyu}, booktitle = {arXiv}, year = {2026}, month = feb, url = {https://arxiv.org/abs/2602.06448}, }