Two co-first-authored papers accepted at NeurIPS 2025 workshops, both focused on making AI practical for materials and experimental discovery.

AI4Mat Workshop: Deep Tree of Research

We introduce a long-horizon deep-research agent with a Deep Tree of Research (DToR) orchestrator. The system adaptively expands and prunes research branches to improve coverage, depth, and coherence while remaining fully deployable on local, consumer-level hardware with on-premises RAG.

Across 27 nanomaterials/device topics, DToR reports outperformed commercial deep-research systems (including ChatGPT-5-thinking, o3, and o4-mini-high Deep Research), achieving approximately 79% mean win rate in pairwise evaluation and leading in dry-lab validations.

• Workshop paper
• arXiv preprint

ML4PS Workshop: Neuromorphic Random Walk

We apply an unsupervised neuromorphic random walk to noisy real-world phosphate-adsorption experiments. The model autonomously segments dynamics into regimes consistent with adsorption theory, demonstrating a practical approach toward energy-efficient, noise-resilient scientific sensing.

• Workshop paper

Collaborators: Rodrigo Ferreira, Rapti Ghosh, Haihui Pu. Advisors: Dr. Junhong Chen and Dr. Yuxin Chen.