Neuron-based Multifractal Analysis of Neuron Interaction Dynamics in Large Models
NeuroMFA characterizes the multifractal organization of AI weight networks across training checkpoints.
DUAL
Discovery and Understanding with AI Lab
Forward ModelingandInverse Design:Quantifying structure and dynamics across scales, and guiding design.AI for ScienceandScience for AI:Understanding the world with AI, and understanding AI itself.
Research highlights
We develop mathematical and AI methods to investigate multiscale structure and multiscale dynamics: how to quantify them, what we can predict, and how they shape function. These questions also inform our growing work in inverse design, for scientific systems and AI models themselves.
Includes the PI's earlier work with collaborators at USC and other institutions.
Multiscale structure
- Multifractal analysis
- Network science
- Multiscale representations
Quantify network organization and heterogeneity across scales, and investigate their relation to function.
Spiking dynamics of individual neurons reflect changes in the structure and function of neuronal networks
Multifractal spiking statistics connect neuronal dynamics to network structure and function.
Deciphering the Generating Rules and Functionalities of Complex Networks
NMFA uses node-centered box growth to characterize multiscale network structure and compare its complexity and heterogeneity.
Multiscale dynamics
- Neural operators
- Physics-informed AI
- Multiwavelets
Analyze and predict evolving fields, coupled dynamics, and systems with memory.
GeoWind2Plan: Mission-Time 3D Urban Wind Prediction for Energy-Efficient UAV Planning
GeoWind2Plan predicts local 3D urban wind with a geometry-conditioned neural operator for energy-efficient UAV planning, with planned routes evaluated under high-fidelity CFD wind.
Coupled Multiwavelet Operator Learning for Coupled Differential Equations
CMWNO learns coupled solution operators through interacting kernels in multiwavelet space.
Multiwavelet-based Operator Learning for Differential Equations
MWT learns PDE solution operators by representing kernels in multiwavelet bases across scales.
Inverse design & AI design
- Network-informed materials
- Neuro-inspired AI
- Optimal control
From network-informed materials selection and neuro-inspired AI models toward physics-constrained inverse design.
Neuro-Inspired Information-Theoretic Hierarchical Perception for Multimodal Learning
ITHP combines a neuro-inspired processing hierarchy with information bottlenecks for multimodal learning.
Biomorphic Structural Batteries for Robotics
Graph and multifractal analysis compare nanofiber networks for structural-battery selection; physical tests assess mechanical performance and ion transport.
News
More newsCongratulations to James on his selection for a UTK Scholarly Development Grant - Research Assistantship (SDG-RA)!
GeoWind2Plan has been accepted as a Spotlight at NeurIPS 2026. Congratulations to Shaoxiang, Yucheng, and our collaborators!
Our collaborative review of complex nanoparticle systems, from synthesis and self-assembly to graph theory and applications, is published in Chemical Reviews.
Toward a Structure-to-Function Science of Foundation Models is accepted as a visionary paper at the ACM AI Leadership Summit 2026. Congratulations to Yucheng and Shaoxiang!
AI Tennessee supports our research on AI for materials.
Our group receives support from NSF COMPASS, with Xiongye Xiao leading the research at UTK as Principal Investigator.
Our group
People
Xiongye Xiao
Principal Investigator
Yucheng Zhao
PhD Student
Yuhang Xie
PhD Student
Tom Olesch
PhD Student (co-advised by Dr. Fei Liu)
Md Enamul Atiq
PhD Student
Shaoxiang Qin
Visiting Scholar; PhD Student at McGill University
James Garrick
Undergraduate Research Assistant
John (JC) Squires
Undergraduate Research Assistant
Sejal Verma
Undergraduate Researcher
Join our group
We welcome students and collaborators interested in AI methods for scientific systems and in the study and design of AI models.
Research support
U.S. National Science FoundationThrough COMPASS, an NSF Science and Technology Center
AI TennesseeUniversity of Tennessee, Knoxville
Our PI is also a member of UTK's Foundational AI cluster.
We gratefully acknowledge support from AI Tennessee. Our COMPASS research is supported by the U.S. National Science Foundation under Award No. 2243104. Any opinions, findings, and conclusions or recommendations expressed here are those of the authors and do not necessarily reflect the views of the U.S. National Science Foundation.