Research

Overview

Data driven machine learning. Modeling and discovery of dynamical systems.

Foundation Models for Multiscale Digital Phenotypic Data Streams

Adapting and applying dynamical systems learning in foundation models for wearable sensors.
Structured Neural ODEs for Multistable Systems and Hysteresis

Learning asymptotically, multistable, dynamical systems and forward tracking control policies from trasient data.

fig 1: Learned control landscape for genetic toggle switch system from (Salas & King, 2026).
Sparse Identification of Nonlinear Dynamics

Developing a new model learning architecture inspired by the widely used (two-step) Sparse Identification of Nonlinear Dynamics (SINDy) that learns a model in scenarios of extremely scarce or noisy data measurements.

fig 2: Observations sampled at \(\Delta t=0.05\) until \(t=10\) with added Gaussian noise of variance \(\sigma^2=4\). Simulated dynamics from Lorenz 63 system are compared against true dynamics from \(t=10\) until \(t=15\). Image from (Hsu et al., 2025).
Reduced Order Modeling for Plume Dynamics

Developing fast reduced order modelings for characterizing plume dynamics directly from video data.

fig 3:Visualization of the plume processing pipeline. Image from (Salas et al., 2025).

Data-driven discovery and control of multistable nonlinear systems and hysteresis via structured Neural ODEs. 2026 [arXiv]
Ike Griss Salas, Ethan King

A joint optimization approach to identifying sparse dynamics using least squares kernel collocation. 2025 [arXiv]
Alexander W. Hsu, Ike Griss Salas, Jacob M. Stevens-Haas, J. Nathan Kutz, Aleksandr Aravkin, Bamdad Hosseini

Data-driven coarse-graining and reduced-order modeling of plume ejection dynamics from video observations. Physics of Fluids 37, 107145 (2025) [arXiv][Journal]
Ike Griss Salas, Megan R. Ebers, Jake Stevens-Haas, J. Nathan Kutz