TAMIDS Spring 2025 SciML Workshop

Workshop Description

The Spring 2025 TAMIDS SciML workshop will be an in-person one-day event in Rudder 302 on Thursday, April 17, from 8:30 am – 3:50 pm. The aim of this informal workshop is to showcase work by the Texas A&M community on scientific machine learning and to foster the formation of new collaborations. Areas of interest include:

• Physics-informed deep neural networks
• Physics-informed Gaussian processes
• Bayesian filtering and inference in physical models
• Data-driven model discovery through large-scale simulation
• ML-guided acceleration of numerical simulations
• Scientific and Engineering applications

Workshop Organization

The workshop is open to all members of Texas A&M. The workshop will comprise a keynote talk by TAMIDS SciML visitor Simo Särkkä (Aalto University, Finland) and short invited talks from Texas A&M speakers, which will consist of a 20-minute presentation plus 5-minute technical discussion. Two 30-minute coffee breaks and a lunch break are included, in order to foster networking and collaboration. For questions, please contact the workshop organizers: Ulisses Braga Neto (ulisses@tamu.edu) and Drew Casey (drew.casey@tamu.edu)

Registration deadline is April 14, 11:59 PM.

Workshop Schedule

8:30-9:00	Check-In and Coffee
9:00-9:10	Welcome and Introduction: Ulisses Braga-Neto (SciML Lab Director) and Drew Casey (TAMIDS Associate Director)
9:10-10:10	Keynote: Simo Särkkä, Aalto University Parallel filtering and smoothing methods for state-space models
10:15-10:40	Eduardo Gildin, Professor, Petroleum Engineering Applications of SciML in surrogate reservoir simulation models
10:40-11:05	Francesco A.B. Silva, Post-Doctoral Researcher, Nuclear Engineering A reduced-basis ensemble Kalman method
11:05-11:35	Coffee Break
11:35-12:00	Ulisses Braga-Neto, Professor, Electrical Engineering and SciML Lab Director Permutation-Invariant In-Context Learning for Solving PDEs
12:00-12:25	Jonathan Siegel, Assistant Professor, Mathematics Nearly Optimal Approximation Rates for Deep Sets
12:25-1:25	Lunch Break
1:25-1:50	Patricia Ning, Assistant Professor, Statistics Sequential Monte Carlo: Applications in Machine Learning and Large Language Models
1:50-2:15	Rui Tuo, Assistant Professor, Industrial Engineering Scalable Gaussian Process Regression with Kernel Packets
2:15-2:45	Coffee Break
2:45-3:10	James Cai, Professor, Veterinary Integrative Biosciences Single-Cell Biology, Machine Learning & Quantum Computing
3:10-3:35	Suparno Bhattacharyya, Post-Doctoral Researcher, Digital Twins Lab (TAMIDS) Uncertainty Quantification with Hyper-Reduced Order Models
3:35-3:45	Closing and Perspectives: Ulisses Braga-Neto (SciML Lab Director)

Keynote Talk

Parallel filtering and smoothing methods for state-space models

Abstract:
State space models (SSMs), including Gaussian state space models, non-linear/non-Gaussian state space models, and hidden Markov models (HMMs), are important tools in target tracking, time series analysis, machine learning, and various other fields. Bayesian filters and smoothers as well as their special cases such as Kalman filters and smoothers or approximations such as particle filters and smoothers are computationally optimal O(T) algorithms for state estimation in these models on classical CPU architectures. However, in parallel setting, such as on GPUs, they are no longer optimal, because they are inherently sequential algorithms. The aim of this talk is to discuss parallel versions of these algorithms. Most of the algorithms are based on so-called associative scans, which are computational primitives that can already be found, for example, in TensorFlow and JAX, and are easily implementable, for example, in Julia/CUDA.jl. These algorithms can be used to make state estimation optimally parallelizable leading to parallel O(log T) span complexity