Data Science · University of Wisconsin–Madison
My research focuses on efficient and reliable machine learning, particularly the effects of reduced numerical precision and computational constraints on model behavior. I study how to identify the computations most sensitive to approximation and preserve them without sacrificing the efficiency gained elsewhere.
Research interests: Efficient Machine Learning · Quantization · Model Compression · Adaptive Computation · Neural Operators · ML Systems
My research is guided by a central question:
How far can we reduce the computational cost of machine-learning systems without losing the internal computations or representations that are important for their capabilities?
I approach this question through numerical precision, selective computation, and the structure of approximation errors. My current research examines these problems in neural operators and inverse-vision models, with broader interests in efficient inference, representation compression, and larger neural architectures.
The Measurement Operator Changes Which Bits Matter: Operator-Conditioned Mixed Precision for Inverse Vision
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027), Algorithms Track. Accepted.
This paper studies how the sensing operator influences precision sensitivity in inverse-vision reconstruction. It introduces Operator-Susceptibility Matching (OSM), which uses calibrated operator information to select mixed-precision schedules without exhaustive per-operator search.
Proceedings forthcoming.
EvidenceLens
ACM Symposium on User Interface Software and Technology (UIST Adjunct 2026), Poster Publication.
Evidence-state interfaces for bounded human-AI visual inspection.
Repository · Paper · ACM DOI
Conditional Precision Bottlenecks and Quantized Defect Correction in Fractional Neural Operators
QDC investigates how quantization affects different components of neural operators and whether numerical defects can be corrected selectively under fixed computational budgets.
The project studies weight and activation sensitivity, physics-based residual diagnostics, and selective precision correction. Its evaluations include fractional-PDE neural operators and GPU inference under low-precision arithmetic.
Repository · Preprint · Software DOI
Risk-aware active evidence acquisition under workflow, resource, safety, and review-capacity constraints.
Controlled generation and auditable state tracking for multi-turn structured interfaces.
Additional projects are available in my GitHub repositories.
WiSys SPARK Symposium 2026
Placement-Aware Quantization for Fractional-PDE Neural Surrogates
DOI
Pi Mu Epsilon Regional Undergraduate Mathematics Conference 2025
Numerical Stability and Convergence of Physics-Constrained Neural Networks in Vibration Modeling
DOI