Data Science student at the University of Wisconsin-Madison.
I study efficient and reliable machine learning, especially what happens when models are pushed under constraints on precision, memory, or compute. My current work looks at where approximation causes important failures and how to preserve or recover the computations that matter without giving up all of the efficiency gains.
Current interests: Efficient ML · Model Compression · Low-Precision Inference · Adaptive Computation · Efficient Inference · ML Systems
Conditional Precision Bottlenecks and Quantized Defect Correction in Fractional Neural Operators
QDC studies how aggressive low-precision inference affects different parts of a neural operator and whether the resulting failures can be corrected selectively rather than by restoring the entire model to high precision.
The project compares activation and weight sensitivity under quantization, uses residual information to identify important defects, applies selective correction under fixed computational budgets, and includes numerical validation and an A100 W8A8 inference evaluation.
Topics: Quantization · Model Compression · Selective Correction · Neural Operators · GPU Inference
Repository · Preprint · Software DOI
The broader question I want to pursue is:
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?
My current QDC work studies this question in neural operators. I am interested in extending the same general problem to model and representation compression, adaptive precision, memory-efficient inference, and larger neural architectures.
Evidence-state interfaces for bounded human-AI visual inspection.
ACM UIST Adjunct 2026 Poster Publication
Repository · Paper · ACM 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 research and engineering projects are available under the Repositories tab.
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
Email: chaewon.yoon.ds@gmail.com