Work

Placement-Aware Quantization for Fractional-PDE Neural Surrogates

Year: 2026

Type: article

Abstract: Research poster presented at the WiSys SPARK Symposium. The work examines component-level quantization sensitivity in Fourier Neural Operators (FNOs) applied to fractional partial differential equation surrogates. Results show that middle-core operator blocks tolerate simulated 8-bit quantization while projection-side layers require higher precision, motivating a placement-aware mixed-precision strategy to reduce compute overhead while preserving numerical accuracy and discrete PDE consistency.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.22044286

Author: Yoon, Chaewon

Institution: University of Wisconsin–Oshkosh

Language: English

Topic: Model Reduction and Neural Networks

Subfield: Statistical and Nonlinear Physics

Field: Physics and Astronomy

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7203859436

FWCI:

Cites:

Cited by:

Related to: