[2P01-29] Poster session

[2P13] GPU-accelerated spectrum-adapted MAP-ECM
algorithm for high-throughput peak fitting of multiple spectra

*Tarojiro Matsumura¹ (1. National Institute of Advanced Industrial Science and Technology (AIST) (Japan))

This study presents a GPU-accelerated framework for high-throughput peak fitting of large-scale spectral datasets. The framework implements the spectrum-adapted MAP-ECM algorithm, which integrates B-spline-based nonlinear background subtraction, peak-number selection, and peak deconvolution within a unified procedure. By implementing objective-function evaluation and parameter-update procedures on the GPU, the method enables parallel fitting across many spectra while preserving the fitting process for each spectrum. The implementation can handle a lot of spectra without time-consuming manual tuning. Benchmark tests using 10,000 synthetic spectra demonstrated more than a tenfold improvement in computational efficiency compared with the conventional CPU implementation. The presentation also discusses the current implementation status and usability improvements toward practical application to experimental spectral data.

Password required to view