A robust principal component analysis for outlier identification in messy microcalorimeter data

TL;DR

This paper explores a robust PCA method called coherence pursuit to automatically identify outliers in microcalorimeter data, enhancing non-linear analysis of sensor pulses for large spectrometers.

Contribution

It introduces the application of coherence pursuit robust PCA for outlier detection in microcalorimeter pulse data, enabling scalable and automated analysis.

Findings

Coherence pursuit effectively identifies outliers in microcalorimeter data.

Robust PCA improves the quality of pulse analysis in large sensor arrays.

The method is computationally efficient and suitable for real-time applications.

Abstract

A principal component analysis (PCA) of clean microcalorimeter pulse records can be a first step beyond statistically optimal linear filtering of pulses towards a fully non-linear analysis. For PCA to be practical on spectrometers with hundreds of sensors, an automated identification of clean pulses is required. Robust forms of PCA are the subject of active research in machine learning. We examine a version known as coherence pursuit that is simple, fast, and well matched to the automatic identification of outlier records, as needed for microcalorimeter pulse analysis.