Graph-based Predictable Feature Analysis

TL;DR

This paper introduces GPFA, a novel graph-based method for unsupervised learning of predictable features from high-dimensional time series, emphasizing low variance in future data given past observations.

Contribution

GPFA provides a new graph embedding approach for extracting predictable features, connecting predictability with information theory and demonstrating competitive performance.

Findings

GPFA effectively captures predictable features in high-dimensional data.

GPFA outperforms or matches existing methods like SFA, FCA, and PFA.

The method is validated on multiple datasets.

Abstract

We propose graph-based predictable feature analysis (GPFA), a new method for unsupervised learning of predictable features from high-dimensional time series, where high predictability is understood very generically as low variance in the distribution of the next data point given the previous ones. We show how this measure of predictability can be understood in terms of graph embedding as well as how it relates to the information-theoretic measure of predictive information in special cases. We confirm the effectiveness of GPFA on different datasets, comparing it to three existing algorithms with similar objectives---namely slow feature analysis, forecastable component analysis, and predictable feature analysis---to which GPFA shows very competitive results.