Contextually Supervised Source Separation with Application to Energy Disaggregation

We propose a new framework for single-channel source separation that lies between the fully supervised and unsupervised setting. Instead of supervision, we provide input features for each source signal and use convex methods to estimate the correlations between these features and the unobserved signal decomposition. Contextually supervised source separation is a natural fit for domains with large amounts of data but no explicit supervision; our motivating application is energy disaggregation of hourly smart meter data (the separation of whole-home power signals into different energy uses). Here contextual supervision allows us to provide itemized energy usage for thousands homes, a task previously impossible due to the need for specialized data collection hardware. On smaller datasets which include labels, we demonstrate that contextual supervision improves significantly over a reasonable baseline and existing unsupervised methods for source separation. Finally, we analyze the case of `2 loss theoretically and show that recovery of the signal components depends only on cross-correlation between features for different signals, not on correlations between features for the same signal.

Authors

• Matt Wytock Carnegie Mellon University
• J. Kolter

Keywords

No keywords are indexed for this paper.