AutoTransfer: Subject Transfer Learning with Censored Representations on Biosignals Data

Abstract

We investigate a regularization framework for subject transfer learning in which we train an encoder and classifier to minimize classification loss, subject to a penalty measuring independence between the latent representation and the subject label. We introduce three notions of independence and corresponding penalty terms using mutual information or divergence as a proxy for independence. For each penalty term, we provide several concrete estimation algorithms, using analytic methods as well as neural critic functions. We propose a hands-off strategy for applying this diverse family of regularization schemes to a new dataset, which we call "AutoTransfer". We evaluate the performance of these individual regularization strategies under our AutoTransfer framework on EEG, EMG, and ECoG datasets, showing that these approaches can improve subject transfer learning for challenging real-world datasets.

Reducing calibration effort for brain-computer interfaces

“…Previous work has investigated the use of censoring penalties in training variational autoencoders [160] and learning disentangled representations [161]. Other work has applied censoring penalties to enforce different notions of conditional independence, using estimation techniques such as kernel density estimation and neural critic functions [110]. Our work extends these approaches by providing a theoretical motivation for each censoring penalty and providing two new methods for estimating censoring penalties that are highly effective and simple to implement.…”

“…Subject Transfer Learning. Note that there are a variety of other approaches to the subject transfer learning problem, such as techniques for regularized pre-training [110], domain adaptation techniques based on Riemannian geometry [243], and weighted model ensembling using unsupervised similarity [244]. Techniques for subject transfer in EMG and related data types such as electroencephalography (EEG) are well-reviewed elsewhere [70], [78].…”