We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

q-bio.NC

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Biology > Neurons and Cognition

Title: "Task-relevant autoencoding" enhances machine learning for human neuroscience

Abstract: In human neuroscience, machine learning can help reveal lower-dimensional neural representations relevant to subjects' behavior. However, state-of-the-art models typically require large datasets to train, so are prone to overfitting on human neuroimaging data that often possess few samples but many input dimensions. Here, we capitalized on the fact that the features we seek in human neuroscience are precisely those relevant to subjects' behavior. We thus developed a Task-Relevant Autoencoder via Classifier Enhancement (TRACE), and tested its ability to extract behaviorally-relevant, separable representations compared to a standard autoencoder for two severely truncated machine learning datasets. We then evaluated both models on fMRI data where subjects observed animals and objects. TRACE outperformed both the autoencoder and raw inputs nearly unilaterally, showing up to 30% increased classification accuracy and up to threefold improvement in discovering "cleaner", task-relevant representations. These results showcase TRACE's potential for a wide variety of data related to human behavior.
Comments: 31 pages, 7 figures, 1 table
Subjects: Neurons and Cognition (q-bio.NC); Machine Learning (cs.LG)
Cite as: arXiv:2208.08478 [q-bio.NC]
  (or arXiv:2208.08478v1 [q-bio.NC] for this version)

Submission history

From: Megan Peters [view email]
[v1] Wed, 17 Aug 2022 18:44:39 GMT (2884kb)

Link back to: arXiv, form interface, contact.