Task relevant autoencoding enhances machine learning for human neuroscience.

Area of Science:

• Neuroscience
• Machine Learning
• Data Science

Background:

• Machine learning models in neuroscience often require large datasets, leading to overfitting with limited human neuroimaging data.
• Identifying behaviorally relevant neural patterns is crucial for understanding brain function.

Purpose of the Study:

• To develop a novel machine learning model, TRACE, for identifying behaviorally relevant neural patterns in human neuroimaging data.
• To evaluate TRACE's performance against standard models using simulated and real functional magnetic resonance imaging (fMRI) data.

Main Methods:

• Developed the Task-Relevant Autoencoder via Classifier Enhancement (TRACE) model.
• Benchmarked TRACE against a standard autoencoder and other models on truncated machine learning datasets.
• Evaluated TRACE on fMRI data from 59 subjects observing animals and objects.

Main Results:

• TRACE demonstrated superior performance compared to alternative models.
• Achieved up to 12% increase in classification accuracy.
• Showcased up to 56% improvement in discovering cleaner, task-relevant neural representations.

Conclusions:

• TRACE is a promising tool for analyzing human neuroimaging data, particularly when datasets are limited.
• The model effectively extracts behaviorally relevant neural patterns, overcoming limitations of traditional autoencoders.
• TRACE has broad applicability for various types of human behavior data.