TRANSFER LEARNING FOR BRAIN DECODING

Date

2023-2-17

Author

Eryol, Erkin

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

240
views

781
downloads

Cite This

Understanding the human brain is a long-standing challenge in science. In this thesis, we focus on the brain decoding problem, where we estimate a cognitive state from functional magnetic resonance imaging (fMRI) images, to uncover the mechanisms in the brain-behavior relationship. However, due to the costly data acquisition process, fMRI studies are generally performed with a limited number of subjects in an experiment. Furthermore, the indirectly taken measurements introduce difficulties in the analysis of brain mechanisms. With the increase in the available brain decoding datasets in recent years, transfer learning methods become applicable on brain decoding studies in neuroscience domain. In this thesis, we utilize the available data and knowledge in the neuroscience domain to improve the performance of a different but related brain decoding study, that we refer as transfer learning for brain decoding. We suggest two approaches on transfer learning for brain decoding. In the first approach, we propose a novel Structured Multi-Layer Perceptron, utilizing a brain atlas. We observe that the Structured MLP model trained only on the target dataset has on-par classification and convergence time performance with the three dimensional convolutional neural network model, that is pre-trained on a large source dataset. In the second approach, we work on transfer learning between small-scale datasets that follows a common experimental paradigm. We propose Hierarchical Group PCA and its supervised variant for transferable feature generation that regards the session, subject and dataset relations. In the experiments, both methods outperform the state-of-the-art method, steadily on all transfer learning cases.

Subject Keywords

fMRI, Canonical Correlation Analysis, Multi-Layer Perceptron, Transfer Learning, Feature Alignment, Brain Decoding

URI

https://hdl.handle.net/11511/103066

Collections

Graduate School of Natural and Applied Sciences, Thesis