Beyin çözümlemesi için bilişsel görevlerin aktarılabilirliğinin grafik üretimi ile öğrenimi

Brain decoding involves analyzing the cognitive states of human brain by using some statistical techniques in order to understand the relations among the cognitive states, based on neuroimaging data. A very powerful tool to acquire the brain data is functional magnetic resonance images (fMRI), which generates three-dimensional brain volume at each time instant, while a subject performs a cognitive task involving social activities, emotion processing, game playing, memory etc. However, it is very difficult and time-consuming to acquire data, which is statistically sufficient to train a deep neural network. An alternative to generate data is to reuse the available statistically similar data, so that the information of the available data can be transferred to relatively small datasets. In this thesis, we propose a pipeline method to create and verify a graph, which shows the relations among well-defined cognitive tasks, based on fMRI data. We measure the affinity between the cognitive tasks on an fMRI dataset gathered from multiple subjects using the adaptation of an existing end-to-end method which compares transferring performance between the cognitive tasks using learned latent representations. However, due to high variance between measurements on different subjects, the results vary greatly. In order to both verify and regulate the differences, we use the performance of the binary classifiers trained on imbalanced data. In the last step, we generate a graph, where the edges are the possible relations between the cognitive tasks which have the potential to improve transfer learning in other datasets.