Decoding Brain Signals with Deep Learning

In House

Manager: William ZhU

Contact: williamzhu@g.ucla.edu

Concepts/skills:  programming (python, keras, tensorflow), Neural networks, deep learning

The inner workings of a brain is one of the greatest mysteries yet to be completely solved by science. Researchers need to make sense of an immense amount of brain signals in order to understand the mechanisms behind higher-order functions—such as perception, feelings, memory, or the sense of self—produced by this three-pound network of neurons. The neural networks used in deep learning, themselves inspired by the brain and designed to process large amounts of information, fit well with this task.

In the fall quarter, we will first gain experience with deep learning through practicing on simple image data sets with convolutional neural networks (CNNs). We will gain exposure to other types of neural net architecture as well, such as recurrent neural networks (RNNs) for time-domain signal analysis. Then, starting in the winter quarter, we will apply the skills and knowledge we have acquired to build a neural net that decodes behavioral and cognitive states from fMRI images (all data will be from public domain datasets, so we won’t work with actual fMRI machines. Sorry to disappoint you!). The exact research topic is subject to modification, depending on how everyone feels about coding or the ideas behind deep learning. At the very least, students will learn very important skills that are both handy in laboratories and sought-after in the job market, including Python, Keras, and TensorFlow.