I joined Stephen McAleer, a UCI Computer Science PhD candidate, in working on the protein folding prediction problem which we intended to solve with Neural nets and reinforcement learning. My role included:
• Helping to find a good representation for a protein in 2D and 3D space and worked on implementing this in python.
• Building a class for the protein environment with methods for creating proteins, making protein folds, ensuring legal moves and methods to calculate or access all the other information which would be necessary for protein prediction.
• Creating a script to generate the first random training samples.
• Experimenting with various architectures for the Neural Net to see what could learn the protein structure well and implemented the designs for our policy/value network using Keras.
• Experimenting with various methods of exploration for reinforcement learning such as e-greedy, bayesian(using dropout), and Boltzmann.
