PhD Research Seminar: Mathematical simulation of protein molecule conformation, meta-gradient reinforcement learning
First talk: Mathematical Simulation of Protein Molecule Conformation
Speaker: Andrei Ignatov, second-year PhD student, Faculty of Computer Science
Protein Folding process has been examined by numerous researchers worldwide for the last 60 years. However, most of the current computational approaches do not provide results of enough accuracy and high application rate. In this research, a new complex method for predicting protein conformation is being developed.
The current stage of the research is focused on protein side-chain folding. Several algorithms are proposed, their strong and weak points are analyzed. The current results present a promising quality level. Directions of further improvement are outlined.
Second talk: Meta-Gradient Reinforcement Learning
Speaker: Alexander Grishin, third-year PhD student, Faculty of Computer Science
Recent advances in supervised and unsupervised learning have been driven by a transition from handcrafted expert features to deep representations; these are typically learned by gradient descent on a suitable objective function to adjust a rich parametric function approximator. As a field, reinforcement learning (RL) has also largely embraced the transition from handcrafting features to handcrafting objectives. Most of these algorithms differ fundamentally in their choice of objective, designed in each case by expert human knowledge. The deep RL version of these algorithms is otherwise very similar in essence: updating parameters via gradient descent on the corresponding objective function. One could instead learn its own objective, and hence its own deep reinforcement learning algorithm, solely from experience of interacting with its environment.