She is particularly interested in building agents that can form and use abstractions from experience. Originally from Yerevan, she received her PhD from VU Brussel in 2017 and her MSc from Oregon State University in 2012.
Generalization in deep reinforcement learning
Modern reinforcement learning continues to deliver successes, in part due to harnessing the power of deep neural networks. Namely, they grant the agent the ability to generalize via the function approximator -- that is, to make inferences about what is unseen based on what is seen. Generalization is both the main challenge and the main appeal of deep reinforcement learning. Due to the RL problem being richer, generalization has many more facets than its analogue in supervised learning, in this talk we will overview this landscape.