Despite their often-discussed advantages, deep learning methods largely disregard theories of both learning and language. This makes their prediction behavior hard to understand and explain. In this talk, I will present a path toward more understandable (but still “deep”) natural language processing models, without sacrificing accuracy. Rational recurrences comprise a family of recurrent neural networks that obey a particular set of rules about how to calculate hidden states, and hence correspond to parallelized weighted finite-state pattern matching. Many recently introduced models turn out to be members of this family, and the weighted finite-state view lets us derive some new ones. I’ll introduce rational RNNs and present some of the ways we have used them in NLP. My collaborators on this work include Jesse Dodge, Hao Peng, Roy Schwartz, and Sam Thomson. ... Read More إقرأ المزيد | Share it now!

Supersizing and Empowering Visual and Robot Learning Abstract: In the last decade, we have made significant advances in the field of computer vision thanks to supervised learning. But this passive supervision of our models has now become our biggest bottleneck. In this talk, I will discuss our efforts towards scaling up and empowering visual learning. First, I will show how the amount of labeled data is a crucial factor in learning. I will then describe how we can overcome the passive supervision bottleneck by self-supervised learning. Next, I will discuss how embodiment is crucial for learning – our agents live in the physical world and need the ability to interact in the physical world. Towards this goal, I will finally present our efforts in large-scale learning of embodied agents in robotics. Finally, I will discuss how we can move from passive supervision to active exploration – the ability of agents to create their own training data. ... Read More إقرأ المزيد | Share it now!