Building a Machine that Can Learn to Understand, Reason and Learn

  • Tsendsuren Munkhdalai | University of Massachusetts

My talk will first briefly review recent advances in memory augmented neural nets and then present my own contribution, Neural Semantic Encoders (NSE). With a special focus on NSE, I show that external memory in conjunction with attention mechanism are very useful in natural language understanding and reasoning. Particularly I will cover a set of real and large-scale NLP tasks ranging from sentence classification to seq2seq learning and question answering, and demonstrate how NSE is effectively applied to them. I am also very excited to share my latest work on meta learning, Meta Networks, that addresses major drawbacks of the neural nets relating to rapid generalization with small data and continual learning of new concepts without forgetting and to demonstrate its rapid generalization abilities on one-shot learning tasks. Lastly, I will discuss about my future research directions that I believe are crucial in developing an AI system that can reason, rapidly acquire different types of skills and knowledge, and broadly communicate with us.

Watch Next