Having won several notable awards, and speaking at several RE•WORK Summits, Ilya Sutskever has become a well-known name in the deep learning field. Since being announced as Research Director at OpenAI, a non-profit company formed with Elon Musk, Sam Altman and Greg Brockman, Ilya has been working on the goal to advance artificial intelligence that benefits humanity. Completing his PhD with the Machine Learning Group of the University of Toronto, working under Geoff Hinton, Ilya went on to co-found DNNresearch (acquired by Google) with Hinton and fellow graduate Alex Krizhevsky, as well as completing postdoctoral work at Stanford University with Andrew Ng's group. Until his appointment at OpenAI, he was a member of the Google Brain team, working as a research scientist.In his talk, at the Deep Learning Summit in San Francisco in January, Ilya will discuss his work on meta learning and self play, as well as surveying some of the exciting recent applications and research frontiers. Meta learning is the idea that learning systems can learn to learn fast and well. Self play systems are intriguing because they can lead to immensely complex behavior even in very simple environments. I will present several results in meta learning applied to different domains, self play results applied to simulated physics environments, and discuss the connection between the two. We caught up with him ahead of the summit to hear more.What are the key factors that have enabled recent advancements in deep learning?

  • Sufficiently fast computers
  • The availability of sufficiently large, high-quality labelled datasets
  • Algorithms, techniques, and skills for training large deep nets

What are the main types of problems now being addressed in the deep learning space?At present, large and deep neural networks are applied to a very large variety of problems. For example, there have been nearly 50 product launches within Google, all to different problems.What are the practical applications of your work and what sectors are most likely to be affected?The practical applications are vast, mainly because deep learning algorithms are largely domain-agnostic. Perception has already been affected. In the near future, I think that robotics, finance, medicine, and human-computer interaction are very likely to be affected. I don't think that this list is exhaustive, however.What developments can we expect to see in deep learning in the next 5 years?We should expect to see much deeper models, models that can learn from many fewer training cases compared to today's models, and substantial advances in unsupervised learning. We should expect to see even more accurate and useful speech and visual recognition systems.What advancements excite you most in the field?I am very excited by the recently introduced attention models, due to their simplicity and due to the fact that they work so well. Although these models are new, I have no doubt that they are here to stay, and that they will play a very important role in the future of deep learning.Ilya Sutskever will be speaking at the RE•WORK Deep Learning Summit in San Francisco, on 24 - 25 January 2018.