Deep Learning: an Interview with Yoshua Bengio
Yoshua Bengio is a renowned figure in the machine learning and specifically deep learning, here is an interview with Yoshua about his thoughts on media interest in the field, future developments and more.
Yoshua Bengio is a renowned figure in the deep learning field. His titles include Full Professor of the Department of Computer Science & Operations Research at the Université de Montréal, head of the Machine Learning Laboratory (MILA), CIFAR Program Co-director of the CIFAR Neural Computation and Adaptive Perception program, and Canada Research Chair in Statistical Learning Algorithms, among many others.
His current interests are centered around a quest for artificial intelligence, through machine learning, and include fundamental questions on deep learning and representation learning, the geometry of generalization in high-dimensional spaces, manifold learning, biologically inspired learning algorithms, and challenging applications of statistical machine learning.
As part of our ongoing Deep Learning Q&A series, I caught up with Yoshua to hear his thoughts on media interest in the field, future developments and more, ahead of his presentation at the RE•WORK Deep Learning Summit in Boston this May.
What do you feel are the leading factors enabling recent advancements and uptake of deep learning?
Research results have greatly improved, due also to improved hardware.
What are your thoughts on the recent surge of media interest surrounding deep learning?
It is fueled by the above progress and the impressive potential for transformative effects on society and business. There are good sides and bad sides. The good side is that it is helps to attract strong researchers (especially students) and funding. The bad side is that journalists tend to exaggerate the progress that any particular paper is making and to ignore all the important research done by a myriad of less known researchers.
Do you think this is beneficial to the field?
Overall, yes.
How can larger corporations working on deep learning ensure that their work benefits others within this field?
By participating in fundamental research in the area, which is done by funding academic research and by establishing their own internal research groups in which researchers publish openly both their papers and their code, go to conferences and engage in an open dialogue with the rest of the scientific community.
Can you tell us more about your collaborative work with IBM Watson?
IBM has been an early player in deep learning (initially for speech recognition) and is moving fast on other areas of application that also interest me, such as natural language understanding. They are also interested in progress in basic science (such as training procedures) and hardware (which will soon become even more important, I believe).
What present or potential future applications of deep learning excite you most?
Natural language understanding.
Which industries do you feel will be most disrupted by deep learning in the future?
Everywhere humans interact with systems, companies, machines, robots. That includes self-driving cars but also personal assistants, search engines, operating systems, customer service, etc. Computer vision will also continue to expand rapidly into products, going beyond industrial vision to applications based on your hand-held devices that can see what is going on and use that to help you, for example in health.
What developments can we expect to see in deep learning in the next 5 years?
I don’t have a crystal ball, but major challenges include improving unsupervised (or semi-supervised) learning, bringing in modelling of causal dependencies, natural language understanding, reasoning, etc.
Yoshua Bengio will be speaking at the RE•WORK Deep Learning Summit in Boston, on 12-13 May 2016. Other speakers include Joseph Durham, Amazon Robotics; Alejandro Jaimes, AiCure; Katherine Gorman, Talking Machines; Hugo Larochelle, Twitter, and more.
The Deep Learning Summit is taking place alongside the Connected Home Summit. For more information and to register, please visit the event website here.
Related:
