Trip Report: ICML 2019


By: Bita Hasheminezhad

A few weeks ago, I had the opportunity to attend the International Conference on Machine Learning (ICML) which is the premier gathering of professionals dedicated to the advancement of machine learning. Thirty-Sixth ICML was held on June 9th to 15th at the Long Beach Convention Center.

It was my first time to be at ICML and it definitely exceeded my expectations. There were about 8,000 individuals at this conference, so as you might guess it is not the largest conference, however,  it is the one I gained the most from. There were 4 sessions of talks every day and in each session we could choose between a large variety of categories: convex optimization, generative adversarial networks, time series, reinforcement learning, supervised learning, kernel methods, information theory and many more. Invited talks were about state-of-the-art methods used to tackle problems in theory and in practice. Unlike other conferences, the poster sessions were crowded and useful. What excited me the most was the diverse atmosphere and attendees of the ICML.

Image of dinner with friends.
Meeting new friends and catching up with the old!

There were a few topics that I was excited to learn about. “Challenging common assumptions in the unsupervised learning of the disentangled representation” is the title of the best paper at 36th ICML. It was fascinating to see learning struggles represented by an explanatory factorization of an image and why well-disentangled models seemingly cannot be identified without supervision.  Although probabilistic prediction techniques are not new approaches, the way they are applied as active learning to applications are certainly a new trend. I really enjoyed the talk on “active learning for probabilistic structured prediction of cuts and matching” by Sima Behour.

Machine learning for robots to think fast was absolutely a well-deserved invited talk at the most prestigious machine learning conference in the world. As a Ph.D. student with an electrical engineering background, majoring in control systems, I realize that the foremost concern in controlling systems is the response time. Many techniques are not practical when a real-time response is needed, but a pre-trained model was essentially helpful as shown by Aude Billard. If you are amazed by seeing robotic arms catch a ball in real-time, do not miss this video: https://www.youtube.com/watch?v=l9UFsRAM_XM

ICML is a great venue to meet new people in the field, make new friends, as well as, visit old ones. During the course of  research in a field, you learn of several popular luminaries. Don’t be surprised to put a face on some of those names!

Image of Yoshua Bengio and me.
Meeting Yoshua Bengio

“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain” (Alan Turing) was the opening quote of another amazing invited talk on what 4 year-olds can do and AI can’t. Studying how children use inference is a novel way to approach the design of machine learning algorithms.

This year I was a recipient of the ICML D&I travel grant. This grant which covered a part of my costs made it possible for me to be a part of the conference. I would like to thank my supervisor, Dr. Kaiser, for his travel support and guidance. I am fortunate to have had this great opportunity to interact with interesting people and to be inspired by several great ideas, which would not have happened without their support.

Image of the Women in Machine Learning Lunch
Lunch with WIML (Women in Machine Learning)



Leave a Reply

Your email address will not be published. Required fields are marked *