New ideas and old friends

Written by Kris Brewer, photos by Tomotake Sasaki

 The annual Conference on Neural Information Processing Systems (NIPS), traditionally held in December, continues to grow since its inception in 1986. The 2016 conference had about 5000 registered participants and substantially grew to 8000 this past December, 2017 making it the largest conference on Artificial Intelligence in the world. Besides machine learning and neuroscience, other fields represented at NIPS include cognitive science, psychology, computer vision, statistical linguistics, and information theory.[1] As the numbers tell, this has become a very popular conference and a great opportunity for scientists in these fields to network and find synergies for collaboration.

Over the course of the week, several symposia occur simultaneously including Kinds of Intelligence: Types, Tests and Meeting the Needs of Society, which was co-organized by Tomaso Poggio, Director of the Center for Brains, Minds and Machines (CBMM). During this particular symposium, MIT’s professor Josh Tenenbaum (also of CBMM) offered "Types of intelligence: Why Human-like AI is Important" for the Mapping the Intelligence Landscape session. The other sessions comprised of The Intelligence Landscape, which contained a talk by Demis Hassabis (CBMM Advisor), and The Intelligence Landscape Society Needs. Each session included presentations from prominent scientists in their fields; including the aforementioned as well as Alison Gopnik, Gary Marcus, Katja Hofmann, Lucia Jacobs, Cynthia Dwork, and David Runciman, and concluded with a Panel/Q&A session of those presenters to merge their work and ideas into conversations the audience enjoyed. Videos and slides for most of this are now available on the symposium’s website - http://kindsofintelligence.org.

Andrés Campero, an MIT PhD candidate in Josh Tenenbaum’s lab, commented that “the conference was so exciting that the lure of the surrounding touristy activities and attractions weren’t even on my radar. I was having too much fun at the conference.” Campero continued stating that he was impressed with a workshop he attended, entitled Learning Disentangled Representations, and how presented research seemed to be going in a new direction combining deep learning and more symbolic approaches, which he personally believes is the direction to pursue for the advancement of AI. This workshop was co-organized in part by Josh Tenenbaum and Tejas Kulkarni (DeepMind and former CBMM graduate student), and included Yoshua Bengio, Finale Doshi-Velez, Ahmed Elgammel, Irina Higgins, Pushmeet Kohli, Doina Precup, Stefano Soatto and Doris Tsao on the list of speakers. Outside of the official conference proceedings, Campero discovered that the conversations did not end at the convention center doors. For example, in his hotel he found that folks from DeepMind were staying there as well, he engaged in a multitude of discussions. Not to mention the chance to reunite with former colleagues and friends over meals. (CBMM group photo by Andrea Tacchetti)

 Charlie Frogner, recent MIT PhD recipient from the Poggio Lab, conveyed the experience being similar to that of “drinking from a firehose, very intense in a good way.” He participated in the Optimal Transport and Machine Learning workshop presenting on Wasserstein Gradient Flows and was very well received. Frogner reminisced for the atmosphere of past NIPS conferences where they were smaller in size allowing for a more singular and community type of feel. The expansion of corporate sponsorship and private corporate parties has changed the landscape a bit, but he says it lends its own sense of excitement. Regardless, Frogner still revels in the opportunity to take part in NIPS as he states “it’s the world’s best machine learning researchers gathering to present their latest work.”