Contrary to what people think, scientific progress is slow & continuous, but social & economic impact can be disruptive - there’s no doubt that we'll reach human level AI but we don’t know how long it’s going to take. We need to make sure that AI & society will bloom for the benefit of all. - Bengio
(PRWEB) October 26, 2017
With Montreal dubbed as the ‘Silicon Valley of Deep Learning’, and after discussing with Yoshua Bengio as well as the Montreal Tourism Board, RE•WORK’s decision to host the first Canadian Deep Learning Summit in Montreal was finalised.
The Panel of Pioneers featuring Bengio, LeCun and Hinton was moderated by the head of the new Facebook Lab in Montreal and Associate Professor of Computer Science at McGill University, Joelle Pineau.
As well as appearing on stage together, all three AI pioneers gave their own presentations where they shared their most cutting edge research. You can watch a preview of the panel discussion, and sign up to watch the pioneers solo presentations here.
Bengio shared some recent use cases of neural networks in deep learning with the use of rectifying nonlinearities (ReL), which enable training deeper networks as well as the use of soft content-based attention. This allows neural nets to go beyond vectors and to process a variety of data structures and led to a breakthrough in machine translation. He spoke about ongoing research that’s now suggesting that brains may use a process similar to backpropagation for estimating gradients and new inspiration from cognition suggests how to learn deep representations which disentangle the underlying factors of variation, by allowing agents to intervene in their environment and explore how to control some of its elements. He explained how “we have an internal model that allows us to generalize to completely new worlds... we need models that can predict the future and the consequences of their actions".
Following on from Bengio, Hinton presented on ‘Dynamic Routing with Capsules’, where he explained how ‘there is a major problem with current neural networks for object classification: they can’t extrapolate to novel orientations, scales, shears, pose changes - they cannot tell if 2 inputs are the same because they need to compare a vector of activities with a vector of weights’. Hinton’s radical proposition is to change the types of neurons we use and the way we assemble them to make networks with capsule routing.
The final thought leader to take to the stage was Yann LeCun who shared some of his recent work on teaching machines to learn as effectively as animals and humans. Currently, the brain learns with an efficiency that none of our machine learning methods can match, but LeCun is striving to unlock the next levels of AI that can build representations of the world and learn how to reason and plan. One of the principal distinctions between humans and machines is that of common sense, and LeCun explained that his intention is to ‘infer to DL models, a ‘common sense’ that simulates the human cognitive ability to predict from whatever information is available, even from partial percepts’ such as predicting under uncertainty and in unsupervised learning.
Additional leading minds at the RE•WORK event included Hugo Larochelle and Jasper Snoek from Google Brain, Raquel Urtasun from Uber, Ira Kemelmacher of Facebook, Eric Humphrey from Spotify, Doina Precup from McGill and many others. You can subscribe to the video hub here to watch presentations and panel discussions from the Deep Learning Summit in Montreal. The event focused on rapidly advancing technologies which will impact business and society. Over the two day summit, we learned about the latest advancements in deep learning technology and also ways in which the technology is being applied across various industries.
The next summits will take place in San Francisco on January 25 & 26. This event will be RE•WORK’s largest to date and will see the Deep Learning Summit and AI Assistants Summit hosting enterprise specific workshops exploring the applications and impact of AI and deep learning on both business efficiency and personal organisation. Confirmed speakers include Ian Goodfellow, Staff Research Scientist, Google Brain, Andrej Karapathy, Director of AI, Tesla, Rushin Shah,Engineering Leader, Facebook, Lionel Cordesses, Innovation Project Manager, Renault, and many more.