François Fleuret from the Idiap research institute here in Martigny Switzerland recently organized a workshop on the tools and methods related to deep learning. What is deep learning? Yoshua Bengio gave a good introduction to this field during the first 10 minutes before diving into the scientific heart of it: talking about back propagation, gradients, derivatives, probabilities, artificial neural networks.
Yoshua Bengio giving an introduction of Deep Learning
At slide 9, Yoshua Bengio explains the same idea than in his interview published in Wired: “Intuition is just the part of the computation going on in our brain for which we don’t have conscious access. It’s really hard to decompose it into little pieces we can explain,” he says. “This is the reason why the traditional AI of the 80s and 70s failed – because it tried to build machines that could explain every single step through reasoning. It turns out it was impossible to do that. It’s much easier to train machines to develop intuitions to make the right decisions.” (by the way, Pete Blackshaw also talks about the importance of intuition at the slide 19 of his talk)
Another interesting quote from Yoshua Bengio related to the question whether Artificial Intelligence will replace knowledge workers over time: “I dream of a world where humans work not for getting a pay but because they are motivated by their contribution to society, because they enjoy their work.” More about it on Quora. We retrieve the same philosophy with Angelo Dalle Molle whose foundation founded the Idiap research institute back in 1991 when he wrote: «Scientific progress in general and progress in computer science in particular should not enslave man but on the contrary be at his service.»
All the talks are available on demand. You can play back presentations from S. Chintala at facebook about Torch, M. Rosca at Google about TensorFlow, A. Lowndes at NVidia and M. Breternitz at AMD.
More information:
- Idiap celebrates its 25 years. Idiap is the birth place of the software Torch originally written by Ronan Collobert, Samy Bengio and Johnny Mariethoz starting around 2001.
- In this article (in French), Yoshua Bengio talks about the difficulty to retain good students at his lab due to the salaries offered by large companies (Google, facebook) in this field.
- Webcasted lectures about neural networks at EPFL, webcast powered by Klewel back in 2011.
- Idiap speaker series, many of these talks are applications of machine learning.
- Related tweets during the event
- And by the way, we are starting a new CTI R&D project at Klewel together with Idiap (involving Jean-Marc Odobez, Florent Monay) and HES-SO Valais (involving Florian Evequoz, Maria Sokhn, Dominique Genoud) which will make use of Deep Learning with the objective of improving the visibility of webcasts
At Klewel we love to help you save the knowledge and gain more visibility by webcasting your events online. If you also want to strengthen the visibility of your event within your specific community, please contact us at +41 27 722 43 42.
