Deep Learning Workshop with Dr. Robb Brown

This workshop will introduce you to the world of Neural Networks and Deep Learning.

Day 1 - total duration: 3 hours 14 minutes.

If videos don’t appear above, click here to watch them.

Day 1 - video segments

  • [Deep Learning D1V01] The idea behind deep learning
  • [Deep Learning D1V02] Perceptrons and processing layers
  • [Deep Learning D1V03] Applications of deep learning; image processing.
  • [Deep Learning D1V04] Character recognition; auto-encoders.
  • [Deep Learning D1V05] Finite-state machines; computation graphs; decision trees
  • [Deep Learning D1V06] Linear regression; nodes as interactions
  • [Deep Learning D1V08] Supervised v. unsupervised learning; sigmoid v. non-linear activation
  • [Deep Learning D1V09] Convolutional networks
  • [Deep Learning D1V10] Convolutional network example; max pool;
  • [Deep Learning D1V11] Convolutional network example (continued); filters
  • [Deep Learning D1V12] Comment: AI and ANN design
  • [Deep Learning D1V13] Optimization; non-convex optimization
  • [Deep Learning D1V14] Optimization (con’t): tricks of the trade, randomness, drop-out
  • [Deep Learning D1V15] Question: how to choose network size?
  • [Deep Learning D1V16] Manifold learning;
  • [Deep Learning D1V17] Distinguishing deep learning & machine learning, more broadly
  • [Deep Learning D1V18] Tensorflow - Google’s machine learning toolkit, and alternatives
  • [Deep Learning D1V19] Hands on: A complete deep learning example with TensorFlow
  • [Deep Learning D1V20] Hands on: softmax regression; stochasticity; …
  • [Deep Learning D1V21] Q&A on running simulations and training network on datasets
  • [Deep Learning D1V22] Over-fitting; generalization error; training data size;
  • [Deep Learning D1V23] Hands on with code: varying network size and parameters to test learning
  • [Deep Learning D1V24] More Q&A from the workshop
  • [Deep Learning D1V25] Hands on: working through image processing challenges and questions

Day 2 - total duration: 1 hours 37 minutes.

If videos don't appear above, [click here](https://vimeo.com/album/4450300/video/206467256) for access.

Day 2 - video segments

  • [Deep Learning D2V01] Fully convolutional ANN; neuroimaging application
  • [Deep Learning D2V02] Autoencoders;
  • [Deep Learning D2V03] Neuroimaging challenge; classifying brain images from open repositories.
  • [Deep Learning D2V04] Neuroimaging challenge (con’t); translational invariance;
  • [Deep Learning D2V05] Question: temporal and spatial encoding and learning
  • [Deep Learning D2V06] Next challenge: The Street View House Numbers Dataset

Workshop details

This workshop will introduce you to the world of Neural Networks and Deep Learning. There will be two sessions:

  • Session I: Thursday, January 19 2017, from 1-5 pm
  • Session II: Thursday, January 26 2017, from 2-6 pm

All of the workshop slides and materials are accessible on this github repository:

Sessions are divided into two halves, theory and practice. Participants are welcome to attend the first, theoretical half, and after a break, we will have a hands-on workshop session. Please bring a laptop computer! Although the principles are broadly applicable in this workshop, the practical sessions will use Google’s TensorFlow.

  • Session I discusses background theory for Machine Learning, graph computation, Artificial Neural Networks (ANNs) and Deep Learning. We will cover perceptron-type ANNs and touch on Convolutional Networks.

  • Session II covers more advanced convolutional and fully convolutional networks, and ends with a challenge problem that is similar to finding numbers written on a page or identifying street address markers from a passing Google car.

About Dr Robert Brown:

Dr. Brown has a BSc in computer science (2001) and a PhD in biomedical engineering (2008). His PhD thesis involved the development and application of the general Fourier family transform (GFT), including an efficient O(N log N) algorithm for calculating it. The Fourier transform, short time Fourier transform and S-transform are all special cases of the GFT. The wavelet transform is also closely related. Currently, Dr. Brown is a post doctoral fellow at the Montreal Neurological Institute in Montreal, Canada. His current research interests include texture analysis, magnetic transfer ratio MR imaging as it relates to multiple sclerosis, and compressive sensing applications in magnetic resonance imaging.

Location: Both sessions will take place at the de Grandpré Communications Centre, Montreal Neurological Institute, 3801 University Street.
Registration: Please email santiago.paiva@mcgill.ca with full name, affiliation, and which session(s) do you plan to attend.
Speaker: Dr. Robert Brown
Cost: Free

The videos were not working on my BIC workstation, so I found a direct link to the playlists:
Day 1
Day 2

which browser are you using? could you try it from another browser just to test it out?

My BIC workstation is on Ubuntu 12.04 and I tried both Chromium Version 37.0.2062.120 and Firefox 52.0. It works fine on my laptop (Ubuntu 14.04) where I have up to date browsers.

1 Like

Before Deep Learning which skill also we need to learn,
Can we directly start to learn Deep Learning.

Please give it a try and let us know if you were able to follow.
Cheers!