University of Guelph MLRG Deep Learning Tutorials

Overview

Date: July 12, 2018

Address: University of Iceland - Sæmundargata 4, 101 Reykjavík

Room: University Centre - Háskólatorg - HT-102

As a part of a collaborative research project, there will be a meeting in Iceland from the 9th to the 12th of July funded by the French Agence Nationale de la Recherche and the Canadian Natural Sciences and Engineering Research Council.

As a part of this meeting, the Machine Learning Research Group from the University of Guelph (Canada) has arranged a day of introductory talks and tutorials for participants from both industry and academia at the University of Iceland on Thursday, July 12th. We will conduct a series of talks, which will include hands-on tutorials on writing code in cutting-edge machine learning software frameworks.

The main objective of this day of tutorials is to give a comprehensive overview of the state of deep learning, as well as its limitations, with practical applications. Learning objectives include:

  • Understanding the building blocks of deep learning and its application across a wide range of domains.
  • Understanding fundamental challenges in computer vision and how deep learning can be used to solve various computer vision tasks.
  • Understanding the challenges involved in applying deep learning to problems in data science.

Several of the sessions will provide code examples that will be worked through interactively. We encourage participants to bring laptops so that they can follow along!

Schedule

Time Title Speaker
08:45 Arrivals/Registration
09:00 Welcoming Remarks
09:15 Introduction to AI & Machine Learning Graham Taylor
09:45 PyTorch I Terrance DeVries
10:30 Refreshments/Discussion
10:50 PyTorch II Adam Balint
11:35 Interpretability of Deep Learning Models Devinder Kumar
12:20 Lunch at STÚDENTAKJALLARINN
Menu
13:40 Transfer Learning Eu Wern Teh
14:25 Model Search Brendan Duke
15:10 Refreshments/Discussion
15:30 Generative Models / Reinforcement Learning / Discrete Optimization Thor Jonsson
16:15 Adversarial Examples/CleverHans Angus Galloway
17:00 Closing Remarks

Tutorial Environment

We have provided a preinstalled, GPU accelerated environment for use in our tutorials. The environment is based on JupyterHub and is mostly isolated, operations on the environment work only on copies, everyone is (mostly) independent. The environments are provided with full access to install additional software and based on Ubuntu 16.04.

Requirements

  • A computer with wireless internet access with a fairly modern browser.
  • A Github account

Usage

  • Tutorial participants will access https://jupyter.co60.ca which will guide them to login through Github.
  • Access to the environment is delegated to the https://github.com/mlrg-public-list organization
  • We will email you an invite to join this organization (you must have, or create a Github account). If you do not receive an invitation, please give us your Github username when you check-in at the start of the day.

Creating a server

  • If slots are available (which should be true during the 12th), you will be able to spawn a server by logging in
  • If you already have a server you can press "Stop Server" to free your server, pressing "Start my Server" will start a new one
  • If you receive an error creating your server press "Home" in the top left or "Control Panel" in the top right and press "My Server" or "Start my Server", errors should only occur if your server is stopping, or there are no slots remaining

Persistence

  • Your server is managed in a cluster of computers, in the event that your server goes does it will not be recoverable and a new server will need to be spawned
  • Your server may be reclaimed to allow a slot for another user if you are inactive during a particular session, data not stored in the persistent directory will be lost
  • Persistent data is shared in a read/writable drive global to all users.

Accessing Notebooks

  • Each session will use different directories within shared/ to help seperate relevant files, the tutorial leader will tell you which directories to enter to find the correct notebook

  • The directories with the tutorials files are read only. If you need to modify and save them you will need to move the files outside the directory

    • The only way to do this is to use a terminal

    • To open a terminal from the "Files" tab click the "New" dropdown and select "Terminal" from the terminal that displays you may copy with

      cp -r <source> <dest>

      • For example copying the Pytorch tutorials would be

        cp -r shared/CFIW-Tutorials/pytorch_tutorial .

Switching the Kernel

  • Due to conflicting libraries some tutorials use different Jupyter Kernels, to change Kernel simply open the Notebook you desire and click "Kernel" Dropdown > Change kernel" then select the Kernel that is required.
    • The base Kernel contains only the bare Python 3 install, for most purposes you will want to use "Python 3 - Pytorch" kernel which contains the most general libraries
    • As our lab primarly uses Python 3, we only support it for this tutorial

Sponsors













         

Machine Learning Research Group, 2018