CA$49 – CA$349

Premium Hands-on Workshop: Deep Learning Without Labels

Event Information

Share this event

Date and Time

Refund Policy

Refund Policy

Contact the organizer to request a refund.

Eventbrite's fee is nonrefundable.

Event description
Learning Without Labels: Unsupervised and Weakly Supervised Learning of Deep Models

About this Event

Workshop Overview

Whilst deep learning has been successfully and widely adopted in many fields, a major bottleneck of these approaches is curating labels for training. In this workshop we will cover solutions that can be leveraged when little, or even no labels are available, along with hands-on examples of using generative models, clustering approaches, custom loss functions etc. Participants will learn how these techniques work in practice, what we sacrifice and gain when no labels are used compared to full supervision, and when it is appropriate to ditch labels altogether.

Important Dates

Early Bird Deadline: November 14, 2019

Refund Deadline: November 15, 2019

Sales end on: November 27, 2019

Please note that this workshop will happen on 3 separate evenings:

Session 1: November 28, 2019

Session 2: December 5, 2019

Session 3: December 12, 2019

"Why should I care about Unsupervised and Weakly-supervised methods?"

Traditionally, training deep models requires a large amount of labeled data which is not often available; However unsupervised and weakly supervised learning can offer simple and effective solutions to overcome this problem, allowing you to use the neural networks with some clever additions.

“How can I attend?”

Join in-person or online

Watch and rewatch the video recordings as often as you need during the 3 weeks of the workshop, and interact with other participants and TAs on our Slack

Need more information?

Send us any questions you might have (, or

Join us for a Q&A session for an opportunity to hear from the instructor and ask them questions about the workshop:

Target Audience

Data Scientists, Machine Learning Engineers, Students, AI Researchers


  • Very comfortable with Python, and familiar with Pytorch
  • Environment: google colab
  • Theoretical knowledge assumed: It is expected that the participants are already familiar with the basic neural network, concept of training of batches and the role of a loss function. I will use mostly CNNs for examples and therefore a basic understanding of CNNs is expected.

Learning Outcomes

The concepts that will be explained during the workshop are designed to make the most out of the data that is available, whilst making some basic mathematical assumptions to train efficiently. Specifically, you will be learning to:

  • Identify *when* you can use deep models if no/little training labels are available
  • Know which unsupervised/weakly supervised models are appropriate when presented with a problem, and understanding the tradeoff between supervised and unsupervised learning.
  • Build and train a deep learning model with unlabeled data using PyTorch.

Pre-workshop reading material

  • “Deep Learning” book by Ian Goodfellow, Yoshua Bengio and Aaron Courville. Free chapters:
  • Basics in PyTorch. (Try to tackle the “Training a classifier” tutorial)
  • The Blurry Lines of Supervised and Unsupervised Learning
  • The Next Frontier in AI: Unsupervised Learning (Yann LeCun)

Learning Material

All online and in-person participants will have access to the following learning material:

  • Slides from the sessions
  • Hands on notebooks
  • Video recording of the sessions (you can use the videos to watch the parts that you missed, or re-watch any parts that are still unclear for you; access to videos beyond one week after the workshop is available to be purchased; see tickets >> add-ons)


Shazia Akbar

Machine Learning Engineer, Altis Labs

Dr. Shazia Akbar is a machine learning engineer at Altis Labs, a startup focused on building intelligent models and tools for healthcare in the field of oncology. She recently completed her postdoctoral training at the Department of Medical Biophysics, University of Toronto, and was also an affiliate of the Vector Institute. She specializes in modifying and applying machine learning techniques to medical images, and holds additional interests in weakly supervised learning and attention models. Shazia received her PhD from the University of Dundee, U.K., in 2015, soon after joining the Department of Radiology at New York University, U.S., before moving to Canada in 2016.

Course Modules

The workshop happens on 3 evenings, and covers the following topics: (This is subject to change)

Day 1: Introduction and Generative Models

  • Introduction to supervised, unsupervised, weakly supervised learning
  • (Deep) Autoencoders: Pros/Cons
  • Hands-on session: Implementation of an autoencoder in PyTorch

Day 2: Embedding

  • Traditional methods for clustering data
  • Clustering techniques in deep learning
  • Hands-on session: Implementing a clustering loss with PyTorch and visualization of embedded spaces

Day 3: Weakly Supervised Learning

  • Transfer Learning: One-shot, zero-shot learning
  • Advanced weak supervision techniques
  • Hands-on session: Implementation of a loss function modeled for weak supervision

█░ Kick-starter style

Please note that our technical workshops are run kick-starter style in the sense that they will only happen if a certain amount of funds are raised by the 2-week deadline prior to the event (November 14, 2019):

Funds raised so far to enable the workshop:

████████░░ 75%

Learning Packages

You can customize your learning package. Click on tickets, select your base package (in-person or online), and then tailor your experience the way you want using add-ons.

There is early bird public discount starting at 40% and decreasing by 10% every week until the end of early bird on November 14th, 2019

Discount Codes

There are discount codes available

  • AISC members receive discounts according to their participation credit; refer to the slack channel for more detail

Referral Program

AISC Members, have personalized referral discount codes up to 50% that they can share with their friends and coworkers. Each code can be used 3 times. In order to ensure that people are using the code with your permission they need to enter your email address when registering

Non-members, can enjoy a group discount (10% on top of the current early bird discount) by forming groups of 3+ and registering together. So, go talk to your friends or coworkers right now so that you all save 10% more together.

Share with friends

Date and Time

Refund Policy

Contact the organizer to request a refund.

Eventbrite's fee is nonrefundable.

Save This Event

Event Saved