MONTRÉAL.AI | Montréal Artificial Intelligence

Montréal.AI Academy


The First World-Class Overview of AI for All

For the newcomers to artificial intelligence, the General Secretariat of MONTREAL.AI introduces, with authority and insider knowledge: “Artificial Intelligence 101: The First World-Class Overview of AI for the General Public“.


This AI 101 tutorial (webinar) harnesses the fundamentals of artificial intelligence for the purpose of providing participants with powerful AI tools to learn, deploy and scale AI. AI opens up a world of new possibilities.

The best way to predict the future is to invent it.“ — Alan Kay

Language: Tutorial (webinar) given in English.

A Well-Crafted Actionable 75 Minutes Tutorial

POWERFUL & USEFUL. This actionable tutorial is designed to entrust participants with the mindset, the skills and the tools to see AI from an empowering new vantage point by : exalting state of the art discoveries and science, curating the best open-source implementations and embodying the impetus that drives today’s AI.

Program overview: “Pioneering an Impactful Understanding of AI”

Opening Address

Session 0 · Getting Started

  • In the Cloud
  • On a Local Machine

Session 1 · Deep Learning

  • Neural Networks
  • Convolution Neural Networks (CNNs)
  • Recurrent Neural Networks (RNNs)
  • Unsupervised Deep Learning (Self-Supervised Learning, Generative Adversarial Nets and Variational Autoencoders)

Session 2 · Autonomous Agents

  • Evolution Strategies
  • Deep Reinforcement Learning
  • Self Play: A Quantum Leap
  • Deep Meta-Learning

Session 3 · Environments

  • OpenAI Gym
  • DeepMind Lab
  • Unity ML-Agents

Closing Remarks

Keynote Speaker

Keynote: Vincent Boucher, Founding Chairman at MONTREAL.AI.

In 1996, Vincent Boucher completed a B. Sc. Theoretical Physics in 1 (one) year, followed by a Master’s degree in Government Policy Analysis (1998) and a Master’s degree in Aerospace Engineering (Space Technology) (2000). From 2000 to 2002, he provided management consulting services for the Canadian Space Agency.

In 2003, Vincent Boucher founded MONTREAL.AI | Montréal Artificial Intelligence.

In 2016+, Vincent Boucher (username: ceobillionaire) ranked #1 worldwide on the OpenAI Gym :

A Legendary History | How It All Began — Learn the Source of an Exceptional Legacy :

A Legendary History | How It All Began

(AI) will rank among our greatest technological achievements, and everyone deserves to play a role in shaping it.“ — Fei-Fei Li

Group Reservation

Group Reservation: :

VIP AI 101 CheatSheet for All

For the purpose of entrusting all sentient beings with powerful AI tools to learn, deploy and scale AI in order to enhance their prosperity, to settle planetary-scale problems and to inspire those who, with AI, will shape the 21st Century, Montréal.AI introduces the “VIP AI 101 CheatSheet for All”.

Curated Open-Source Codes, Implementations and Science

MONTREAL.AI is Preparing a Global Network of Education Centers to Pioneer an Impactful Understanding of AI and to Foster a Vector for Safe Humanitarian Artificial General Intelligence (AGI).

Montréal.AI created the largest artificial intelligence community in Canada. Join us, learn and discuss at !

Curated Open-Source Codes, Implementations and Science

You are qualified for a career in machine learning!

0. Getting Started

AI opens up a world of new possibilities. Today’s artificial intelligence is powerful, useful and accessible to all.

Tinker with Neural Networks : Neural Network Playground — TensorFlow

In the Cloud

Free GPU compute via Colab.

Colaboratory is a hosted Jupyter notebook environment that is free to use and requires no setup.

On a Local Machine

Install Anaconda and Launch ‘Anaconda Navigator’.

JupyterLab is Ready for Users.

Update Jupyterlab and Launch the Application. Under Notebook, Click on ‘Python 3’.

In the Browser

TensorFlow.js: a library for developing and training ML models in JavaScript.


Making it easier to discover datasets.

Google Dataset Search

Preliminary Readings

When you first study a field, it seems like you have to memorize a zillion things. You don’t. What you need is to identify the 3-5 core principles that govern the field. The million things you thought you had to memorize are various combinations of the core principles.“ — J. Reed

The Loss Landscape. Source: Visuals by Javier Ideami.

1. Multiply things together 2. Add them up 3. Replaces negatives with zeros 4. Return to step 1, a hundred times.“ — Jeremy Howard

Types of Learning, by Alex Graves at NeurIPS 2018

An embedding is a mapping from discrete objects, such as words, to vectors of real numbers.

1. Deep Learning

Deep learning allows computational models that are composed of multiple processing layers to learn REPRESENTATIONS of (raw) data with multiple levels of abstraction. At a high-level, neural networks are either encoders, decoders, or a combination of both.

DL is essentially a new style of programming–”differentiable programming”–and the field is trying to work out the reusable constructs in this style. We have some: convolution, pooling, LSTM, GAN, VAE, memory units, routing units, etc.“ — Thomas G. Dietterich

1.1 Neural Networks

Neural networks” are a sad misnomer. They’re neither neural nor even networks. They’re chains of differentiable, parameterized geometric functions, trained with gradient descent (with gradients obtained via the chain rule). A small set of highschool-level ideas put together.“ — François Chollet

Forward pass in a neural networks with two hidden layers

Backward pass in a neural networks with two hidden layers

1.1.1 Universal Approximation Theorem

The universal approximation theorem states that a feed-forward network with a single hidden layer containing a finite number of neurons can solve any given problem to arbitrarily close accuracy as long as you add enough parameters.

The chain rule applied to one single unit. Image source: Markus Völk.

Neural Networks + Gradient Descent + GPU:

When a choice must be made, just feed the (raw) data to a deep neural network (universal function approximator).

1.2 Convolution Neural Network

In images, local combinations of edges form motifs, motifs assemble into parts, and parts form objects.

The deep convolutional network, inspired by Hubel and Wiesel’s seminal work on early visual cortex, uses hierarchical layers of tiled convolutional filters to mimic the effects of receptive fields, thereby exploiting the local spatial correlations present in images.

Architecture of LeNet-5, a Convolutional Neural Network. LeCun et al., 1998

A ConvNet is made up of Layers. Every Layer has a simple API: It transforms an input 3D volume to an output 3D volume with some differentiable function that may or may not have parameters.

I admire the elegance of your method of computation; it must be nice to ride through these fields upon the horse of true mathematics while the like of us have to make our way laboriously on foot.“ — A. Einstein

Interactive Inceptionv3 created by TensorSpace

1.3 Recurrent Neural Networks

For sequential inputs. Recurrent neural networks are networks with loops in them, allowing information to persist. RNNs process an input sequence one element at a time, maintaining in their hidden units a ‘state vector’ that implicitly contains information about the history of all the past elements of the sequence.

RNN Layers Reuse Weights for Multiple Timesteps

Google Smart Reply System is built on a pair of recurrent neural networks. Diagram by Chris Olah

  • The Unreasonable Effectiveness of Recurrent Neural Networks — Andrej Karpathy
  • Massive Exploration of Neural Machine Translation Architectures arXiv | Docs | Code — Denny Britz, Anna Goldie, Minh-Thang Luong, Quoc Le
  • A TensorFlow implementation of : “Hybrid computing using a neural network with dynamic external memory” GitHub — Alex Graves, Greg Wayne, Malcolm Reynolds, Tim Harley, Ivo Danihelka, Agnieszka Grabska-Barwińska, Sergio Gómez Colmenarejo, Edward Grefenstette, Tiago Ramalho, John Agapiou, Adrià Puigdomènech Badia, Karl Moritz Hermann, Yori Zwols, Georg Ostrovski, Adam Cain, Helen King, Christopher Summerfield, Phil Blunsom, Koray Kavukcuoglu & Demis Hassabis

I feel like a significant percentage of Deep Learning breakthroughs ask the question “how can I reuse weights in multiple places?”
– Recurrent (LSTM) layers reuse for multiple timesteps
– Convolutional layers reuse in multiple locations.
– Capsules reuse across orientation.
“ — Trask

1.4 Capsules

1.5 Unsupervised Learning

True intelligence will require independent learning strategies. Unsupervised learning is a paradigm for creating AI that learns without a particular task in mind: learning for the sake of learning. It captures some characteristics of the joint distribution of the observed random variables (learn the underlying structure).

Self-supervised learning is derived form unsupervised learning where the data provides the supervision.

1.5.1 Generative Adversarial Network

Simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake.

$$\min_{\theta_g} \max_{\theta_d} [{\rm IE_{x\sim p_{data}(x)}} [log D_{\theta_d}(x)] + {\rm IE_{z\sim p_z(z)}} [log(1 - D_{\theta_d}(G_{\theta_g}(z)))]]$$

What I cannot create, I do not understand.“ — Richard Feynman

This framework corresponds to a minimax two-player game. Generative Adversarial Nets — Goodfellow et al.

GAN: Neural Networks Architecture Pioneered by Ian Goodfellow at University of Montreal (2014)

StyleGAN: A Style-Based Generator Architecture for Generative Adversarial Networks Paper | Code

Demo of BigGAN in an official Colaboratory notebook (backed by a GPU)

1.5.2 Variational Auto-Encoders (VAEs)

Variational Auto-Encoders (VAEs) are powerful models for learning low-dimensional representations.

Variational Autoencoders (VAEs): Powerful Generative Models.

I think transfer learning is the key to general intelligence. And I think the key to doing transfer learning will be the acquisition of conceptual knowledge that is abstracted away from perceptual details of where you learned it from.“ — Demis Hassabis

2. Autonomous Agents

Reinforcement learning (RL) studies how an agent can learn how to achieve goals in a complex, uncertain environment.

Artificial Intelligence 101: The First World-Class Overview of AI for the General Public | Autonomous Agents

An autonomous agent is any device that perceives its environment and takes actions that maximize its chance of success at some goal. At the bleeding edge of AI, autonomous agents can learn from experience, simulate worlds and orchestrate meta-solutions. Here’s an informal definition of the universal intelligence of agent `\pi`

$$\Upsilon(\pi) := \sum\limits_{\mu \in E} 2^{-K(\mu)} V^{\pi}_{\mu}$$

Intelligence measures an agent’s ability to achieve goals in a wide range of environments.“ — Shane Legg

2.1 Evolution Strategies

Evolution is a slow learning algorithm that with the sufficient amount of compute produces a human brain.“ — Wojciech Zaremba

Evolution and neural networks proved a potent combination in nature. Natural evolutionary strategy directly evolves the weights of a DNN and performs competitively with the best deep reinforcement learning algorithms. Neuroevolution enables capabilities that are typically unavailable to gradient-based approaches.

Exploiting Potential Energy to Locomote. Source: The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities.

Flexible Muscle-Based Locomotion for Bipedal Creatures. Source: ACM Transactions on Graphics, Vol. 32, Nr. 6 (Proc. of SIGGRAPH Asia 2013).

Neural architecture search has advanced to the point where it can outperform human-designed models.

… evolution — whether biological or computational — is inherently creative, and should routinely be expected to surprise, delight, and even outwit us.“ — The Surprising Creativity of Digital Evolution, Lehman et al.

2.2 Deep Reinforcement Learning

In reinforcement learning, an agent interacts with an environment through a Markov decision process.

An Agent Interacts with an Environment

The goal in reinforcement learning is to train the agent to maximize the sum of future rewards, called the return.

Asynchronous Advantage Actor-Critic (A3C). Source: Petar Velickovic

Open-source RL

2.2.1 Model-Based RL

In Model-Based RL, the agent generates predictions about the next state and reward before choosing each action.

No superintelligent AI is going to bother with a task that is harder than hacking its reward function.“ — The Lebowski theorem

2.3 Self Play

Self-play mirrors similar insights from coevolution.

AlphaGo Zero showed that algorithms matter much more than big data and massive amounts of computation:

AlphaGo Zero : Algorithms matter much more than big data and massive amounts of computation

Self-Play is Automated Knowledge Creation.“ — Carlos E. Perez

Transfer learning is the key to go from self-play to the real world.

2.4 Multi-Agent Populations

2.5 Deep Meta-Learning

Learning to Learn. The goal of meta-learning is to train a model on a variety of learning tasks, such that it can solve new learning tasks using only a small number of training samples. A meta-learning algorithm takes in a distribution of tasks, where each task is a learning problem, and it produces a quick learner — a learner that can generalize from a small number of examples.

The notion of a neural “architecture” is going to disappear thanks to meta learning.“ — Andrew Trask

3. Environments

3.1 OpenAI Gym

3.2 Unity ML-Agents

3.3 DeepMind Control Suite

3.4 Brigham Young University | Holodeck

  • BYU Holodeck: A high-fidelity simulator for deep reinforcement learning Website | GitHub | Documentation — Brigham Young University

3.5 Facebook’s Horizon

  • Horizon: Facebook’s Open Source Applied Reinforcement Learning Platform Paper | GitHub | Blog — Jason Gauci, Edoardo Conti, Yitao Liang, Kittipat Virochsiri, Yuchen He, Zachary Kaden, Vivek Narayanan, Xiaohui Ye

3.6 PhysX

  • PhysX SDK, an Open-Source Physics Engine GitHub | Blog — NVIDIA

4. General Readings, Ressources and Tools

A Map of Ethical and Right-Based Approaches. Source:

I believe that the answer here is to figure out how to create superintelligent A.I. such that even if – when – it escapes, it is still safe because it is fundamentally on our side because it shares our values. I see no way around this difficult problem.“ — Nick Bostrom

ML paper writing pro-tip: you can download the raw source of any arxiv paper. Click on the “Other formats” link, then click “Download source”. This gets you a .tar.gz with all the .tex files, all the image files for the figures in their original resolution, etc.“ — Ian Goodfellow

Deep Reinforcement Learning with OpenAI Gym 101

Deep Reinforcement Learning with OpenAI Gym 101

Language: Tutorial (webinar) given in English.

Intelligence is the computational part of the ability to predict and control a stream of experience.“ — Rich Sutton

Customized Artificial Intelligence Training for Businesses

Organizations Adopting New AI Tools Need to Develop Internal Skills

Customized Artificial Intelligence Training for Businesses

MONTREAL.AI provides made-to-measure training programs in digital intelligence for select organizations.

For any questions about our made-to-measure training programs in digital intelligence :

✉️ Email Us :
🌐 Website :
🐦 Twitter :
📝 LinkedIn :
🏛 Headquarters : 350, PRINCE-ARTHUR STREET W., SUITE #2105, MONTREAL [QC], CANADA, H2X 3R4 *Administrative Head Office

#AIFirst #ArtificialIntelligence101 #MontrealAI #MontrealAIAcademy #MontrealArtificialIntelligence

Your browser is out-of-date!

Update your browser to view this website correctly. Update my browser now