Unsupervised, Self-supervised and Reinforcement Learning | SGInnovate
March052020
March062020

Location

BASH, Level 3,
79 Ayer Rajah Crescent
Singapore 139955

Price

Early Bird [Ends on 15 February 2020] (Ticket Inclusive of G.S.T) - $1,524.75
Module 4 (Ticket Inclusive of G.S.T) - $1605

Unsupervised, Self-supervised and Reinforcement Learning

Presented by SGInnovate and Red Dragon AI

Together with Red Dragon AI, SGInnovate is pleased to present the fourth module of the Deep Learning Developer Series. In this module we dive deeper into some of the latest techniques for using Deep Learning through unsupervised, self-supervised and reinforcement learning.

About the Deep Learning Developer Series:

The Deep Learning Developer Series is a hands-on and cutting-edge series targeted at developers and data scientists who are looking to build Artificial Intelligence applications for real-world usage. It is an expanded curriculum that breaks away from the regular eight-week long full-time course structure and allows for modular customisation according to your own pace and preference. In every module, you will have the opportunity to build your own Deep Learning models as part of your main project. You will also be challenged to use your new skills in an application that relates to your field of work or interest.

Start here

    • Have an interest in Deep Learning?
    • Join us if you are able to read and follow codes
    • This module is compulsory before you take the advanced modules
     
    • You will need to take module 1 before this module
     
    • You will need to take module 1 before this module
     
    • You will need to take module 1 AND module 2 OR 3 before this module
     
    • You will need to take module 1, 2 AND 3 before this module
    • Attain a “Deep Learning Specialist” certification when you complete all five modules
     
    • Have an interest in Deep Learning?
    • Join us if you are able to read and follow codes
    • This module is compulsory before you take the advanced modules
     
    • You will need to take module 1 before this module
     
    • You will need to take module 1 before this module
     
    • You will need to take module 1 AND module 2 OR 3 before this module
     
    • You will need to take module 1, 2 AND 3 before this module
    • Attain a “Deep Learning Specialist” certification when you complete all five modules
     

About this module:

This module looks at some of the latest developments in Deep Learning research; it will cover the latest and most promising developments in Machine Learning and AI.

One issue with many of the current techniques used in Machine Learning is the requirement for lots of labelled data, which is both costly and time-consuming to create.

There are several techniques to extract information from raw unlabelled data such as Autoregressive models, Representational Learning and Cycle Consistency. We will examine each of these in-depth to give the student an understanding of how they work and how they can be applied.

We will also look at a variety of techniques such as Generative Adversarial Networks and how they are being used to produce realistic and lifelike examples, such as generated pictures of faces. We will explore how Autoregressive models such as Wavenet, PixelRNN, GPT2 can help create self-labelling systems. We will also touch on Variational Auto Encoders, the concepts of latent representations, and how to extract representational learnings from data.

We will dive into the field of Reinforcement Learning, which has led to breakthroughs such as DeepMind’s AlphaGo, Alpha Star and OpenAI’s DOTA models. We will look at how you can examine problems within a game and develop algorithms to tackle them.

As with all the other Deep Learning Developer modules, you will have the opportunity to build multiple models yourself. These include your main project, which gives you the ability to take these new skills and apply them to your field of work or interest.

This workshop is in the process of funding support application.

In this course, participants will learn key techniques such as:

  • Unsupervised Learning
  • Representational Learning (RL)
  • Auto Regressive Self-supervised Learning
  • Reinforcement Learning
  • Generative Adversarial techniques
  • Cycle Consistency

Some of the models we will look at include:

  • Generative Adversarial Networks (GAN)
  • StyleGAN
  • BigGAN
  • CycleGAN
  • InfoGAN
  • Auto encoders, VAEs (CVAE, BetaVAE)
  • RL Q Learning
  • Actor-critic Models, PPO

Recommended Prerequisites:

 


 

Day 1 Morning (5 March 2020)

  • Presenting and understanding information — traditional Unsupervised Learning
    • PCA / Clustering
    • TSNE / UMAP
  • AutoEncoders
    • Code (denoising)
  • Progress in Unsupervised and Semi-supervised Learning
    • An overview
  • Autoregressive models
    • CharRNN
    • PixelRNN
    • Conditional PixelCNN
    • Language Models, GPT2
    • WaveNet

Day 1 Afternoon (5 March 2020)

  • VAEs
    • Code
    • BetaVAE
    • Disentangling representations
  • Latent Spaces
    • Embeddings for words
    • Embed all the Things
    • StyleGAN Notebook
    • Language Translation matching
  • WaveNet (audio)

Day 2 Morning (6 March 2020)

  • Introduction to Generative Adversarial Networks (GAN)
    • Basic GAN concepts
    • Coding your first GAN
  • ConditionalGAN
    • Code walkthrough
    • Examples of bigger conditional GANs
    • BigGAN
  • CycleGAN
    • Code walkthrough
    • Uses of CycleGAN eg. Unsupervised Image Segmentation
  • SGAN: Self-supervised GAN for Data Augmentation
    • Code walkthrough
  • StyleGAN
    • Finding a latent space in a GAN
    • Uses of StyleGAN
    • StyleGAN2

Day 2 Afternoon (6 March 2020)

  • Introduction to Reinforcement Learning (RL)
    • AlphaGo / AlphaZero
    • Exploitation vs Exploration
    • Bubble Breaker
    • Q-Learning
    • Proximal Policy Optimisation
    • Actor-critic Models

Early Bird (Ends on 15 February 2020) - $1,524.75 (inclusive of GST)
Normal Ticket Module 4 - $1,605 (inclusive of GST)

Dr Martin Andrews

Martin has over 20 years’ experience in Machine Learning and has used it to solve problems in financial modelling and has created AI automation for companies. His current area of focus and speciality is in natural language processing and understanding. In 2017, Google appointed Martin as one of the first 12 Google Developer Experts for Machine Learning. Martin is also one of the co-founders of Red Dragon AI. 

Sam Witteveen

Sam has used Machine Learning and Deep Learning in building multiple tech start-ups, including a children’s educational app provider which has over 4 million users worldwide. His current focus is AI for conversational agents to allow humans to interact easier and faster with computers. In 2017, Google appointed Sam as one of the first 12 Google Developer Experts for Machine Learning in the world. Sam is also one of the co-founders of Red Dragon AI. 

Topics: Artificial Intelligence / Deep Learning / Machine Learning / Robotics