Practical Deep Learning for coders (

This is in my opinion the best free course on getting into the state of the art in deep learning. It is a site that offers a free 7 week learning experience for deep learning. taught by 2 year in a row Kaggle winner, entrepreneur and generally nice guy Jeremy Howard and Math PhD/Data scientist/Full stack developer/Forbes Featured Rachel Thomas, two amazing people in AI. Their approach to teaching Deep Learning for Coders is that it shall be accessible to as many people as possible and not to a selected few. So instead of abstract mathy lectures, they allow you to get your hands dirty from the first lecture and improve your intuition of the field, thus enabling you to create state of the art deep learning solutions from day one.

After starting the course, i immediately realized that these are very talented educators that are sincere about their goal to make AI accessible to everyone, and to make it benefit others. What i especially like about the course is the way they approach the topic pedagogically. Their method is inspired by the book “Making Learning Whole: How Seven Principles of Teaching can Transform Education” by author David Perkins. Perkins compares todays education with learning baseball:

If you would learn baseball the way that math is taught, you would first learn about the shape of a parabola, and then you would learn the material science behind the stitchings in baseballs and so forth. And twenty years later after you have completed your PhD and post-doc, you would be taken to your first baseball game and you would be introduced to the rules of baseball. And then 10 years later you might get to hit. The way that in practice baseball is taught is we take the kid down to the baseball diamond and we say “These people are playing baseball, would you like to play?” And they say, “Yeah! Sure I would”. “Perfect, stand here, i’m gonna throw this. Hit it. Ok, great, now run. Good you’re playing baseball”

That is why the first class of the course they demonstrate that here are 7 lines of code that you can use to perform state of the art image classification using deep learning. And to do any image classification you want as long as you structure it the right way. You may not understand most of it, but as you need to adapt the tasks to your needs, you will need to learn more details, and thus you learn.

The course consists of a 2 hour lecture each week, detailed lecture notes, a community contributed wiki and jupyter notebooks which you also will do your assignments in. (There is also setup instructions for getting a GPU equipped machine up and running on AWS) In The first weeks assigment you will submit an entry into the Kaggle competition for classifying cat and dog images. By taking advantage of what you learn you will outperform what was the state of the art when the competition was launched 2013.

Hybrid Reward Architecture breaks world record AI and Human for Pac Man

Maluuba, a microsoft bought up AI startup achieved the highest possible score (999 990) for the very difficult to beat Ms Pac-Man. It used a divide and conquer like reinforcement learning where responsibilities for each positive and negative reward giving elements in the game are assigned an individual agent that seeks to suggest to the player a move that is best for reaching that particular local goal. A “manager” agent recieves these suggestinos and decides what move the user shall perform in order to achieve the maximum reward.


Read the paper here

Paper: Introduction to Convolutional Nerual Networks by Jianxin Wu

In this recently published paper, Jianxin Wu helps the reader understand
how a CNN runs at the mathematical level. It is self contained and you should not need any further material to understand it from a mathematical viewpoint.

With CNN, the important part is understanding what happens when you adjust the different parameters. Bu in order to make sense of those it is much easier when you know the underlying principles behind it.

Here you go:

Another Deep Learning School

I recently posted about Deep Learning Summer School 2016

Here is a link to another Bay Area Deep Learning School one that was the same year in September at Stanford CA.


Introduction to Feedforward Neural Networks​

Hugo Larochelle​

I will cover some of the fundamental concepts behind feedforward neural networks. I’ll start by briefly reviewing the basic multi-layer architecture of feedforward networks, as well as backpropagation from automatic differentiation and stochastic gradient descent (SGD). Then, I’ll discuss the most recent ideas that are now commonly used for training deep neural networks, such as variants of SGD, dropout, batch normalization and unsupervised pretraining.

Video, Slides

Deep Learning for Computer Vision

Andrej Karpathy

I will cover the design of convolutional neural network (ConvNet) architectures for image understanding, the history of state of the art models on the ImageNet Large Scale Visual Recognition Challenge, and some of the most recent patterns of developments in this area. I will also talk about ConvNet architectures in the context of related visual recognition tasks such as object detection, segmentation, and video processing.

Video, Slides

Deep Learning for NLP

Richard Socher

I will describe the foundations of deep learning for natural language processing: word vectors, recurrent neural networks, tasks and models influenced by linguistics. I will end with some recent models that put together all these basic lego blocks into a very powerful deep architecture called dynamic memory network.

Video, Slides

Tensorflow Tutorial

Sherry Moore

Video, Slides

Foundations of Deep Unsupervised Learning

Ruslan Salakhutdinov

Building intelligent systems that are capable of extracting meaningful
representations from high-dimensional data lies at the core of solving many Artificial Intelligence tasks, including visual object recognition, information retrieval, speech perception, and language understanding. In this tutorial I will discuss mathematical basics of many popular unsupervised models, including Sparse Coding, Autoencoders, Restricted Boltzmann Machines (RBMs), Deep Boltzmann Machines (DBMs), and Variational Autoencoders (VAE). I will furtherdemonstrate that these models are capable of extracting useful hierarchical representations from high dimensional data with applications in visual object recognition, information retrieval, and natural language processing. Finally, time permitting, I will briefly discuss models that can generate natural language descriptions (captions) of images, as well as generate images from captions using attention mechanism.


Nuts and bolts of applying deep learning

Andrew Ng


Deep Reinforement Learning

John Schulman, OpenAI

I’ll start by providing an overview of the state of the art in deep reinforcement learning, including recent applications to video games (e.g., Atari), board games (AlphaGo) and simulated robotics. Then I’ll give a tutorial introduction to the two methods that lie at the core of these results: policy gradients and Q-learning. Finally, I’ll present a new analysis that shows the close similarity between these two methods. A theme of the talk will be to not only ask “what works?”, but also “when does it work?” and “why does it work?”; and to find the kind of answers that are actionable for tuning one’s implementation and designing better algorithms.

Video, Slides

Theano Tutorial

Pascal Lamblin

Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently, on CPU or GPU. Since its introduction, Theano has been one of the most popular frameworks in the machine learning community, and multiple frameworks for deep learning have been built on top of it (Lasagne, Keras, Blocks, …). This tutorial will focus first on the concepts behind Theano and how to build and evaluate simple expressions, and then we will see how more complex models can be defined and trained.

Video, Slides

Deep Learning for Speech

Adam Coates

Traditional speech recognition systems are built from numerous modules, each requiring its own challenging engineering. With deep learning it is now possible to create neural networks that perform most of the tasks of a traditional engine “end to end”, dramatically simplifying the development of new speech systems and opening a path to human-level performance. In this tutorial, we will walk through the steps for constructing one type of end-to-end system similar to Baidu’s “Deep Speech” model. We will put all of the pieces together to form a “scale model” of a state of the art speech system; small-scale versions of the neural networks now powering production speech engines.

Video, Slides

Torch Tutorial

Alex Wiltschko

Torch is an open platform for scientific computing in the Lua language, with a focus on machine learning, in particular deep learning. Torch is distinguished from other array libraries by having first-class support for GPU computation, and a clear, interactive and imperative style. Further, through the “NN” library, Torch has broad support for building and training neural networks by composing primitive blocks or layers together in compute graphs. Torch, although benefitting from
extensive industry support, is a community owned and community developed ecosystem. All neural net libraries, including Torch NN, TensorFlow and Theano, rely on automatic differentiation (AD) to manage the computation of gradients of complex compositions of functions. I will present some general background on automatic differentiation (AD), which is the fundamental abstraction of gradient based optimization, and demonstrate
Twitter’s flexible implementation of AD in the library torch-autograd

Video, Slides

Sequence to Sequence Learning for NLP and Speech

Quoc Le

I will first present the foundations of sequence to sequence (seq2seq) learning and attention models, and their applications in machine translation and speech recognition. Then I will discuss attention with pointers and functions. Finally I will describe how reinforcement learning can play a role in seq2seq and attention models.

Video, Slides

Foundations and Challenges of Deep Learning

Yoshua Bengio

Why is deep learning working as well as it does? What are some big challenges that remain ahead? This talk will first survey some key factors in the success of deep learning. First, from the context of the no-free lunch theorem, we will discuss the expressive power of deep netwroks to capture abstract distributed representations. Second, we will discuss our surprising ability to actually optimize the parameters of neural networks in spite of their non-convexity. We will then consider a few challenges ahead, including the core representation question of disentangling the underlying explanatory factors of variation, especially with unsupervised learning, why this is important for bringing reinforcement learning to the next level, and optimization questions that remain challenging, such as learning of long-term dependencies, understanding the optimization landscape of deep networks, and how learning in brains remain a mystery worth attacking from the deep learning perspective.

Video, Slides