Deep Learning

A summary of the content from taking this course to prepare for Grad School: https://www.udacity.com/course/intro-to-tensorflow-for-deep-learning--ud187

Basic Terms

Artificial Intelligence - Computer science trying to make computers perform intelligence tasks like humans.

• Machine Learning - Computation techniques where computers are trained

  • Neural Network - Inspired literally by networks of neurons found in biological brains. Fundamental component of Deep Learning

    • Deep Learning - A subset of ML using Neural Networks

Supervised Learning - You know what you want to teach the computer Unsupervised Learning - You let the computer figure out what can be learned

Feature - Input(s) to an ML model Labels - The output a model predicts

Layer - A collection of nodes together within a network Model - The representation of your neural network (or whatever Machine Learning paradigm you're going for) Learning Rate - The "step size" for loss improvement during gradient descent Batch - Set of examples used during training of the neural network Epoch - A full pass over the entire training dataset Forward pass - Computation of output values from input Backward pass (backpropagation) - The calculation of internal variable adjustments according to the optimizer algorithm, starting from the output layer and working back through each layer to the input.

Current Applications

• Deep Learning Model detects cancer from growths better than dermatologists: https://www.nature.com/articles/nature21056

• Self-driving cars trained on data from a few sensors: http://www.princeton.edu/~alaink/Orf467F14/Deep%20Driving.pdf

• Training software on Starcraft games that can beat world-class players: https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/

What is Machine Learning?

Traditional Software Development

The input and the algorithm is known, and you write a function to produce an output:

• Input data

• Apply logic to it

• Produce a result

# Convert Celsius to Fahrenheit
def function(C):     # Input
    F = C * 1.8 + 32 # Algorithm
    return F         # Output

Machine Learning

You know the input and output, but you don't know the algorithm that creates the output given the input.

• Take pairs of input and output data

• Figure out the algorithm

Neural networks are used to learn a relation between inputs and outputs. Each layer is a self-defining set of algorithms and output networking. We train these layers by making the model attempt to solve equations.

Machine Learning tunes variables to map inputs to outputs