Machine Learning

An Introduction

Machine Learning

Definitions?

The science of getting computers to act without being explicitly programmed

The construction and study of algorithms that can learn from and make predictions on data

The creation of a model from example inputs in order to make data-driven predictions or decisions

Learning from experience either with or without supervision from humans

Uses

Text classification: spam detection
Ranking items: priority inbox
Regression models: predicting page views
Regularization: text regression
Optimization: code breaking
Unsupervised learned: building a stock market index
Spatial similarity: clustering US Senators by the voting records
Recommendation system: suggesting R packages to users
Social network analysis: who to follow on Twitter
Underwriting: approve loans

Main Source: Machine Learning for Hackers

Flavours

Reinforcement Learning + Interact with an Environment
Data + Supervised Learning
Data + Unsupervised Learning

Algorithms

K-Nearest Neighbours
Decision Trees
Support Vector Machines
Neural Networks
Genetic Algorithms
Evolutionary Programs
Distributed ...
...

Games

War Games

Checkers

By Arthur Samuels, IBM, 1959

Tree Search

TD-Gammon

http://www.bkgm.com/articles/tesauro/tdl.html

Neural Networks

Artificial Neuron

Many Layers

## Creating and Training a Network ```js var network = sn.network([6, 10, 3]); var outputs = network.outputs([0, 0, 1, 1, 0, 0]); console.dir(outputs); for (var k = 0; k < 20; k++) { network.train([0, 0, 1, 1, 0, 0], [0, 1, 0], 0.1); // .... } ```

Digits

https://github.com/ajlopez/SimpleNeuron/tree/master/samples/digits

ConvNetJS

Deep Learning in your browser

http://cs.stanford.edu/people/karpathy/convnetjs/

https://github.com/ajlopez/JavaScriptAI/tree/master/static

ConvNetJS Cifar10

https://github.com/ajlopez/JavaScriptAI/tree/master/static

Genetic Algorithms

Search for a Solution to a Problem

Stochastic Search Algorithms

Essentially, Reinformence Learning Algorithms: Fitness

Genotype

Represents a solution/individual
Composed by "genes"
Evaluation of Performance
Mutation
Crossover

Generic Structure

## Creating and Running Populations ```js for (var k = 0; k < 1000; k++) genotypes.push(Path.createGenotype(/* ... */)); var mutators = []; for (var k = 0; k < 4; k++) mutators.push(Path.createMutator()); engine = new simplega.Engine({ minimize: true }); engine.setPopulation(genotypes); engine.setMutators(mutators); // var newgenotypes = engine.nextPopulation(); best = simplega.getBestGenotype(newgenotypes, { minimize: true }); ```

Path Sample

https://github.com/ajlopez/SimpleGA/blob/master/samples/path/path.html

Evolutionary Programs

Evolving Animals

https://github.com/ajlopez/SimpleGA/tree/master/samples/evol

Machine Learning

An Introduction