Skip to main content

Introduction to Machine Learning

Neural Networks

Instead of writing program by hand, give computer a lot of examples / data with the output, so computer can learn to improve

Machine Learning​

Machine learning is concerned with computer programs that automatically improve their performance through experience

Study of computer algorithms that can improve automatically through experiences without being explicitly programmed

  • Top-down/Classic: model all different functions and wire all these 'agents' together
  • Bottom-up / Deep learning: give the system a lot of data

Artificial Neural Networks​

  • Simulate, in the computer, neural networks in the brain
  • McCulloh and Pits (1943): 1st neural networks
    • Combine simple inputs, idea of threshold
    • Weights design by humans
  • Hebb (1949): First learning rule
    • Would adjust weights as it was ran
  • 50s/60s
    • Perceptrpm: great excitement
    • Coverage to correct weights: learning →\to thinking
  • 1969: death/winter of ANN
    • Perceptron cant learn certain type of important functions (non- linear separable)
  • Mid 80s: discovered multi-layer networks to solve problem of non-linear separable

Neurons​

  • Signals "move" between neurons
  • Sum of inputs >= threshold, neuron fires (output of 1)
  • Long-term firing patters - basing learning
  • Theory behind first neural network
  • McCulloch and Pitts produced the 1st neural networks in 1943

First Neural Networks​

  • Fundamental processing unit of a brain: a neuron
  • A processing element - (neuron)
  • Inputs (dendrites)
  • Output (axon)
  • Weights (synapses)

Positive weight: excitatory, otherwise inhibitory

8f7d9d6504f97d72a58c200294c3014e.png As long as the sum is above the threshold its correct

2edf4d1b6db1e258e068d9f261d7b02a.png

Training a NN​

Weights are normally randomly assigned

  • Epoch: Entire training set feed into the neural network. The AND function: an epoch consists of four sets of inputs(patterns) feed into the network
  • Training Value, T: Value that we require the network to produce
  • Error, Err: The amount the output by the network O differs from the training value T
  • X_i: Inputs to Neuron
  • w_i: Weight from input X_i, to the output
  • LR (Learning Rate) How quickly the network converges. It is set by the experimentation, typically 0.1. This is the adjustment amount
While epoch produces an error
    Check next inputs(pattern) from epoch
    Err = T - O
    If Err <> 0 then
        w_i = w_i + LR * X_i * Err
    End If
End While

Performance Measure:

  • Mean squared error = [(T-O)^2]/n
    • (n) number of items

Linearly Separable: Functions which can be separated. Only linearly separable functions can be represented by a single layer NN, i,e, perception

If you can separate your inputs and outputs with a straight line then you have found the weights

15786801da80d2bbcf14b1b38be2beaf.png

Current AI Developments​

AlphaCode - Writes computer programs at a competitive level. Ranked top 54% in real-world programming competitions

Deep Learning - Fathers of Deep Learning, Handwriting image recognition, language processing, vision etc.

Learning From Experience - Deep neural networks learn by adjusting the strengths of their connections to better convey input signals through multiple layers to neurons associated with the right general concepts