How do you get here? https://www.youtube.com/watch?v=dk3oc1Hr62g
Pattern Recognition & Machine Learning
Patterns Humans are excellent at recognizing patterns
Patterns Even if we can't explain how we do it…
Trick 1: Nearest Neighbor Task : predict what houses are most likely to donate to an election
Nearest Neighbor Task : predict what houses are most likely to donate to an election Know some voter registrations
Nearest Neighbor Task : predict what houses are most likely to donate to an election What should we predict for the ? marks
Nearest Neighbor Task : predict what houses are most likely to donate to an election Should we consider more than one neighbor?
Other Nearest Neighbor Nearness as pixel difference:
Trick 2: Decision Trees Sequnce of choices to make a decision Do I need an umbrella?
Learning a Decision Tree Is an email important?
Machine Learning Machine Learning : Build a general algorithm to LEARN specific patterns
Learning a Decision Tree http://aispace.org/dTree/
Human Involvement Still need to determine possible questions, things to look at
Human Involvement Still need to determine possible questions, things to look at What should we look at for these???
Trick 3: Neural Networks Biologically inspired computation
Neural Networks Biologically inspired computation
Neural Networks A simple "take umbrella" network:
Neural Networks
Sunglasses Network Image recognition network:
Sunglasses Network Image recognition network:
Enhanced Neurons Signals can be any value 0-1
Enhanced Neurons Signals can be any value 0-1 Inputs can be weighted
Enhanced Neurons Signals can be any value 0-1 Inputs can be weighted Threshold function is not all or nothing Produces values 0-1
Learning Neural network learns via training Guess for lots of known examples Update weights based on success/failure No Yes Yes No No Yes
Samples https://playground.tensorflow.org
Result One neuron's weights:
Other samples https://cs.stanford.edu/people/karpathy/convnetjs/