with Rob McDiarmid
Note: speakers are paraphrased
#What is an artificial neural network? A computational model that you can train to recognize patterns in data.
Used in image identification, text analysis, recommendation engines, spam/fraud detection, voice analysis, natural language and even using a neural network to train another network (AutoML by Google).
#Why Python?
- python is flexible and concise in it's syntax
- fast, so long as you aren't actually running things in python: most uses for neural networks are actually just using python as a wrapper for C/C++ libraries and offloading work to gpu's
Example: Predict House Price
- house is 10 years old and has 5 rooms
- feed in a list of houses with rooms and age values in the input layer
- use hidden layers to store additional information, nodes to represent a bonus value based on the newness of the house, and another node to store the base price of the house
- this is a basic model that is manually created
- with a larger more complex network you need a way for the network to adjust itself
#Gradient Decent
- using different algorithms that can work their way down a slope of possible values to find the optimal value
Using Keras to build the neural network in Python.
Math for neural networks often works best when using values between 0 and 1.
Note: had to leave early, notes are incomplete :(