AI basics
Opened this issue · 1 comments
- AI types
- Heuristics
- Search algorithm
Notes on
https://www.3blue1brown.com/lessons/neural-networks
https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
Following this page: https://gist.github.com/rain-1/eebd5e5eb2784feecf450324e3341c8d
Transformers
Neural Networks
Example: classifying an image as a number between 0-9
Neurons - Thing that holds a number
Hidden Layers
Designing middle layers - if we're thinking about how to determine if a picture is of a given number, we would want to have a way we could imagine doing it.
We'll want to have an input, which from a picture, it might be a brightness value for the pixel on the image, which it uses to determine if anything has been written in that pixel, representing part of the number.
Sigmoid curve - a curve that measures how likely something is to to be true under certain circumstances. inputs are weights with a set bias to help set desired range of optimal inputs
Weights - the total values of the pixel's brightness in the first layer, to
Biases
So if we look at most numbers, they have small visual patterns that make them up, like loops and lines. For it to figure out how a loop might look, it could count the number of edges, figure and based on the rule
- Activation - a threshold
Network
Each neuron corresponds to a component of the input, so each neuron in 3b1b's example could be a pixel with a value like how bright the pixel is (0-1). So the list of values for all of the pixels in an image could be the inputs to the first layer of a neural network. Your final layer could be 10 neurons, one for each digit.