Aim
- binary classification for any input data, either discrete or continuous
How does the brain learn?
Components of human brain: ~10 000 000 000 of neurones connected through synapses
Each neuron has:
- a body (soma), an axon and more dendrites
and can be in a given state:
- active state - if the input information is over a given stimulation threshold
- passive state - otherwise
A synapse is a link between the axon of a neuron and the dendrites of other neurons and takes part in information exchange between neurons. On average, there are 5000 connections/ neuron.
What is an ANN?
-
a structure similar to a biological neural network
-
a set of nodes (units, neurons, processing elements) located in a graph with more layers
Nodes:
- have inputs and outputs
- perform a simple computing through an activation function
- connected by weighted links
Layers:
- input layer - contains m nodes (number of attributes of a data)
- output layer - contains r nodes (number of outputs)
- intermediate layers of different structures and sizes
Biological NN vs artificial NN
| BNN |
ANN |
| Soma |
Node |
| Dendrite |
Input |
| Axon |
Output |
| Activation |
Processing |
| Synapse |
Weighted connection |
How does a network learn?
A training set of n data
Form an ANN with m input nodes, r output nodes and an internal structure
- some hidden layers, each layer having a given number of nodes, with weighted connections between every 2 nodes of consecutive layers