Neural Networks

86.54 - Basic concepts of neural networks. Hopfield Networks, Ising Model, Simple-Layer Perceptron, Multi-Layer Perceptron, Genetic Algorithms, Kohonen Networks, Simulated Annealing.

Using Hopfield Network to clean image noise

Hopfield Networks can be trained with a set of images/data and recognise noisy/modified versions of the same images/data. Nevertheless, they can get confused if too much noise is applied (see middle column pictures below).

Using multi-layer perceptron to estimate function output

A multi-layer perceptron can estimate a function's output based on an input and has the potential to make more efficient computations on some functions.

Using Kohonen Network to solve salesman problem

Solving the salesman problem where we need to find the optimal route through all points in a city is a complex task. Despite of that, Kohonen Networks can solve the problem pretty fast.