Artificial Neural Networks course

Authors: Kenza Bouzid, Agnieszka Miszkurka, Tobias Höppe

Solutions for labs for ANN course at KTH. Each lab contains implementation of neural networks algorithms as well as notebooks with experiments.

Lab1 - From Single Layer to Multi Layer Networks

Part 1

• perceptron learning, delta rule
• Multi Layer Perceptron with generalised delta rule (backprop)

Part 2

• Multi-layer perceptron network for chaotic time-series prediction

Lab2 - Radial basis functions, competitive learning and self-organisation

• Function approximation with RBF Network
• Radial Basis Function trained with competitive learing
• Self-organising maps used for data cluserisation and visualisation

Lab3 - Hopfield networks

• Hopfield network with Hebbian learnign
• Experiments concerning various properties of hopfield networks such as:
  • capacity
  • convergence
  • distortion resistance
  • energy
  • sequential vs batch update
  • sparsity of patterns

Lab4 - Restricted Boltzmann Machines and Deep Belief Nets

• RBM trained with contrastive divergence
• DBN trained using greed layer-wise pretraining of RBMs
• Using RBM and DBN for classification and generating new samples with MNIST data set