Comparison between Oja's and BCM neural networks models in finding useful projections in high-dimensional spaces

This thesis presents the concept of a neural network starting from its corresponding biological model, paying particular attention to the learning algorithms proposed by Oja and Bienenstock Cooper & Munro. A brief introduction to Data Analysis is then performed, with particular reference to the Principal Components Analysis and Singular Value Decomposition. The two previously introduced algorithms are then dealt with more thoroughly, going to study in particular their connections with data analysis. Finally, it is proposed to use the Singular Value Decomposition as a method for obtaining stationary points in the BCM algorithm, in the case of linearly dependent inputs.