• Current opened records

  • PCA and Competitive learning

Awards
Author(s):
Category:
Institution:
Region:
Winner Category:
Year:
Abstract:
  • This lab report describes the process of performing Principal Component Analysis (PCA) to reduce the dimensionality of datasets. Using the MNIST dataset as an example PCA is applied to the 784 dimensional data to reduce it down to 3 dimensions. The best features to represent the data are chosen by inspection using 3-Dimensional plots. Following this, the report describes how competitive learning can be used to cluster the numbers in the MNIST dataset and how the unsupervised training method can be optimised. The experiments found that using the first, third and fourth components separated the MNIST the best in the reduced feature space. Adding noise to the learning was shown to improve the performance of the competitive learning algorithm implemented.
Attached Documents: