This preview shows page 1. Sign up to view the full content.
Unformatted text preview: es and futures prices. The RBF network has two 3 hidden layers, with the first hidden layer being trained using the K-means clustering
algorithm which is thus, an unsupervised learning algorithm. Once an initial solution for
the means and standard deviations of each neuron is found using the clustering algorithm,
a supervised learning algorithm is applied to fine tune the parameters of both the hidden
layers. Gradient descent is used for the supervised learning. No details are given regarding
the inputs used or the feature extraction.
Multi-component trading system using S&P 500 prediction
Obradovic et al  use two neural networks, to predict the returns on S&P 500 stock
index. One network is trained using upward trending data while another is trained for
downward trending data. The test data is fed to both the neural networks after filtering. A
complex filtering scheme is used based on traditional direction indicators. The outputs of
the two networks are combined using a high level decision rule base to obta...
View Full Document
- Spring '12