Training Kohonen Maps for Financial Market Data
Summary
The article introduces Kohonen self-organizing layers as a way to group input vectors into regions of similar patterns. It explains winner-take-all selection, in which the neuron with the closest weight vector responds, and adjusts that winner toward the input using a learning rate that decreases over training. It also discusses how neuron count affects classification and why input scaling and initial weights matter.
For financial examples, the article proposes using lagged price or indicator observations as input data. It describes normalizing vectors, optionally centering positive-valued samples, and several weight initialization approaches, including random and example-based initialization. The discussion is introductory and grounded in conceptual illustrations and established references; it does not provide empirical evidence that the technique produces profitable forecasts. It explicitly cautions that neural networks are not a universal solution for trading.
Key ideas
- A Kohonen layer assigns each input vector to the neuron with the closest weight vector.
- Training moves the winning neuron's weights toward the input, with the learning rate declining over time.
- Too few neurons can increase classification error, while random initialization may leave some neurons poorly placed to learn.
- Normalizing or centering inputs can change how the network organizes financial observations.
- The article explains network mechanics but does not demonstrate trading profitability.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.