Perceptron Structure, Backpropagation, and Trading Signals in MQL5

Summary

The article introduces feedforward neural networks through the perceptron, explaining how input, hidden, and output layers pass information forward. Each neuron combines weighted inputs and applies an activation function; the author describes sigmoid and hyperbolic tangent outputs, with the latter mapped conceptually to sell, buy, and uncertain signals. Layer sizes are linked to the number of inputs and desired outputs, while additional hidden layers increase complexity and computational needs.

It outlines supervised, unsupervised, and reinforcement learning, then focuses on supervised backpropagation. The network compares its output with a reference answer, propagates error backward, and adjusts connection weights using stochastic gradient descent; the article also discusses continued online training. MQL5 classes are used to build and store a perceptron without external libraries. This is an introductory implementation example, not a rigorous trading evaluation: no out-of-sample results, benchmark comparisons, or risk analysis are provided, and the author presents the perceptron as one limited model among many.

Key ideas

• A feedforward network passes inputs through layers of weighted neurons to produce outputs.
• Activation functions transform weighted sums, and the selected function should fit the output task.
• Supervised backpropagation adjusts weights using the difference between predictions and reference answers.
• The MQL5 example implements a perceptron and supports saving its connection data.
• The article offers no empirical evidence that the example produces profitable trading signals.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.