Neural network – A flexible function approximator composed of layers of interconnected “neurons” or nodes. Each neuron applies a weight and bias to inputs and passes the result through an activation. Neural networks (especially deep ones) automatically learn complex patterns in data. In forex, deep networks (including LSTM networks) can model sequential price data. As noted, neural networks “process training data with layers of nodes that mimic the human brain”. They are powerful for both regression and classification but require careful tuning (e.g. learning rate, model.fit(), many epochs).

Back to Robots Glossary Index