A fundamental optimization algorithm for minimizing a model’s loss function by iteratively moving weights in the direction of the negative gradient. In neural network training, gradient descent (and its variants) adjusts the network’s weights to reduce prediction error. Forex AI systems using neural nets rely on gradient descent (often via backpropagation) to learn from data. Proper tuning of learning rate and algorithm variant (SGD, Adam, etc.) is crucial to converge efficiently.

Back to Robots Glossary Index