Topic
Neural networks
A neural network is a parameterized model composed of connected computational layers.
At a glance
- Adjustable values
- Model parameters
- Training signal
- Loss function
Overview
A neural network is a parameterized model composed of connected computational layers. Training adjusts parameters to reduce a defined loss on examples. The architecture, data, and training process together influence what relationships the model can learn and how it behaves afterward.
From input to prediction
An input passes through model operations to produce an output. Layers transform representations, and the final output is interpreted according to the task, such as a category score or predicted value.
Learning from error
Training compares predictions with the target through a loss. Automatic differentiation computes gradients used by an optimizer to update parameters; separate evaluation checks whether the resulting model generalizes.
Sources and review
MOOR's explanatory text is supported by the following source links.
- Build the neural network — PyTorch
- Optimize model parameters — PyTorch