Starting with Machine Learning on devices like MCU - Part 8 (Is a model and its architecture fixed when we start training it?)

To know about this, we need to understand what process is involved.

• Model architecture is to be fixed before training (Number of layers, number of neurons per layer)
• During the training, weights are updated
• Hyperparameters -> these are not learnt/controlled by the model, these are controlled externally which control how model learns

The architecture of the model is fixed before training starts, this meant,

• Number of layers are fixed
• Type of layers
• Connections between layers
• how functions run

During training, the model structure remains the same, weights and biases are updated inside the layers, the goal is to reduce the loss and achieve accuracy. Note that weights and biases are collectively known as parameters.