When deploying machine learning systems, developers must distinguish between the phase where a neural network learns and the phase where it is put to work. What is the fundamental difference between the training phase and the inference phase of a model's lifecycle?

The machine learning lifecycle is divided into training and inference phases. During training, a model learns the underlying patterns in a dataset by iteratively updating its parameters (weights and biases) via backpropagation to minimize a loss function. This phase is computationally intensive and requires substantial hardware resources, such as GPUs or TPUs. Once trained, the model's parameters are frozen. In the inference phase, the trained model is deployed to make predictions or decisions on new, unseen data. Inference requires only a single forward pass, making it significantly faster and less resource-intensive than training. - Distractor B is incorrect because training is highly resource-intensive due to backpropagation, while inference is computationally lightweight. - Distractor C is incorrect because training uses the training dataset to build the model, while testing and inference are performed on separate datasets to evaluate and use the model. - Distractor D is incorrect because both training and inference are required for both classification and regression tasks.