The defining computational activity of model training is the iterative adjustment of the model's trainable parameters—principally weights and biases —to reduce an objective or loss function. During training, input examples are processed through the model, prediction error is calculated, gradients are derived, and an optimization algorithm updates parameters. Repetition across batches and epochs gradually produces a model that captures useful patterns and relationships in the training data.
Cisco identifies Training as a specific AI/ML workload type and separately requires knowledge of the overall AI lifecycle. Cisco's AI/ML operations guidance describes model training as the process through which a model learns patterns, features, representations, and relationships from data and places training before deployment and production operation.
Option A represents deployment or application integration after training. Option B normally belongs to problem definition and project-planning stages before model development. Option D is a production-monitoring activity used to determine whether a deployed model's behavior has degraded or changed because incoming data or underlying relationships have shifted.
Therefore, C identifies the central learning mechanism occurring during the training stage.
Study Guide Reference: 1.0 AI Fundamentals and Applications — 1.1.b Training; 1.2 Describe the AI lifecycle.
===============
