Case context: An engineering team is building an autonomous driving system to detect pedestrians and predict their walking direction. Instead of training a single end-to-end model, they decide to decompose the task: first, segment the pedestrian's body; second, locate their feet; and third, estimate their orientation. They explicitly code these stages as sequential components in a pipeline.

Question: Explain why the team's decision to explicitly code these subtask steps aids the overall learning algorithm, according to Ng's principles of task decomposition.

Sample answer: By breaking the complex task (pedestrian direction prediction) into simpler subtasks (segmentation, feet location, orientation estimation) and explicitly coding the execution path, the team supplies the learning algorithm with crucial prior knowledge. This prior knowledge constraints the structure, enabling the system to learn the full task more efficiently than an end-to-end model starting from scratch.

Key points:

• The complex prediction task is decomposed into simpler subtasks.
• Coding subtask steps explicitly gives the algorithm prior knowledge.
• The resulting prior knowledge helps the model learn the target task more efficiently.

Rubric: The response must identify that: 1. The team has decomposed a complex task into simpler subtasks (segmentation, feet location, orientation). 2. Explicitly coding these steps embeds prior knowledge into the pipeline. 3. This prior knowledge makes the learning process more efficient.