Google

A pipeline can fail because a later component is doing as well as it can given its inputs, but those inputs do not contain enough information. Pipeline redesign should ask what other information a skilled human would need.

Missing Information in Pipeline Inputs

Machine Learning Yearning, a free ebook from Andrew Ng, teaches you how to structure Machine Learning projects.
https://www.deeplearning.ai/machine-learning-yearning/

Note: The content of the book is aligned with the Coursera Deeplearning.ai specialization.  https://www.deeplearning.ai/deep-learning-specialization/ 

Machine Learning Yearning (Deeplearning.ai)

If a skilled human driver needs lane-marking locations, the self-driving pipeline can be redesigned by adding a lane-marking detector. This supplies important missing information to path planning.

Adding a Lane-Marking Detector to a Self-Driving Pipeline

Best Question for Pipeline Redesign

Component Performance vs. Input Quality

Using a _____ as a Baseline for Missing Info

Concepts in Pipeline Information Deficiencies

Resolving Missing Information in Pipelines

Question:
Explain why a pipeline component might fail despite functioning optimally given its inputs, and describe the recommended approach for redesigning the pipeline in this scenario.

Sample answer:
A pipeline component might fail even if it is performing perfectly on its inputs because those inputs simply do not contain enough information to solve the problem. For example, if a path-planning algorithm only knows the location of cars but not pedestrians, it will fail to drive safely, even if the algorithm itself is flawless. The recommended approach for redesign is to ask what other information a skilled human would need to perform the same task. By identifying the information a human uses, engineers can determine what new inputs or preceding components need to be added to the pipeline.

Key points:
- A component can only perform as well as its inputs allow.
- Missing information causes failure regardless of the algorithm's quality.
- Redesign should involve asking what a skilled human would need.
- The pipeline must be updated to provide this newly identified information.

Rubric:
A strong answer will explain that optimal performance is constrained by input quality and will explicitly mention using a 'skilled human' as a benchmark for determining what new information to add.

Diagnosing Pipeline Component Failures

Case context:
You are building a self-driving car pipeline. The 'Plan path' component receives the location of pedestrians and other cars from earlier components. However, the path planning is still erratic and unsafe. You evaluate the planner and determine the 'Plan path' algorithm is mathematically optimal for the inputs it currently receives.

Question:
Based on the concept of missing information, what is likely the cause of this failure, and what specific question should you ask to guide your pipeline redesign?

Sample answer:
The likely cause is that the inputs (locations of pedestrians and cars) do not contain enough information to plan a safe path. For example, it might be missing lane markings or traffic light states. To guide the redesign, I should ask: 'What other information, other than the outputs from the earlier components, would a skilled human driver need to plan a safe path?'

Key points:
- The current inputs are insufficient for the task.
- The algorithm is not the problem.
- Ask what a skilled human driver would need.
- Identify potential missing inputs (e.g., lane markings).

Rubric:
The answer must correctly diagnose the issue as missing input information and state the key question regarding what a skilled human driver would need.

Self-Driving Path Planner Case Study

Question:
If a pipeline component is failing because its inputs lack sufficient information, what specific benchmark or comparison should you use to determine what new information to add to the pipeline?

Sample answer:
You should use a skilled human as a benchmark by asking what additional information a skilled human would need to successfully complete the same task.

Key points:
- Use a skilled human as a reference point.
- Ask what information the human requires for the task.

Rubric:
The answer must explicitly mention asking what a skilled human would need to perform the task.

Learn Before

Related