For autonomous driving, machine learning can be used to detect cars and pedestrians, and labeled car or pedestrian image data is relatively easy to obtain. By contrast, a pure end-to-end approach would require a large dataset of image and steering-direction pairs that is time-consuming and expensive to collect, making it difficult to train. This data availability supports considering intermediate car and pedestrian detectors in a multi-stage, non-end-to-end pipeline.

Autonomous Driving Data Availability Favors Intermediate Detectors

Why consider a multi-stage pipeline?

Is a multi-stage pipeline structure potentially superior?

Available data for _____ in a pipeline

Match multi-stage pipeline concepts

Sequence the decision process for a multi-stage pipeline

Question:
Explain why having a lot of training data available for intermediate modules makes a multi-stage pipeline worth considering. How does this impact the potential superiority of the pipeline?

Sample answer:
A multi-stage pipeline becomes an attractive option when there is abundant data for intermediate modules because this data can be directly utilized to train those specific components. This structure can be superior because each intermediate module can be highly optimized and accurate due to the dedicated training data, which might be more effective than trying to train a single end-to-end model without such targeted intermediate supervision.

Key points:
- Abundant data available for intermediate modules
- Ability to train specific components effectively
- Superiority of the resulting pipeline structure

Rubric:
The essay should clearly state that available data allows for the training of individual intermediate modules, making the multi-stage approach viable and potentially superior due to targeted optimization.

Analyze the impact of intermediate module data

Case context:
You are building a system for autonomous vehicles. You have access to a massive dataset of bounding boxes around pedestrians, but only a moderate amount of data mapping raw camera pixels directly to steering commands.

Question:
Based on the availability of data for intermediate modules, what structural decision should you make for your machine learning pipeline, and why?

Sample answer:
I should consider using a multi-stage pipeline. The massive dataset of bounding boxes around pedestrians is an example of having a lot of data available for an intermediate module (a pedestrian detector). Because this structure allows me to use all that available data to effectively train the pedestrian detector, it could be superior to attempting an end-to-end approach with limited direct end-to-end data.

Key points:
- Bounding box data serves as intermediate module data
- A multi-stage pipeline is the recommended structure
- The structure uses the available data to train the intermediate detector

Rubric:
The learner must identify the bounding box data as intermediate module data and conclude that a multi-stage pipeline is the correct choice, citing the ability to use that data for training.

Decide on a pipeline structure based on available data

Question:
What specific condition regarding training data can make a multi-stage pipeline superior?

Sample answer:
A multi-stage pipeline can be superior if there is a lot of data available specifically to train its intermediate modules.

Key points:
- Availability of a lot of data
- Data is used for training intermediate modules

Rubric:
Answers must mention the availability of data for training intermediate modules.

What makes a multi-stage pipeline superior?

Which is an intermediate module?

Does limited data favor multi-stage pipelines?

If a lot of training data is available for intermediate modules in a pipeline, a multi-stage pipeline is worth considering. This structure can be superior because the available data can be used to train those intermediate modules.

Google

When choosing components for a non-end-to-end pipeline, one important factor is whether data can be collected easily to train each component.

Training Data Availability for Pipeline Component Selection

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/ 

Learn Before

Related

Learn After