When can a staged pipeline outperform a single direct model?
Question: What data condition can make a staged pipeline a better choice than a single end-to-end model?
Sample answer: A staged pipeline can be better when there is abundant training data for the intermediate tasks, so each stage can be learned reliably before the final output is produced.
Key points:
- Enough data exists for the intermediate steps
- The intermediate parts can be trained with reliable supervision
- That extra data can make the staged approach stronger than a direct one
Rubric: Answers must state that plenty of data is available to train the intermediate stages or modules.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Why Labeled Perception Data Can Make a Staged Driving System Easier
When is a staged model design especially attractive?
A staged design can be better when each stage has usable training data.
Available data for _____ in a staged workflow
Match staged system concepts
Choosing a Multi-Stage Pipeline When Subtask Labels Are Abundant
When a Modular Pipeline Can Be Competitive
Choose a pipeline structure when intermediate labels are abundant
When can a staged pipeline outperform a single direct model?
Which example is an intermediate module in a larger machine learning pipeline?
Limited data makes a multi-stage pipeline the best choice.