Analyzing Data Requirements for End-to-End Systems
Question: Why is a pure end-to-end approach particularly difficult to train for autonomous driving, as opposed to tasks where paired data is more naturally available?
Sample answer: A pure end-to-end approach for autonomous driving requires a massive dataset of paired (Image, Steering Direction) data. Gathering this specific type of data is highly time-consuming and expensive because it cannot be easily scraped or synthesized. Instead, it requires deploying a fleet of specially-instrumented cars and having people drive them extensively to capture a sufficiently wide range of possible driving scenarios.
Key points:
- Requires large datasets of (Image, Steering Direction) pairs
- Data collection is highly time-consuming and expensive
- Requires a fleet of specially-instrumented cars
- Requires immense amounts of driving to cover diverse scenarios
Rubric: 1 point for identifying the need for (Image, Steering Direction) pairs. 1 point for stating data collection is time-consuming and expensive. 1 point for mentioning the requirement of specially-instrumented cars. 1 point for noting the huge amount of driving required to cover various scenarios.
0
1
References
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Machine Learning Yearning (Deeplearning.ai)
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
What type of labeled data pairs does a pure end-to-end autonomous driving system require for training?
End-to-end learning systems are not always a good choice, even when they can be remarkably successful with abundant data.
To train a pure end-to-end autonomous driving system, you need a large dataset of (Image, _____) pairs.
Match each end-to-end learning domain to the specific labeled input-output data pairs required to train it.
Order the causal chain explaining why a pure end-to-end autonomous driving system is difficult to train.
Why does collecting training data for a pure end-to-end autonomous driving system require specially instrumented cars?
End-to-end learning systems tend to perform well when large amounts of labeled data exist for both the input end and the output end.
When sufficient labeled input-output data is not available, you should approach end-to-end learning with great _____.
Match each data collection challenge for end-to-end autonomous driving to the consequence described in Machine Learning Yearning.
Order the reasoning steps a practitioner should follow when evaluating whether to use a pure end-to-end approach for a new problem.
Analyzing Data Requirements for End-to-End Systems
Evaluating End-to-End Learning Viability
Resource Requirements for Autonomous Data Collection