Should a better human trajectory be used when no optimal trajectory is known?
Case context: A learning algorithm produces a pilot trajectory. The team has a human pilot trajectory that performs better, but the team cannot show that the human trajectory is optimal.
Question: Decide whether the team can use the human trajectory for Optimization Verification and explain what the test can indicate.
Sample answer: Yes. The team can use the human trajectory as y* because it is superior to the current learning algorithm's trajectory, even though it is not known to be optimal. The resulting test can indicate whether improving the optimization algorithm or improving the scoring function is more promising.
Key points:
- The human trajectory is better than the current algorithm's output.
- Proven optimality is unnecessary.
- The human trajectory can serve as y*.
- The test compares the promise of improving optimization and scoring.
Rubric: The response should approve using the human trajectory based on its superiority, avoid claiming that it is optimal, and identify the optimization-versus-scoring decision supported by the test.
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
What makes a human pilot trajectory suitable for Optimization Verification?
Optimization Verification requires a provably optimal comparison output.
The comparison output must be _____ to the current algorithm's output.
Match each Optimization Verification concept with its role.
Order the reasoning for using a non-optimal human trajectory.
Explain why superiority, rather than optimality, is sufficient for this test.
Should a better human trajectory be used when no optimal trajectory is known?
What minimum property must y* have in Optimization Verification?
What conclusion is justified when T_human is good but not optimal?
A better-than-current output can guide the optimization-versus-scoring decision.