Learn Before
Why beam search is useful but not exact during decoding
Question: Explain why beam search is used to approximate the best output during inference and why it can still miss the true highest-scoring result.
Sample answer: Beam search is a heuristic search method for decoding. At each step, it keeps only the B most promising partial candidates according to a scoring function such as J(z), instead of expanding every possible continuation. Because some options are pruned early, the search may never examine the globally best complete output, so beam search is not guaranteed to return the score-maximizing result.
Key points:
- Beam search is a heuristic, approximate search method used during inference.
- It retains only the B highest-scoring partial candidates at each step.
- Because it prunes alternatives, it may miss the global maximum.
Rubric: The response should identify beam search as approximate search, explain that it keeps only the top B candidates during expansion, and state that this pruning means the optimum is not guaranteed.
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 set of candidates does beam search keep after each expansion?
True or False: Beam search is guaranteed to return the highest-scoring sequence under its scoring rule.
Beam search is a form of _____ search that keeps only the K highest-scoring partial candidates at each step.
In beam search, which partial candidates are kept after each expansion step?
Beam search is always guaranteed to return the sequence with the highest value of Score_A.
In beam search, the algorithm retains only the top _____ partial candidates at each step.
Match each beam-search concept with its description.
Put the beam-search inference steps in a sensible order.
Why is beam search considered approximate rather than exact?
Beam Search as an Approximate Decoder
Beam search may fail to _____ the sequence with the maximum Score_A(S).
Match each beam search property to its main consequence.
Put the beam search reasoning in the right order.
Why beam search is useful but not exact during decoding
Why a beam-search run can miss the best-scoring result
Beam search does not guarantee the best sequence