Case Study

Find out why an approximate decoder missed a better-scoring output.

Case context: A handwriting recognition system receives an image x and assigns each candidate text string T a score Scorex(T) = P(T|x). Because checking every possible string is infeasible, the team uses an approximate decoding method to search for the best output. On one test image, the system returns T1. Later, an engineer verifies that another valid transcription T2 has a higher score, so Scorex(T2) > Scorex(T1).

Question: Why did the decoder return T1 instead of T2, and what does this tell you about the decoding method?

Sample answer: The decoder returned T1 because the approximate search procedure did not find the highest-scoring transcription. This means the decoding method is not guaranteed to recover the true maximizer of Scorex(T) even when a better-scoring candidate exists.

Key points:

  • The approximate decoder missed the best-scoring transcription.
  • Approximate search methods can return a good but not optimal output.
  • The higher-scoring transcription was missed because of search limitations, not because the score was computed incorrectly.

Rubric: The response must explain that: 1. The approximate search method failed to identify T2 even though it had the larger score. 2. This shows the method does not guarantee finding the absolute best output under the score function.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Data Science

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI