A deep learning model for speech recognition suffers from adversarial noise. Which mechanism limits its ability to generalize from training data?

💡 Explanation

Acoustic invariance promotes over-reliance on superficial acoustic features, and adversarial noise exacerbates this. This degrades the distributional consistency because the model learns spurious correlations that do not generalize. Therefore, acoustic invariance limits robustness to novel noisy data, rather than improving feature selection or gradient stability.

• Why does a lexicographer consult a large text corpus when updating a dictionary entry?
• Why does communication using only emoji fail to convey complex legal arguments online?
• Why does a double dissociation study effectively localize distinct cognitive functions after brain damage?
• Why does inconsistent character spacing within a digital headline render it less readable, even if the letterforms themselves are clear?
• What distinguishes coarticulation involving nasal consonants from vowel-vowel coarticulation in vocal production?
• Why does a statistical parser, trained on a limited corpus of medical texts, perform poorly on general English text?