A deep neural network's performance degrades due to adversarial noise injected during training. Which consequence dominates if the network lacks explicit denoising layers?

💡 Explanation

If a neural network lacks denoising layers, adversarial noise directly corrupts learned features, because the network cannot distinguish signal from noise. Therefore, robustness to perturbations will sharply decrease, rather than overfitting, which is a separate concern, or gradients vanishing, which can be independent.

• A robot processes spoken commands to manipulate objects in a shared workspace. If the referring expression 'put that there' is uttered, which consequence follows?
• During a video conference with international participants, which outcome most likely explains misinterpretations of co-speech gestures that point to locations on a shared digital map?
• Why does a speaker of American English struggle to distinguish certain Hindi retroflex consonants?
• Why does head-final order correlate harmonically with SOV word order, rather than SVO or VSO, in cross-linguistic typology?
• Why does a neural machine translation (NMT) system, despite achieving a high BLEU score, still sometimes produce translations that are semantically inaccurate or nonsensical, especially with idiomatic expressions?
• Within a convolutional error-correcting code used in satellite communication, which mechanism explains why increasing the constraint length improves bit error rate?