NLP - Sequence Models for NLP
To enhance an RNN text classifier's ability to capture long-range dependencies and reduce vanishing gradient issues, which architecture change is most effective?
To enhance an RNN text classifier's ability to capture long-range dependencies and reduce vanishing gradient issues, which architecture change is most effective?
hard📝 Application Q9 of 15
Step-by-Step Solution
Solution:
Step 1: Understand vanishing gradient problem
SimpleRNNs struggle with long sequences due to vanishing gradients.Step 2: Identify architectures that mitigate this
LSTM and GRU units have gating mechanisms to preserve gradients over longer sequences.Step 3: Evaluate other options
Increasing units or adding Dense layers doesn't solve gradient issues; batch size affects training but not vanishing gradients directly.Final Answer:
Replace SimpleRNN with an LSTM or GRU layer -> Option BQuick Check:
LSTM/GRU handle long dependencies better [OK]
Quick Trick: Use LSTM/GRU to fix vanishing gradients [OK]
Common Mistakes:
MISTAKES- Thinking more units fix vanishing gradients
- Adding Dense layers improves sequence memory
- Believing batch size affects gradient flow
Master "Sequence Models for NLP" in NLP
Want More Practice?
15+ quiz questions · All difficulty levels · Free
Free Signup - Practice All QuestionsMore NLP Quizzes