This Blog is about how deep learning models work.
Design and implement experiments to understand various recurrent
neural network architectures and training challenges. Students will:
1. Implement a basic Recurrent Neural Network (RNN) and
train it on a sequential dataset.
2. Apply Backpropagation Through Time (BPTT) and
analyze the effect of long sequence lengths on gradient
behavior.
3. Observe Vanishing and Exploding Gradients and apply
mitigation strategies such as gradient clipping.
4. Implement Truncated BPTT and compare its performance
with full BPTT.
5. Implement and train LSTM and GRU models,
demonstrating
Tools/Technology To Be Used – Python, TensorFlow / Keras or
PyTorch, NumPy, Matplotlib / Seaborn, NLTK / spaCy
Total Hours of Problem Definition Implementation: 2 Hours
Total Hours of Engagement: 6 Hours
Post Laboratory Work Description: Implementation files for
RNN, GRU, LSTM, encoder–decoder, and attention models.
Share
X / Twitter