Bidirectional RNN in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a Bidirectional RNN?

A Bidirectional RNN is a type of recurrent neural network that processes data in both forward and backward directions. This helps the model understand context from past and future data points.

beginner

Why use Bidirectional RNNs instead of regular RNNs?

Bidirectional RNNs capture information from both past and future inputs, improving understanding of context, which is useful in tasks like speech recognition and language modeling.

intermediate

How does TensorFlow implement a Bidirectional RNN?

TensorFlow uses the `tf.keras.layers.Bidirectional` wrapper around RNN layers like LSTM or GRU to create a Bidirectional RNN that runs two RNNs in opposite directions and combines their outputs.

intermediate

What are the outputs of a Bidirectional RNN layer?

The outputs are usually the concatenation of the forward and backward RNN outputs at each time step, giving richer information for the next layers.

beginner

Write a simple TensorFlow code snippet to create a Bidirectional LSTM layer.

import tensorflow as tf
model = tf.keras.Sequential([
  tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(64, return_sequences=True), input_shape=(None, 10))
])

What does a Bidirectional RNN process?

A Random data order

B Only forward data

C Only backward data

D Data in both forward and backward directions

Bidirectional RNNs process data in both forward and backward directions to capture full context.

Which TensorFlow layer wraps an RNN to make it bidirectional?

A tf.keras.layers.Bidirectional

B tf.keras.layers.Dense

C tf.keras.layers.Conv1D

D tf.keras.layers.Dropout

The Bidirectional wrapper layer makes an RNN run forwards and backwards.

What is a common use case for Bidirectional RNNs?

A Speech recognition

B Image classification

C Sorting numbers

D Simple linear regression

Bidirectional RNNs are great for tasks like speech recognition where context from both past and future is important.

What does the output of a Bidirectional RNN usually contain?

A Only forward outputs

B Concatenated forward and backward outputs

C Only backward outputs

D Random noise

Outputs from both directions are combined to provide richer information.

In TensorFlow, which argument is important to keep the sequence output in an LSTM layer inside a Bidirectional wrapper?

A dropout=0.5

B activation='relu'

C return_sequences=True

D units=32

Setting return_sequences=True ensures the LSTM outputs a sequence for each time step.

Explain how a Bidirectional RNN works and why it is useful.

Think about reading a sentence forwards and backwards to understand it better.

Describe how to implement a Bidirectional RNN in TensorFlow with an example.

Remember the wrapper layer that runs two RNNs in opposite directions.