Intermediate Artificial Intelligence & Machine Learning
Q47 / 100

What is the vanishing gradient problem?

Correct! Well done.

Incorrect.

The correct answer is B) During backpropagation, gradients shrink exponentially through layers, making it hard to train early layers in deep networks

B

Correct Answer

During backpropagation, gradients shrink exponentially through layers, making it hard to train early layers in deep networks

Explanation

Sigmoid/tanh saturate with small gradients (≤ 0.25). Multiplying many <1 values → 0. Solutions: ReLU activations, residual connections (ResNet), batch normalization, LSTM gates, gradient clipping, careful initialization.

Progress
47/100