Why is exploding gradient occurred?1 pointComputing gradient wrt hidden states requires many factors of weights and repeated gradient computationComputing gradient wrt inputs requires many factors of outputs and repeated gradient computationComputing gradient wrt weights requires many factors of hidden states and repeated gradient computationNone of the above

The exploding gradient problem occurs primarily due to the nature of backpropagation in training deep neural networks. Here's a step-by-step explanation:

1. Backpropagation: In the training process of neural networks, we use a method called backpropagation to compute the gradient of the loss function with respect to the weights of the network. This gradient is then used to update the weights to minimize the loss function.

2. Chain Rule: The backpropagation algorithm is essentially an application of the chain rule from calculus. It computes the gradient by propagating the gradient of the loss function backwards through the network. At each layer, the gradient is multiplied by the weights of the current layer and the derivative of the activation function.

3. Long sequences: When dealing with long sequences, the gradients have to be propagated through many layers. Each time the gradient is propagated to the previous layer, it is multiplied by the weights of the current layer.

4. Exploding gradients: If the weights are large or the sequence is long, the gradient can become very large, because it is repeatedly multiplied by the weights. This is known as the exploding gradient problem. It can lead to unstable training process and poor performance, because the large gradient can cause the weights to be updated in a very large step, potentially overshooting the minimum of the loss function.

So, the correct answer is: "Computing gradient wrt hidden states requires many factors of weights and repeated gradient computation".