With the inverted dropout technique, at test time:

In the inverted dropout technique, at test time, the neural network uses all of its neurons. This is different from the training phase, where dropout is applied and a certain percentage of neurons are randomly turned off to prevent overfitting.

Here are the steps:

1. During training, for each hidden layer, for each training sample, for each iteration, certain neurons are dropped or turned off randomly. This is done by setting their activations to zero during the forward pass. The probability of each neuron being dropped is usually set to a value such as 0.5.

2. The dropped neurons do not contribute to the training in both the forward and backward passes of the iteration.

3. The idea behind this is to prevent the model from relying too much on any one neuron and therefore overfitting the training data.

4. At test time, we want to get the most accurate predictions. So, we keep all the neurons active. However, to compensate for the increased number of active neurons, we scale down the output by the same dropout rate used during training. This is why it's called "inverted dropout".

5. So, if a dropout rate of 0.5 was used during training, the outputs are halved at test time. This ensures that for any given neuron, the expected output (on average) is the same as it was during training.

6. This technique allows the neural network to be less sensitive to the specific weights of neurons, resulting in a network that is capable of better generalization and is less likely to overfit the training data.