The function is used to convert logits into probabilities in a multi-class classification problem.

The function used to convert logits into probabilities in a multi-class classification problem is the Softmax function.

Here are the steps:

1. Compute the logits: These are the raw predictions that a classification model generates, which are typically in the form of real numbers, positive or negative, and not constrained in any way.

2. Apply the softmax function: The softmax function takes an N-dimensional vector of real numbers and transforms it into a vector of real number in range (0, 1) which add up to 1. The function is given by the formula:

   S(y_i) = e^(y_i) / Σ e^(y_j) for j = 1 to N

   where:

   • S(y_i) is the output of the softmax function for the i-th element
   • y_i is the i-th element of the input vector
   • e is the base of the natural logarithm (approximately equal to 2.71828)
   • N is the number of classes (i.e., the length of the input vector)

3. Interpret the output: The output of the softmax function can be interpreted as probabilities. The higher the value, the more likely the corresponding class is the correct one.

This is how logits are converted into probabilities in a multi-class classification problem.