Closed SSDUT-Caiyq closed 8 months ago
If the prototype is a simple linear layer, how can the procedure for computing the exponent (exp(1/\tao_s ...) / \sum k' exp(...) ) in Equation 4 be implemented?
Actually, we do not recommend using linear layers, which could easily overfit to Old classes and find it hard to discover novel ones.
If the prototype is a simple linear layer, how can the procedure for computing the exponent (exp(1/\tao_s ...) / \sum k' exp(...) ) in Equation 4 be implemented?