Activations
NeuralVerifier.Encoding.relu — Methodrelu(x)Apply the ReLU activation function.
NeuralVerifier.Encoding.sigmoid — Methodsigmoid(x)Apply a piecewise linear approximation to x.
Output Activations
NeuralVerifier.Encoding.sigmoid_bool — Methodsigmoid_bool(x)Apply the piecewise softmax function with 2 splits for binary classification.
NeuralVerifier.Encoding.softmax — Methodsoftmax(x::Array)Apply the softmax operation. As we do not require the gradients, this is simply checking if each input is_argmax, if true, return the index label.