Overview

Overview

A neural net consists of multiple layers, each of which (potentially) operates on input differently. We represent these objects with NeuralNet and Layer.

Index

PublicInterface

MIPVerify.Layer — Type.
abstract type Layer

Supertype for all types storing the parameters of each layer. Inherit from this to specify your own custom type of layer. Each implementation is expected to:

  1. Implement a callable specifying the output when any input of type JuMPReal is provided.

source
MIPVerify.NeuralNet — Type.
abstract type NeuralNet

Supertype for all types storing the parameters of a neural net. Inherit from this to specify your own custom architecture. Each implementation is expected to:

  1. Implement a callable specifying the output when any input of type JuMPReal is provided

  2. Have a UUID field for the name of the neural network.

source
MIPVerify.chain — Method.

An array of Layers is interpreted as that array of layer being applied to the input sequentially, starting from the leftmost layer. (In functional programming terms, this can be thought of as a sort of fold).

source