Supporting Input Convex Neural Networks in MathOptAI.jl

MathOptAI.jl is a Julia package for embedding machine learning models into optimization problems built using JuMP. These embeddings work as optimization surrogates and replace the terms in the original formulations with approximations represented by the added variables and constraints. Input convex neural networks have some useful properties such as convexity that enable approximations with similar quality with much simpler embedded formulations. This GSOC project will extend MathOptAI to add support for input convex neural networks.