Universality conjecture for overparameterized neural networks as neural channels

An overparameterized neural network is a neural network with more parameters than are needed to fit its training data. A Neural Channel is a model that can approximately preserve its original behavior while efficiently storing and enabling the reception of arbitrary information through an associated transmission and reception procedure.

Neural-channel universality conjecture. Every overparameterized neural network is a Neural Channel.

The paper introduces Neural Channels as storage and communication channels implemented by trained models and indicates that the conjecture will be investigated empirically. Its resolution is not given in the supplied text.