2021 BRN Discussion, page-21563

Software.
US10783433B1 Method for training and self-organization of a neural network
https://worldwide.espacenet.com/patent/search/family/072516815/publication/US10783433B1?q=US10783433B1

Claim 7. A system for configuring a neural network for training and self-organization, the system comprising
a memory containing computer-readable instructions and a processor to execute the computer-readable instructions to cause the system to:receive sets of digital attributes at inputs of the neural network,
the neural network being adapted to perform multidimensional regression and comprising
a plurality of neurons, each neuron being capable of being directly connected to any other neuron;
randomly generate a training set and a testing set from an initial data set;
generate a control data set that is not included in the training and testing sets;
define a plurality of activation functions and interconnections of the neurons, each activation function being defined in terms of a plurality of activation function parameters;
define a plurality of p model parameters of the neural network, the model parameters being divided into groups of η model parameters, wherein all η model parameters in each group are calculated in parallel at m points, and wherein at each of a plurality of q iterations of training a search interval for each model parameter in each group is reduced by up to twice a predefined discretization value Δ q , the discretization value being calculated according to the formula
Δ q = 2 q - 1 ⁢ interval ( m - 1 ) q *,where interval denotes a relative interval value; ...


Thanks to our proprietary algorithm and disruptive machine learning technology, models built on Neuton are super compact, meaning that they consist of relatively few neurons and coefficients. The *actual algorithm is our IP, therefore, we cannot disclose it. Neuton results were compared against Caffe2, Tensor Flow+Keras, CNTK, Torch, Theano. Those networks showed very similar results.

A corollary to Ella's famous quote is that you can't keep secrets in patents.