Supervised Learning in Non-Markovian Environments with MongoDB and Node.js

Source: Deep Learning on Medium


TL;DR;
Every decision made by neurons, and networks — as a whole — , is stored in MongoDB. Decisions are timestamped pairs of inputs and outputs of neurons and networks. One could pick a previous decision made by a network, correct it, and use the new input and expected value pair to retrain the network. This effectively allows one to train neural networks retroactively.