Nonparametric Neural Networks via Retrieval-Centric Deep Learning

This method replaces fixed-size weight matrices with a growing network that stores key-value pairs for every training data point. It utilizes functional gradient-based learning rules for kernelized attention layers to enable efficient retrieval and recombination during inference.