Stefan Liejnen' Abstract

Title: A Taxonomy of the Neural Network Zoo


Abstract: The past decade has witnessed a spectacular revival of academic and

commercial interest in deep learning. Numerous applications of artificial intelligence in

general and neural networks model in particular, have led to a variety of network types

and topologies. The Asimov Institute has recently published an update of the Neural

Network Zoo, an overview of the most commonly used models in deep learning.

Whereas some models are engineered to provide optimal learning of a particular task,

others trade performance for universality by staying close to their biological inspiration.

Upon further inspection, this taxonomy lays out several dimensions of depth, and points

to a more general dynamics that could underly networked information processing

systems.