Controlling Recurrent Neural Networks by Conceptors

TL;DR

This paper introduces conceptors, a neurodynamical mechanism that enables recurrent neural networks to learn, store, and manipulate multiple patterns and concepts robustly, bridging nonlinear dynamics with cognitive abstraction.

Contribution

It presents a novel conceptor framework that unifies nonlinear neural dynamics with conceptual abstraction, allowing flexible pattern management within a single neural system.

Findings

Enables learning and recognition of multiple patterns without interference.

Automatically filters neural noise during pattern processing.

Supports adding new patterns without disrupting existing ones.

Abstract

The human brain is a dynamical system whose extremely complex sensor-driven neural processes give rise to conceptual, logical cognition. Understanding the interplay between nonlinear neural dynamics and concept-level cognition remains a major scientific challenge. Here I propose a mechanism of neurodynamical organization, called conceptors, which unites nonlinear dynamics with basic principles of conceptual abstraction and logic. It becomes possible to learn, store, abstract, focus, morph, generalize, de-noise and recognize a large number of dynamical patterns within a single neural system; novel patterns can be added without interfering with previously acquired ones; neural noise is automatically filtered. Conceptors help explaining how conceptual-level information processing emerges naturally and robustly in neural systems, and remove a number of roadblocks in the theory and applications of recurrent neural networks.