Emergent weight morphologies in deep neural networks

TL;DR

This paper demonstrates that deep neural networks develop emergent, periodic weight structures during training, independent of data, which influence their performance and raise considerations for AI security.

Contribution

The study introduces a theoretical framework predicting emergent weight morphologies in neural networks, supported by numerical experiments across various datasets.

Findings

Emergent periodic channel structures in weights during training

Theoretical prediction of instability leading to emergence

Impact on neural network performance

Abstract

Whether deep neural networks can exhibit emergent behaviour is not only relevant for understanding how deep learning works, it is also pivotal for estimating potential security risks of increasingly capable artificial intelligence systems. Here, we show that training deep neural networks gives rise to emergent weight morphologies independent of the training data. Specifically, in analogy to condensed matter physics, we derive a theory that predict that the homogeneous state of deep neural networks is unstable in a way that leads to the emergence of periodic channel structures. We verified these structures by performing numerical experiments on a variety of data sets. Our work demonstrates emergence in the training of deep neural networks, which impacts the achievable performance of deep neural networks.