Emergent weight morphologies in deep neural networks
This work addresses understanding deep learning mechanisms and security risks for AI systems, but it appears incremental as it builds on analogies to condensed matter physics without broad SOTA impact.
The authors tackled the problem of whether deep neural networks exhibit emergent behavior by showing that training leads to emergent weight morphologies, independent of data, resulting in periodic channel structures verified through numerical experiments.
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.