Graph Neural Networks for Learning Equivariant Representations of Neural Networks

TL;DR

This paper introduces a graph neural network approach to learn permutation-equivariant representations of neural networks, effectively handling diverse architectures and outperforming existing methods in various tasks.

Contribution

The authors propose representing neural networks as computational graphs to leverage GNNs and transformers for permutation-equivariant encoding, addressing limitations of prior approaches.

Findings

Outperforms state-of-the-art methods across tasks

Handles diverse neural network architectures

Effective in classification, editing, and generalization prediction

Abstract

Neural networks that process the parameters of other neural networks find applications in domains as diverse as classifying implicit neural representations, generating neural network weights, and predicting generalization errors. However, existing approaches either overlook the inherent permutation symmetry in the neural network or rely on intricate weight-sharing patterns to achieve equivariance, while ignoring the impact of the network architecture itself. In this work, we propose to represent neural networks as computational graphs of parameters, which allows us to harness powerful graph neural networks and transformers that preserve permutation symmetry. Consequently, our approach enables a single model to encode neural computational graphs with diverse architectures. We showcase the effectiveness of our method on a wide range of tasks, including classification and editing of implicit neural representations, predicting generalization performance, and learning to optimize, while consistently outperforming state-of-the-art methods. The source code is open-sourced at https://github.com/mkofinas/neural-graphs.