NAR-Former V2: Rethinking Transformer for Universal Neural Network Representation Learning

TL;DR

This paper introduces NAR-Former V2, a Transformer-based model that effectively learns neural network representations for both cell-structured and entire networks, outperforming GNN-based methods in latency estimation and matching state-of-the-art accuracy predictions.

Contribution

The paper proposes a novel Transformer-based model that incorporates GNN capabilities for universal neural network representation learning, improving generalization and performance over existing methods.

Findings

Surpasses GNN-based NNLP in latency estimation on NNLQP dataset.

Achieves comparable accuracy prediction results on NASBench101 and NASBench201 datasets.

Enhances Transformer with graph encoding and inductive learning for better generalization.

Abstract

As more deep learning models are being applied in real-world applications, there is a growing need for modeling and learning the representations of neural networks themselves. An efficient representation can be used to predict target attributes of networks without the need for actual training and deployment procedures, facilitating efficient network deployment and design. Recently, inspired by the success of Transformer, some Transformer-based representation learning frameworks have been proposed and achieved promising performance in handling cell-structured models. However, graph neural network (GNN) based approaches still dominate the field of learning representation for the entire network. In this paper, we revisit Transformer and compare it with GNN to analyse their different architecture characteristics. We then propose a modified Transformer-based universal neural network representation learning model NAR-Former V2. It can learn efficient representations from both cell-structured networks and entire networks. Specifically, we first take the network as a graph and design a straightforward tokenizer to encode the network into a sequence. Then, we incorporate the inductive representation learning capability of GNN into Transformer, enabling Transformer to generalize better when encountering unseen architecture. Additionally, we introduce a series of simple yet effective modifications to enhance the ability of the Transformer in learning representation from graph structures. Our proposed method surpasses the GNN-based method NNLP by a significant margin in latency estimation on the NNLQP dataset. Furthermore, regarding accuracy prediction on the NASBench101 and NASBench201 datasets, our method achieves highly comparable performance to other state-of-the-art methods.