Graph Knowledge Distillation to Mixture of Experts

TL;DR

This paper introduces Routing-by-Memory, a Mixture-of-Experts model designed for graph knowledge distillation, achieving more consistent node classification performance compared to traditional MLPs.

Contribution

The paper proposes a novel MoE-based student model, Routing-by-Memory, that improves the consistency of knowledge distillation from GNNs across various datasets.

Findings

Routing-by-Memory outperforms MLP-based distillation in accuracy.

Experts in RbM specialize on different regions of the representation space.

The approach achieves more stable performance across multiple datasets.

Abstract

In terms of accuracy, Graph Neural Networks (GNNs) are the best architectural choice for the node classification task. Their drawback in real-world deployment is the latency that emerges from the neighbourhood processing operation. One solution to the latency issue is to perform knowledge distillation from a trained GNN to a Multi-Layer Perceptron (MLP), where the MLP processes only the features of the node being classified (and possibly some pre-computed structural information). However, the performance of such MLPs in both transductive and inductive settings remains inconsistent for existing knowledge distillation techniques. We propose to address the performance concerns by using a specially-designed student model instead of an MLP. Our model, named Routing-by-Memory (RbM), is a form of Mixture-of-Experts (MoE), with a design that enforces expert specialization. By encouraging each expert to specialize on a certain region on the hidden representation space, we demonstrate experimentally that it is possible to derive considerably more consistent performance across multiple datasets. Code available at https://github.com/Rufaim/routing-by-memory.