Policy-GNN: Aggregation Optimization for Graph Neural Networks

Abstract

Graph data are pervasive in many real-world applications. Recently, increasing attention has been paid on graph neural networks (GNNs), which aim to model the local graph structures and capture the hierarchical patterns by aggregating the information from neighbors with stackable network modules. Motivated by the observation that different nodes often require different iterations of aggregation to fully capture the structural information, in this paper, we propose to explicitly sample diverse iterations of aggregation for different nodes to boost the performance of GNNs. It is a challenging task to develop an effective aggregation strategy for each node, given complex graphs and sparse features. Moreover, it is not straightforward to derive an efficient algorithm since we need to feed the sampled nodes into different number of network layers. To address the above challenges, we propose Policy-GNN, a metapolicy framework that models the sampling procedure and message passing of GNNs into a combined learning process. Specifically, Policy-GNN uses a meta-policy to adaptively determine the number of aggregations for each node. The meta-policy is trained with deep reinforcement learning (RL) by exploiting the feedback from the model. We further introduce parameter sharing and a buffer mechanism to boost the training efficiency. Experimental results on three real-world benchmark datasets suggest that Policy-GNN significantly outperforms the state-of-the-art alternatives, showing the promise in aggregation optimization for GNNs.

Graph neural architecture search: A survey

“…RS involves generating random submodels from the search space. RS has been tested by some researchers [24,46,66] for Graph-NAS. Unfortunately, even if this method can provide good results in terms of efficiency, it is not widely adopted because of its hazardous results, which are often less optimal than those generated by other methods.…”

“…However, this method entails the use of hundreds of GNN performance distribution and graph data characteristics to build a neural predictor. One can also use the buffer mechanism for graph representation learning [66] .…”

Reinforced Neighborhood Selection Guided Multi-Relational Graph Neural Networks

“…Overview. To address these problems, we propose a novel Recursive and Scalable Reinforcement Learning framework RSRL, upon traditional Reinforcement Learning based approaches [21,51,63], to not only update strategies through the learning environment but also the recursive structure can be used to quickly and accurately meet the accuracy requirements of different relations. Figure . 4 depicts the forest-based learning architecture.…”

“…This model uses a recurrent network to generate variable-length strings that describe the architectures of graph neural networks, and then trains the recurrent network with reinforcement learning to maximize the expected accuracy of the generated architectures. • Policy-GNN [51]: is a meta-policy framework that adaptively learns an aggregation policy to sample diverse iterations of aggregations for different nodes. To accelerate the learning process, we also use a buffer mechanism to enable batch training and parameter sharing mechanism to decrease the training cost.…”

“…GraphNAS [24] employs a search space covering sampling functions, aggregation functions and gated functions and uses RL to search graph neural architectures. Policy-GNN [51] formulates the GNN training problem as a Markov Decision Process, and can adaptively learn an aggregation policy to sample diverse iterations of aggregations for different nodes. However, neither GraphNAS nor Policy-GNN models considers heterogeneous neighborhoods in aggregation although they pay more attention to neural architecture searching.…”