Recent works have proven the feasibility of fast and accurate time series classification methods based on randomized convolutional kernels [5, 32]. Concerning graph-structured data, the majority of randomized graph neural networks are based on the Echo State Network paradigm in which single layers or the whole network present some form of recurrence [7, 8]. This paper aims to explore a simple form of a randomized graph neural network inspired by the success of randomized convolutions in the 1-dimensional domain. Our idea is pretty simple: implement a no-frills convolutional graph neural network and leave its weights untrained. Then, we aggregate the node representations with global pooling operators, obtaining an untrained graph-level representation. Since there is no training involved, computing such representation is extremely fast. We then apply a fast linear classifier to the obtained representations. We opted for LS-SVM since it is among the fastest classifiers available. We show that such a simple approach can obtain competitive predictive performance while being extremely efficient both at training and inference time.
An Untrained Neural Model for Fast and Accurate Graph Classification
Gallicchio C.;Sperduti A.
2023-01-01
Abstract
Recent works have proven the feasibility of fast and accurate time series classification methods based on randomized convolutional kernels [5, 32]. Concerning graph-structured data, the majority of randomized graph neural networks are based on the Echo State Network paradigm in which single layers or the whole network present some form of recurrence [7, 8]. This paper aims to explore a simple form of a randomized graph neural network inspired by the success of randomized convolutions in the 1-dimensional domain. Our idea is pretty simple: implement a no-frills convolutional graph neural network and leave its weights untrained. Then, we aggregate the node representations with global pooling operators, obtaining an untrained graph-level representation. Since there is no training involved, computing such representation is extremely fast. We then apply a fast linear classifier to the obtained representations. We opted for LS-SVM since it is among the fastest classifiers available. We show that such a simple approach can obtain competitive predictive performance while being extremely efficient both at training and inference time.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
