The paper proposes AD-NEv++, a three-stage neuroevolution-based method for multivariate anomaly detection. The method extends the previous AD-NEv framework by incorporating graph autoencoders as a new type of neural architecture in the neuroevolution process and defining a new layer of abstraction to consider optional layers, such as attention, skip connection, and dense connection.
The framework involves the simultaneous evolution of two populations: models and subspaces. The model population consists of neural network architectures that evolve during the neuroevolution process, whereas the subspace population defines subsets of input features. After the neuroevolution process, the framework yields a bagging technique-based ensemble model derived from single optimized architectures.
The experimental evaluation on widely adopted multivariate anomaly detection benchmark datasets shows that the models generated by AD-NEv++ outperform well-known deep learning architectures and neuroevolution-based approaches for anomaly detection. The results also demonstrate that AD-NEv++ can improve and outperform the state-of-the-art GNN (Graph Neural Networks) model architecture in all anomaly detection benchmarks.
Egy másik nyelvre
a forrásanyagból
arxiv.org
Mélyebb kérdések