Graph Embedding for Neural Architecture Search with Input-Output Information

0560713 - ÚI 2023 RIV US eng C - Konferenční příspěvek (zahraniční konf.)
Suchopárová, Gabriela - Neruda, Roman
Graph Embedding for Neural Architecture Search with Input-Output Information.
Auto-ML Conf 2022: Accepted Papers: Late-Breaking Workshop. Baltimore: AutoML Conference, 2022.
[Auto-ML 2022: International Conference on Automated Machine Learning /1./. Baltimore (US), 25.07.2022-27.07.2022]
Grant ostatní: Ministerstvo školství, mládeže a tělovýchovy - GA MŠk(CZ) LM2018140
Institucionální podpora: RVO:67985807
Klíčová slova: machine learning * neural architecture search * meta-learning * graph neural networks * representation learning
Obor OECD: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

ZÁKLADNÍ ÚDAJE: Auto-ML Conf 2022: Accepted Papers: Late-Breaking Workshop. Baltimore: AutoML Conference, 2022. KONFERENCE: Auto-ML 2022: International Conference on Automated Machine Learning /1./. Baltimore (US), 25.07.2022-27.07.2022. ABSTRAKT: Graph representation learning has been widely used in neural architecture search as a part of performance prediction models. Existing works focused mostly on neural graph similarity without considering functionally similar networks with different architectures. In this work, we address this issue by using meta-information of input images and output features of a particular neural network. We extended the arch2vec model, a graph variational autoencoder for neural architecture search, to learn from this novel kind of data in a semi-supervised manner. We demonstrate our approach on the NAS-Bench-101 search space and the CIFAR10 dataset, and compare our model with the original arch2vec on a REINFORCE search task and a performance prediction task. We also present a semi-supervised accuracy predictor, and we discuss the advantages of both variants. The results are competitive with the original model and show improved performance.
Trvalý link: https://hdl.handle.net/11104/0333566