Search for: All records

Self-supervised learning of graph neural networks (GNN) is in great need because of the widespread label scarcity issue in real-world graph/network data. Graph contrastive learning (GCL), by training GNNs to maximize the correspondence between the representations of the same graph in its different augmented forms, may yield robust and transferable GNNs even without using labels. However, GNNs trained by traditional GCL often risk capturing redundant graph features and thus may be brittle and provide sub-par performance in downstream tasks. Here, we propose a novel principle, termed adversarial-GCL (\textit{AD-GCL}), which enables GNNs to avoid capturing redundant information during the training by optimizing adversarial graph augmentation strategies used in GCL. We pair AD-GCL with theoretical explanations and design a practical instantiation based on trainable edge-dropping graph augmentation. We experimentally validate AD-GCL by comparing with the state-of-the-art GCL methods and achieve performance gains of up-to~14\% in unsupervised, ~6\% in transfer and~3\% in semi-supervised learning settings overall with 18 different benchmark datasets for the tasks of molecule property regression and classification, and social network classification. 

Abstract Despite major progress in the investigation of boron cluster anions, direct experimental study of neutral boron clusters remains a significant challenge because of the difficulty in size selection. Here we report a size‐specific study of the neutral B₉cluster using threshold photoionization with a tunable vacuum ultraviolet free electron laser. The ionization potential of B₉is measured to be 8.45±0.02 eV and it is found to have a heptagonal bipyramidD_7hstructure, quite different from the planar molecular wheel of the B₉^‐anionic cluster. Chemical bonding analyses reveal superior stability of the bipyramidal structure arising from delocalized σ and π bonding interactions within the B₇ring and between the B₇ring and the capping atoms. Photoionization of B₉breaks the single‐electron B‐B bond of the capping atoms, which undergo off‐axis distortion to enhance interactions with the B₇ring in the singlet ground state of B₉⁺. The single‐electron B‐B bond of the capping atoms appears to be crucial in stabilizing theD_7hstructure of B₉. This work opens avenues for direct size‐dependent experimental studies of a large variety of neutral boron clusters to explore the stepwise development of network structures.