
Graph neural networks (GNNs) are often miscalibrated for link prediction tasks, exhibiting a mixed behavior of overconfidence in negative predictions and underconfidence in positive ones. The proposed method, IN-N-OUT, effectively calibrates GNNs for link prediction tasks.


coremsg

in-n-out-calibrating-graph-neural-networks-for-link-prediction


In-n-Out: Calibrating Graph Neural Networks for Link Prediction


title_rewrite


GNNs for link prediction exhibit miscalibration, requiring calibration methods like IN-N-OUT to improve accuracy.
