핵심 개념
GNNs for link prediction exhibit miscalibration, requiring calibration methods like IN-N-OUT to improve accuracy.
초록
Deep neural networks, especially GNNs, often show miscalibration in link prediction tasks.
IN-N-OUT proposes a novel method to calibrate GNNs for link prediction by adjusting temperature scaling.
Experimental results demonstrate the effectiveness of IN-N-OUT in improving calibration and outperforming baselines.
The study highlights the importance of accurate calibration for reliable graph ML methods.
통계
GNNs are often miscalibrated in link prediction tasks.
IN-N-OUT consistently outperforms baselines in calibration experiments.
인용구
"IN-N-OUT significantly improves the calibration of GNNs in link prediction."
"GNNs are often overconfident in positive predictions and underconfident in negative ones."