Adaptive Negative Scheduling for Graph Contrastive Learning

Adnan Ali, Jinlong Li, Syed Muhammad Israr, Ali Kasif Bashir · 2026

Read Adaptive Negative Scheduling for Graph Contrastive Learning by Adnan Ali, Jinlong Li, Syed Muhammad Israr, Ali Kasif Bashir (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-03076.txt.

Description

Adnan Ali, Jinlong Li, Syed Muhammad Israr, Guanghua Yang, Ali Kashif Bashir, and Mohammed Ameen improve graph contrastive learning by scheduling which negative examples a model sees, and when. Negatives drive this self-supervised method, yet past work focused on picking them, not on controlling exposure during training.

Their AdNGCL framework keeps a bounded pool per anchor and rechecks hardness as representations change. A scheduler adjusts exposure to easy, medium, or hard negatives by category response, while refresh steps drop suspect items and add variety. Across ten benchmarks it led on eight datasets by mean score.

Catalog record

CREATORAdnan Ali, Jinlong Li, Syed Muhammad Israr, Ali Kasif Bashir
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03076

Browse the full library · About the Tower of Babel