Generalizability and robustness evaluation of attribute-based zero-shot learning

The paper introduces the concepts of generalizability and robustness in attribute-based zero-shot learning (ZSL) in order to test how much the performance results of ZSL models are affected by the “splits” chosen and the statistical properties of the classes and attributes used in training.
The paper reports the results of a series of experiments to test ZSL models with different splits.

The article is available here: doi.org/10.1016/j.neunet.2024.106278