<ul><li>T-CIL is a novel temperature scaling approach for class-incremental learning without a validation set for old tasks.</li><li>It leverages adversarially perturbed exemplars from memory to improve model confidence calibration.</li><li>The key idea of T-CIL is to perturb exemplars more strongly for old tasks than for the new task based on feature distance.</li><li>T-CIL outperforms various baselines in terms of calibration and can be integrated with existing class-incremental learning techniques.</li></ul>

T-CIL: Temperature Scaling using Adversarial Perturbation for Calibration in Class-Incremental Learning

Discover more