🤖 AI Summary
This work addresses catastrophic forgetting and unreliable pseudo-labeling in semi-supervised class-incremental learning without storing past samples by proposing a continual learning framework based on a frozen backbone network and learnable adapters. The method generates hallucinated features as substitutes for historical data through soft-weighted geometric calibration and introduces a geometric structure alignment objective to stabilize the topology of the feature space, effectively mitigating feature drift. It integrates graph-propagated labeling, Gaussian sampling, teacher-student head matching, and classifier prototype anchoring. Evaluated on CIFAR-100, CUB-200, and ImageNet-R, the approach significantly outperforms existing semi-supervised and supervised class-incremental methods using only 1%–10% labeled data, with particularly notable gains in fine-grained, low-label regimes.
📝 Abstract
Semi-supervised Class Incremental Learning (SSCIL) is a severe challenge for neural networks, and it is hardest in the exemplar-free setting where no past data may be stored. Existing methods forget catastrophically due to feature drift, and their pseudo-labels become increasingly unreliable as the label space grows. In this paper, we propose MAGIC (Manifold Anchoring and Geometric Incremental Calibration), a framework that stabilizes plasticity without storing exemplars. MAGIC's design centers on two components. The first is Soft-Weighted Geometry Calibration (SWGC), which uses graph-based label propagation on the learner's plastic feature space to weight and calibrate class means and variances computed on the frozen backbone; from these calibrated Gaussians, we sample phantom features that stand in for data from previous tasks. The second is a Geometric Structural Alignment (GSA) objective that preserves representation topology by matching the relational structure of student and teacher heads and aligning feature anchors with the fixed classifier prototypes, locking the orientation of the feature space. Together, these constraints keep the adapter from drifting, so geometric relations between classes remain stable as new classes arrive. We implement MAGIC with a frozen ResNet-18 backbone and a learnable plastic adapter. Across CIFAR-100, CUB-200, and ImageNet-R, at label ratios of 1%, 5%, and 10%, MAGIC improves average incremental accuracy over most of the supervised CIL methods equipped with FixMatch and native SSCIL baselines; the largest gains occur in the fine-grained, low-label setting, where confidence thresholding fails most clearly.