# Revisiting Prototype Rehearsal for Exemplar-Free Continual Learning: Manifold-Aware Boundary Sampling with Adaptive Class-Balanced Loss

arXiv: https://arxiv.org/abs/2606.05695 · HTML: https://mlsanity.com/paper/2606.05695/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 48 references that are in the map.
- First posted on arXiv 4 Jun 2026.

## Builds on (most cited first, 27 of 27)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Learning without Forgetting](https://mlsanity.com/paper/1606.09282/) (2016)
- [A Comprehensive Survey of Continual Learning: Theory, Method and Application](https://mlsanity.com/paper/2302.00487/) (2023)
- [Class-incremental learning: survey and performance evaluation on image classification](https://mlsanity.com/paper/2010.15277/) (2020)
- [Deep Long-Tailed Learning: A Survey](https://mlsanity.com/paper/2110.04596/) (2021)
- [Online Continual Learning in Image Classification: An Empirical Survey](https://mlsanity.com/paper/2101.10423/) (2021)
- [Semantic Drift Compensation for Class-Incremental Learning](https://mlsanity.com/paper/2004.00440/) (2020)
- [Improving Calibration for Long-Tailed Recognition](https://mlsanity.com/paper/2104.00466/) (2021)
- [FeTrIL: Feature Translation for Exemplar-Free Class-Incremental Learning](https://mlsanity.com/paper/2211.13131/) (2022)
- [The Majority Can Help The Minority: Context-rich Minority Oversampling for Long-tailed Classification](https://mlsanity.com/paper/2112.00412/) (2021)
- [PyCIL: A Python Toolbox for Class-Incremental Learning](https://mlsanity.com/paper/2112.12533/) (2021)
- [A Survey on Long-Tailed Visual Recognition](https://mlsanity.com/paper/2205.13775/) (2022)
- [A Comprehensive Survey of Forgetting in Deep Learning Beyond Continual Learning](https://mlsanity.com/paper/2307.09218/) (2023)
- [A survey on learning from imbalanced data streams: taxonomy, challenges, empirical study, and reproducible experimental framework](https://mlsanity.com/paper/2204.03719/) (2022)
- [Elastic Feature Consolidation for Cold Start Exemplar-Free Incremental Learning](https://mlsanity.com/paper/2402.03917/) (2024)
- [Gradient Reweighting: Towards Imbalanced Class-Incremental Learning](https://mlsanity.com/paper/2402.18528/) (2024)
- [Long-Tailed Class Incremental Learning](https://mlsanity.com/paper/2210.00266/) (2022)
- [Resurrecting Old Classes with New Data for Exemplar-Free Continual Learning](https://mlsanity.com/paper/2405.19074/) (2024)
- [Exemplar-free Continual Representation Learning via Learnable Drift Compensation](https://mlsanity.com/paper/2407.08536/) (2024)
- [Steering Prototypes with Prompt-tuning for Rehearsal-free Continual Learning](https://mlsanity.com/paper/2303.09447/) (2023)
- [Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning](https://mlsanity.com/paper/2409.18265/) (2024)
- [Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers](https://mlsanity.com/paper/2406.18451/) (2024)
- [Task-Adaptive Saliency Guidance for Exemplar-free Class Incremental Learning](https://mlsanity.com/paper/2212.08251/) (2022)
- [Efficient Augmentation for Imbalanced Deep Learning](https://mlsanity.com/paper/2207.06080/) (2022)
- [Exemplar-Free Continual Transformer with Convolutions](https://mlsanity.com/paper/2308.11357/) (2023)
- [Ferret: An Efficient Online Continual Learning Framework under Varying Memory Constraints](https://mlsanity.com/paper/2503.12053/) (2025)

## Abstract (arXiv)

Exemplar-free class-incremental learning (EFCIL) aims to acquire new classes over time without storing raw data. Historically, prototype rehearsal, which samples around stored class prototypes and mixes them with current-task data, has been a popular strategy to reduce catastrophic forgetting. However, recent drift-compensation methods that explicitly realign prototypes in the evolving feature space consistently outperform prototype-based rehearsal, raising the question of whether rehearsal itself is fundamentally limited. We argue that the performance gap stems not from the idea of prototype rehearsal per se, but from how it is typically instantiated: existing approaches treat prototypes as isolated class summaries that ignore information from nearby enemy classes, and fail to correct the emerging class imbalance between a handful of synthetic old-class samples and hundreds of real instances from newly introduced classes. Building on this hypothesis, we revisit prototype rehearsal and propose a manifold-aware variant that restores its competitiveness in EFCIL. First, we introduce Constrained Expansive Over-Sampling, which interpolates each old-class prototype toward its nearest enemy features from new classes, generating boundary-aware rehearsal samples that better follow the underlying data manifold while preserving inter-class separation. Second, we design an Adaptive Class-Balanced loss that performs time-based class weighting, amplifying gradients from older prototypes when they are most informative and gradually annealing their influence as richer supervision from later tasks accumulates. Together, these components turn prototype rehearsal into a drift-resilient, imbalance-aware mechanism that closes, and often reverses, the gap to recent drift-compensation methods, achieving state-of-the-art performance across multiple EFCIL benchmarks.