Generality of the convolutional-network drift ordering in recurrent systems

Determine whether the ordering of representational drift profiles observed across continual-learning methods in convolutional networks generalizes to recurrent neural networks.

Background

The convolutional-network experiments compared naive sequential training, elastic weight consolidation, Learning without Forgetting, and experience replay, whereas the recurrent-network analysis reliably compared only naive training and replay. Elastic weight consolidation and Learning without Forgetting did not provide sufficiently reliable continual-learning performance in the recurrent setting for an interpretable drift comparison.

Consequently, the relative ordering of methods by retention, plasticity, and representational stability established in the convolutional experiments has not been shown to hold for recurrent architectures. Establishing this generalization would require reliable, interpretable comparisons among the corresponding continual-learning methods in recurrent systems.

References

EWC and LwF did not achieve sufficiently reliable continual-learning performance in the recurrent network for an interpretable drift comparison; whether the convolutional-network ordering generalizes to recurrent systems remains unknown.

— Continual-learning rules shape representational drift  (2608.16141 - Si et al., 17 Aug 2026) in Discussion, Limitations

Whether M3's advantage in ageing rate reflects recurrent architectures in general, or is specific to this particular accuracy gap, is accordingly still an open question.

— Template Ageing and Longitudinal Verification in Fixed-Text Keystroke Dynamics: A Subject-Disjoint Study Across Eight Weeks  (2609.29851 - Parkinson et al., 24 Sep 2026) in Section 5.2, “Limitations,” subsection “Mechanism comparability”