Latent spectral regularization for continual learning

Emanuele Frascaroli, Riccardo Benaglia, Matteo Boschini, Luca Moschella, Cosimo Fiorini, Emanuele Rodolà, Simone Calderara

Abstract: Highlights•We study the geometry of a model’s latent space in a Continual Learning setting.•We propose Continual Spectral Regularizer, a geometrically motivated regularizer.•We combine CaSpeR with SOTA rehearsal-based CL approaches in standard benchmarks.•We compare our proposal with recent contrastive-based CL approaches.•We reveal that CaSpeR achieves increased accuracy and reduced forgetting.