analiza aiqbench
This paper proposes Sparse Autoencoder Geometric Anchoring (SAGA) as a novel method to mitigate catastrophic forgetting in continual learning by leveraging sparse autoencoders to constrain representation drift. The approach establishes a local geometric relationship between tasks, allowing models to retain previously learned information while adapting to new data without retaining the full SAE architecture. Experimental results on standard vision benchmarks demonstrate significant accuracy improvements and reduced forgetting compared to existing baselines, highlighting the utility of low-rank geometry in balancing stability and plasticity.
Sparse Autoencoders Can Induce Local Geometry for Mitigating Catastrophic Forgetting
arXiv:2610.22439v1 Announce Type: new Abstract: Catastrophic forgetting remains a challenge in supervised continual learning, as adapting to new tasks causes representation drift and loss of previously learned information. This raises the question of which representational changes should be constrained while preserving plasticity. We therefore introduce Sparse Autoencoder Geometric Anchoring (SAGA), which converts a Sparse Autoencoder (SAE) trained at each task boundary into local, low-rank constraints on representation drift. Replay examples specify where the constraints apply, while leading right singular directions of the encoder Jacobian specify which changes are penalized. The resulting geometry is input dependent and can be retained without keeping the SAE for subsequent tasks' training. We establish its local relation to latent code matching, define its spectral truncation error, and derive a conditional bound on frozen-readout drift. With replay and snapshot history, the method achieves the highest mean Class-IL accuracy and lowest mean forgetting among the evaluated ER, DER++, ER-ACE, and SGD baselines on three pretrained ViT-B/16 benchmarks, with accuracy gains of 12.15, 4.79, and 3.22 percentage points over the strongest conventional baseline on CIFAR-100, CUB-200, and ImageNet-R, respectively. From scratch results are more mixed, while rank ablations and geometry controls show that SAE-induced low-rank geometry offers useful stability-plasticity trade-offs.
Abstract:Catastrophic forgetting remains a challenge in supervised continual learning, as adapting to new tasks causes representation drift and loss of previously learned information. This raises the question of which representational changes should be constrained while preserving plasticity. We therefore introduce Sparse Autoencoder Geometric Anchoring (SAGA), which converts a Sparse Autoencoder (SAE) trained at each task boundary into local, low-rank constraints on representation drift. Replay examples specify where the constraints apply, while leading right singular directions of the encoder Jacobian specify which changes are penalized. The resulting geometry is input dependent and can be retained without keeping the SAE for subsequent tasks' training. We establish its local relation to latent code matching, define its spectral truncation error, and derive a conditional bound on frozen-readout drift. With replay and snapshot history, the method achieves the highest mean Class-IL accuracy and lowest mean forgetting among the evaluated ER, DER++, ER-ACE, and SGD baselines on three pretrained ViT-B/16 benchmarks, with accuracy gains of 12.15, 4.79, and 3.22 percentage points over the strongest conventional baseline on CIFAR-100, CUB-200, and ImageNet-R, respectively. From scratch results are more mixed, while rank ablations and geometry controls show that SAE-induced low-rank geometry offers useful stability-plasticity trade-offs.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2610.22439 [cs.LG]
(or arXiv:2610.22439v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.22439 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 292 cuvinte
Subiecte conexe
Reflections on unlearning and inoculation
LessWrong · 20 sept. 2026
NIAR Researchers Show QFI Improves Quantum Continual Learning
Quantum Zeitgeist · 20 aug. 2026
PIRAMID: Progress and Plans
LessWrong · 27 iul. 2026
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · 12 iun. 2026