<

Putting a Face to Forgetting: Continual Learning meets Mechanistic Interpretability

Sergi Masip, Gido M. van de Ven, Javier Ferrando, Tinne Tuytelaars

CoLLAs • 2026

Paper

We introduce a mechanistic framework that gives a geometric interpretation of catastrophic forgetting as transformations to how individual features are encoded — a new, feature-centric vocabulary for continual learning.

Illustration of an experiment where a human dissects a robot brain. Circa 2026 (colorized)

Accepted to CoLLAs 2026 — camera-ready and full proceedings citation coming soon. Linked below is the arXiv preprint.

Catastrophic forgetting in continual learning is often measured at the performance or last-layer representation level, overlooking the underlying mechanisms. We introduce a mechanistic framework that offers a geometric interpretation of catastrophic forgetting as the result of transformations to the encoding of individual features. These transformations can lead to forgetting by reducing the allocated capacity of features (worse representation) and disrupting their readout by downstream computations.

Analysis of a tractable model formalizes this view, allowing us to identify best- and worst-case scenarios. Through experiments on this model, we empirically test our formal analysis and highlight the detrimental effect of depth.

Finally, we demonstrate how our framework can be used in the analysis of practical models through the use of Crosscoders. We present a case study of a Vision Transformer trained on sequential CIFAR-10. Our work provides a new, feature-centric vocabulary for continual learning.

Citation

BibTeX
@misc{masip2026puttingface,
  title = {Putting a Face to Forgetting: Continual Learning meets Mechanistic Interpretability},
  author = {Sergi Masip and Gido M. van de Ven and Javier Ferrando and Tinne Tuytelaars},
  year = {2026},
  eprint = {2601.22012},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url = {https://arxiv.org/abs/2601.22012}
}