Sketched Lanczos Uncertainty Score: A Low-Memory Summary of the Fisher Information

Abstract

Current uncertainty quantification is memory and compute expensive, which hinders practical uptake. To counter, we develop Sketched Lanczos Uncertainty (SLU): an architecture-agnostic uncertainty score that can be applied to pre-trained neural networks with minimal overhead. Importantly, the memory use of SLU only grows logarithmically with the number of model parameters. We combine Lanczos' algorithm with dimensionality reduction techniques to compute a sketch of the leading eigenvectors of a matrix. Applying this novel algorithm to the Fisher information matrix yields a cheap and reliable uncertainty score. Empirically, SLU yields well-calibrated uncertainties, reliably detects out-of-distribution examples, and consistently outperforms existing methods in the low-memory regime.

PDF NeurIPS OpenReview Semantic Scholar

Cite

Text

Miani et al. "Sketched Lanczos Uncertainty Score: A Low-Memory Summary of the Fisher Information." Neural Information Processing Systems, 2024. doi:10.52202/079017-0728

Markdown

[Miani et al. "Sketched Lanczos Uncertainty Score: A Low-Memory Summary of the Fisher Information." Neural Information Processing Systems, 2024.](https://mlanthology.org/neurips/2024/miani2024neurips-sketched/) doi:10.52202/079017-0728

BibTeX

@inproceedings{miani2024neurips-sketched,
  title     = {{Sketched Lanczos Uncertainty Score: A Low-Memory Summary of the Fisher Information}},
  author    = {Miani, Marco and Beretta, Lorenzo and Hauberg, Søren},
  booktitle = {Neural Information Processing Systems},
  year      = {2024},
  doi       = {10.52202/079017-0728},
  url       = {https://mlanthology.org/neurips/2024/miani2024neurips-sketched/}
}