State-Space Unlearning for Non-Stationary Bias in Land Surface Forecasting

Anidipta Saha, Author 2, Author 3, Author 4, Author 5

machine unlearninggeosciencestate space modelsforecasting

2610.02248

Builder Relevance

80%

1h ago

Abstract

This paper introduces SSU-LSF, a machine-unlearning framework for geoscientific Mamba-based SSMs that addresses non-stationary confounding events in land surface forecasting.

Reality Card

Core Claim

SSU-LSF achieves confounding reduction rates of up to 0.859 while maintaining a worst-case clean-domain RMSE degradation of only 4.2% on ERA5.

Method / Result

Achieves confounding reduction rates of 0.859 on ERA5 with 8.4x lower GPU-cost per unlearning request than full retraining.

Limitations

The paper does not provide extensive details on the implementation of the EKFac influence functions, which may affect reproducibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.

← Back to all papers