AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Decomposing Earth Embeddings with Sparse Autoencoders

Vitus Benson, Fanny Yang, Markus Reichstein

NeurIPS 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Applies sparse autoencoders to decompose Earth observation foundation model embeddings into more interpretable sparse features.

Problem

Earth observation foundation model embeddings are opaque because each dimension entangles unrelated physical concepts, hurting interpretability and robustness for practitioners and scientists.

Approach

Proposes using sparse autoencoders as a principled way to decompose dense Earth embeddings into sparse, disentangled concept-like features.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use trained SAE instead of PCA at matched sparsityEarth embeddings · AlphaEarth reconstruction L0=16fraction of variance explained: 0.70 -> 0.78 (frozen decoder)Improved
train the SAE decoder instead of freezing a random dictionary (ablation)BatchTopK SAE · AlphaEarth reconstruction L0=16unexplained variance (%): 22 -> 2.7Improved
train the SAE decoder instead of freezing random dictionary (ablation)BatchTopK SAE · Tessera reconstruction L0=16unexplained variance (%): 47 -> 8.4Improved
use SAE features instead of raw dimensionsEarth embeddings · AlphaEarth concept alignmentmedian groups aligned: 12 -> 88% of features align with at most one groupImproved
train the decoder instead of freezing a random dictionary (ablation)SAE · AlphaEarth concept alignmentshare aligned with at most one group: 88% vs 89%No clear effect
use SAE features instead of raw dimensionsEarth embeddings · AlphaEarth spatial scalewithin-patch variance share: 9% -> 23-32%Improved
train decoder instead of freezing random dictionary (ablation)SAE · AlphaEarth spatial scalemedian correlation length: 23 km (frozen) vs 20-33 kmNo clear effect
use SAE features for concept detection instead of raw dimensionsAlphaEarth · concept detectionmean folded AUROC: 0.84 -> 0.75Hurt
compute AoA distance in SAE code instead of raw embeddingridge regressor on canopy height · OOD error rankingSpearman rho: +0.17 (AlphaEarth), +0.21 (Tessera) median gainImproved
compute AoA distance in SAE code instead of raw embeddingridge regressor on elevation · OOD error rankingSpearman rhoNo clear effect
use trained SAE instead of untrained random sparse code for OOD ranking (ablation)AoA distance · canopy heightSpearman rho gain: random code gains as much on AlphaEarth; +0.07 on TesseraNo clear effect
increase dictionary size M from 512 to 2048 (ablation)SAE at L0=16 · reconstructionFVE: <0.01 changeNo clear effect

Experimental setup

Builds on: sparse autoencoders

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy