Research map / Architectures and efficiency / Pruning and efficient tuning
SIMPLER: Efficient Foundation Model Adaptation via Similarity-Guided Layer Pruning for Earth Observation
Víctor Barreiro, Johannes Jakubik, Francisco Argüello, Dora Heras
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CKA instead of Jaccard/SVCCA for layer selection (ablation) | SIMPLER · MADOS | mIoU: 50.7 -> 62.8 | Improved |
| add ImageNet-style pre-training (Prithvi) versus from scratch (ablation) | SIMPLER 5-block · MADOS | mIoU: 44.1 -> 62.8 | Improved |
| add pre-training to full baseline (ablation) | Prithvi-EO-2 300M · MADOS | mIoU: 46.7 -> 66.9 | Improved |
| prune 5-block depth instead of full 24 blocks, from scratch (ablation) | Prithvi-EO-2 300M · MADOS (from scratch) | mIoU: 44.1 vs 46.7 | No clear effect |
| select depth with SIMPLER instead of 40% magnitude pruning | Prithvi-EO-2 300M · MADOS | mIoU: 47.9 -> 62.8 | Improved |
| increase CKA samples from 10 to 500 (ablation) | SIMPLER · MADOS | cutoff std (blocks): 2.95 -> 0.53 | Improved |
| increase CKA samples from 500 to 1000 (ablation) | SIMPLER · MADOS | computation time: 218s -> 678s | Hurt |
| apply SIMPLER to TerraMind-Small | TerraMind-Small · MADOS (TerraMind) | mIoU: 53.0 -> 53.6 | Improved |
| apply SIMPLER to TerraMind-Large | TerraMind-Large · MADOS (TerraMind) | mIoU: 70.1 -> 58.8 | Hurt |
| apply SIMPLER to ViT-MAE on CIFAR-100 | ViT-MAE · CIFAR-100 | accuracy: 88.8 -> 72.8 | Hurt |
| apply SIMPLER to Prithvi-EO-2 300M | Prithvi-EO-2 300M · BigEarthNetv2 | mAP: 73.4 -> 71.2 | Hurt |
| apply SIMPLER to Prithvi-EO-2 300M | Prithvi-EO-2 300M · Sen4Map | F1-macro: 66.6 -> 63.8 | Hurt |
Closest papers by meaning
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.89)
- REOBench: Benchmarking Robustness of Earth Observation Foundation Models (NeurIPS 2025 · similarity 0.88)
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.88)
- EoS-FM: Can an Ensemble of Specialist Models act as a Generalist Feature Extractor? (ECCV 2026 · similarity 0.88)
- ExPLoRA: Parameter-Efficient Extended Pre-Training to Adapt Vision Transformers under Domain Shifts (ICML 2025 · similarity 0.86)
- Meta-Learning Hyperparameters for Parameter Efficient Fine-Tuning (CVPR 2025 · similarity 0.86)
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.
