LAMP: Layerwise Adaptive Mixing of Predictors for Training-Free Few-Shot Vision-Language Adaptation
We propose LAMP, Layerwise Adaptive Mixing of Predictors, a training-free framework that combines global and regional information from multiple encoder layers.
ProofPaper ↗
Key points
- Few-shot adaptation often relies on an encoder's final global representation, leaving complementary information in intermediate layers underused.
- Global and support-calibrated regional kernel predictors independently fit the same label residuals.
- LAMP achieves state-of-the-art mean accuracy among the compared published training-free ResNet-50 methods at 4, 8, and 16 shots, reaching 78.73% at 16 shots.
- These findings show that combining layer-specific predictions can exploit complementary features from frozen encoders to improve few-shot classification.
Sources (1)
- [1]LAMP: Layerwise Adaptive Mixing of Predictors for Training-Free Few-Shot Vision-Language AdaptationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 04:35 PM
We propose LAMP, Layerwise Adaptive Mixing of Predictors, a training-free framework that combines global and regional information from multiple encoder layers.
Few-shot adaptation often relies on an encoder's final global representation, leaving complementary information in intermediate layers underused.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026Data-Efficient ConvNeXt Pretraining for High-Resolution Sentinel-1 GRD SAR Representation Learning
- Oct 15, 2026VesperMUSE: Multi-Modal Fusion Vision Transformer for Image Semantic Segmentation Under Diverse Light Conditions
- Oct 15, 2026Spurious Attention Heads in Vision Transformers: Mechanistic Identification of Annotation-Driven Bias in Medical Chest X-Ray Analysis
- Oct 14, 2026Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation
- Oct 14, 2026Continual pretraining Prostate MRI ViT Encoder for task alignment
- Oct 14, 2026Your Transformer Is Secretly Elastic