LAMP: Layerwise Adaptive Mixing of Predictors for Training-Free Few-Shot Vision-Language Adaptation

Name: LAMP: Layerwise Adaptive Mixing of Predictors for Training-Free Few-Shot Vision-Language Adaptation
Uploaded: 2026-10-20T12:11:19.865490+00:00
Duration: 261 s
Description: LAMP is a training-free method that improves few-shot vision-language adaptation by combining global and regional information from multiple encoder layers. It uses independently fitted kernel predictors to fit label residuals, then selects the best mixture using labeled validation data. This approach leverages complementary features from frozen encoders to achieve state-of-the-art accuracy.

LAMP is a training-free method that improves few-shot vision-language adaptation by combining global and regional information from multiple encoder layers. It uses independently fitted kernel predictors to fit label residuals, then selects the best mixture using labeled validation data. This approach leverages complementary features from frozen encoders to achieve state-of-the-art accuracy.

Open the full episode