GEAR aligns Ricci curvatures in latent spaces of models to create a unified transfer learning architecture, achieving 14.4% and 8.3% performance gains on 23 molecular task pairs under random and scaffold splits.
Franco Scarselli, Marco Gori, Ah Tsoi, Markus Hagenbuchner, and Gabriele Monfardini
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
representative citing papers
Adding register tokens to Vision Transformers eliminates high-norm background artifacts and raises state-of-the-art performance on dense visual prediction tasks.
VTAB is a 19-task benchmark that measures representation quality by few-shot adaptation performance across diverse vision domains, with a controlled large-scale comparison of popular pretraining methods.
Relative rankings of online EWC, LwF, SI and GEM are not consistently preserved across five trainable-depth regimes on five datasets and eleven task orders.
PEFT-MedSAM adapts MedSAM by training only its mask decoder on ISIC 2018 skin lesion data, achieving Dice 0.9411 and outperforming U-Net (0.8715) and zero-shot MedSAM (0.8997), with PH2 validation (0.9467) and 98.27% Grad-CAM pointing accuracy.
citing papers explorer
-
Geometric Embedding Alignment via Curvature Matching in Transfer Learning
GEAR aligns Ricci curvatures in latent spaces of models to create a unified transfer learning architecture, achieving 14.4% and 8.3% performance gains on 23 molecular task pairs under random and scaffold splits.
-
Vision Transformers Need Registers
Adding register tokens to Vision Transformers eliminates high-norm background artifacts and raises state-of-the-art performance on dense visual prediction tasks.
-
A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark
VTAB is a 19-task benchmark that measures representation quality by few-shot adaptation performance across diverse vision domains, with a controlled large-scale comparison of popular pretraining methods.
-
Fine-Tuning Regimes Define Distinct Continual Learning Problems
Relative rankings of online EWC, LwF, SI and GEM are not consistently preserved across five trainable-depth regimes on five datasets and eleven task orders.
-
PEFT-MedSAM: Efficient Fine-Tuning of Medical Foundation Models for Explainable Skin Lesion Segmentation
PEFT-MedSAM adapts MedSAM by training only its mask decoder on ISIC 2018 skin lesion data, achieving Dice 0.9411 and outperforming U-Net (0.8715) and zero-shot MedSAM (0.8997), with PH2 validation (0.9467) and 98.27% Grad-CAM pointing accuracy.