ST-LoRA: Single Trajectory LoRA Ensemble for Uncertainty Aware Agricultural Segmentation

Name: ST-LoRA: Single Trajectory LoRA Ensemble for Uncertainty Aware Agricultural Segmentation
Uploaded: 2026-10-09T05:06:51.146009+00:00
Duration: 408 s
Description: ST-LoRA is a method combining Low-Rank Adaptation (LoRA) with snapshot ensembling to create an efficient ensemble for agricultural segmentation. It uses a single training run and different random initializations to generate multiple models, achieving performance comparable to full ensembles while significantly reducing training costs and improving robustness against distribution shifts.

ST-LoRA is a method combining Low-Rank Adaptation (LoRA) with snapshot ensembling to create an efficient ensemble for agricultural segmentation. It uses a single training run and different random initializations to generate multiple models, achieving performance comparable to full ensembles while significantly reducing training costs and improving robustness against distribution shifts.

Open the full episode