Learning Transferable Predictability Representations
Researchers have developed a new model called the Gauge-Fixed Ordinal Network (GON) to score trajectory windows based on their predictability. This model aims to provide a consistent numerical interpretation of predictability across different systems, unlike existing methods that are limited to single systems. The GON uses a temporal convolutional model and an anchor-and-variance objective to achieve this, operating on local trajectory geometry features. Experiments show that initializing GON with a pretrained checkpoint significantly improves performance across various window sizes and systems, demonstrating its cross-system transferability. AI
IMPACT Introduces a novel method for assessing and transferring predictability scores across diverse dynamical systems, potentially improving forecasting and diagnostics.