Learning from Synthetic Data via Provenance-Based Input Gradient Guidance

Researchers propose a new machine learning framework that uses provenance information from synthetic data generation to improve model training. The method uses input gradient guidance to suppress learning from non-target regions, reducing spurious correlations and improving discrimination accuracy across multiple AI tasks.

• →New framework addresses spurious correlations in synthetic data training by using provenance information to guide learning.
• →Input gradient guidance suppresses model reliance on non-target regions during training.
• →Method shows effectiveness across weakly supervised object localization, spatio-temporal action localization, and image classification.
• →Approach directly promotes learning of discriminative representations for target regions rather than indirect improvements.
• →Framework helps reduce synthesis biases and artifacts that can mislead model training.