Zono-Conformal Prediction: Zonotope-Based Uncertainty Quantification for Regression and Classification Tasks

Researchers introduce zono-conformal prediction, a new uncertainty quantification method for machine learning that uses zonotope-based prediction sets instead of traditional intervals. The approach is more computationally efficient and less conservative than existing conformal prediction methods while maintaining statistical coverage guarantees for both regression and classification tasks.

• →Zono-conformal prediction addresses computational expense and data-intensity limitations of current conformal prediction methods.
• →The method uses zonotopic uncertainty sets that better capture dependencies in multi-dimensional outputs compared to interval-based approaches.
• →Zono-conformal predictors can be identified through a single, data-efficient linear program.
• →The approach works for both regression and classification tasks with probabilistic coverage guarantees.
• →Experimental results show the method is less conservative than standard conformal prediction while achieving similar test data coverage.