PATH: A New Autoregressive Method for Interval Prediction on Tabular Data
A new research paper on arXiv (ID: 2608.08078) introduces PATH, a method for interval prediction on tabular data. The approach frames interval construction as next-interval prediction, using an autoregressive decoder to refine branch probabilities in a binary tree hierarchy. This method aims to produce shorter intervals while maintaining target coverage levels, addressing limitations of post hoc calibration rules in conformal regression. The paper is authored by researchers and is available at https://arxiv.org/abs/2608.08078.
Key facts
- Paper ID: arXiv:2608.08078
- Title: PATH: Next-Interval Prediction via Autoregressive Tree Hierarchy on Tabular Data
- Method: PATH learns probability mass flow from intervals to nested subintervals
- Uses autoregressive decoder to refine branch probabilities
- Targets interval prediction with coverage calibration
- Addresses limitations of post hoc rules in conformal regression
- Available on arXiv
Entities
Institutions
- arXiv