Uncertainty Modeling for Multi-Objective RTA Interception with Distillation Acceleration

TL;DR

This paper introduces UMDA, a joint modeling framework with distillation that efficiently estimates traffic quality and uncertainties in real-time RTA interception, improving accuracy and speed.

Contribution

It presents a novel theoretical analysis and a combined multi-objective and uncertainty modeling approach, enhanced by knowledge distillation for real-time efficiency.

Findings

UMDA provides more effective samples for downstream tasks.

Distilled UMDA retains uncertainty-sharing capabilities.

Achieves a tenfold increase in inference speed.

Abstract

Real-Time Auction (RTA) Interception aims to filter out invalid or irrelevant traffic to enhance the integrity and reliability of downstream data. However, two key challenges remain: (i) the need for accurate estimation of traffic quality together with sufficiently high confidence in the model's predictions, typically addressed through uncertainty modeling, and (ii) the efficiency bottlenecks that such uncertainty modeling introduces in real-time applications due to repeated inference. To address these challenges, we first provide a theoretical analysis of the intrinsic mechanism underlying uncertainty estimation. Building on this analysis, we propose a joint modeling framework that integrates multi-objective learning with uncertainty modeling, named UMDA, which yields both traffic quality predictions and reliable confidence estimates. We further apply knowledge distillation to UMDA, enabling the model to produce both aleatoric and epistemic uncertainties in a single forward pass, thereby substantially reducing the computational overhead of uncertainty modeling, while largely preserving predictive accuracy and retaining the benefits of multiple-forward-pass uncertainty estimation. Experiments on the JD and Criteo datasets demonstrate that UMDA provides more effective samples for downstream tasks through uncertainty sharing, and the distilled model retains the original uncertainty-sharing capability while delivering a tenfold increase in inference speed.