Selective Test-Time Compute Scaling for Click-Through Rate Prediction via Uncertainty-Triggered Feature Path Exploration

2026-05-24

AI Summary

This work addresses the challenge that click-through rate (CTR) prediction models often yield unreliable estimates when encountering sparse feature combinations and struggle to dynamically handle instance-level uncertainty during inference. To this end, the authors propose the UTTSI framework, which introduces test-time computation scaling into CTR prediction for the first time. UTTSI employs a dual-signal mechanism—combining model confidence and data frequency priors—to disentangle epistemic from aleatoric uncertainty, enabling adaptive decisions: high-uncertainty instances trigger stochastic feature-path exploration and consistency-based ensemble weighting to enhance predictions, while confident instances bypass additional computation. Notably, UTTSI requires no retraining and is model-agnostic. Evaluated across four datasets and three backbone models, it consistently outperforms training-phase baselines, achieving a statistically significant 5.3% relative CTR gain (p < 0.01) in online A/B tests with only a 2.8× average computational overhead.

Abstract

Scaling test-time compute has proven highly effective for language models, yet this opportunity remains largely unexplored for industrial Click-Through Rate (CTR) prediction. CTR models suffer from a fundamental asymmetry: feature combinations well-represented in training yield confident predictions, while sparsely observed ones produce unreliable outputs. Existing training-phase solutions such as adaptive gating learn a fixed selection function subject to the same sparsity, offering no per-instance recourse at deployment.We propose UTTSI (Uncertainty-Triggered Test-Time Selective Inference), a training-free model-agnostic framework that scales inference depth proportionally to per-instance uncertainty. A dual-signal estimator combining model logit confidence with a data-level frequency prior distinguishes epistemic uncertainty from aleatoric ambiguity. Every instance undergoes adaptive feature filtering to remove unreliable embeddings; uncertain instances additionally receive stochastic feature-path explorations whose predictions are aggregated via consistency-weighted ensembling. Confident instances bypass exploration entirely, keeping average overhead at approximately $2.8\times$ base model cost with worst-case latency unchanged.Experiments on four datasets with three backbone architectures demonstrate consistent, statistically significant gains over all training-phase baselines. A seven-day online A/B test further confirms a 5.3% relative CTR gain ($p < 0.01$), establishing selective test-time compute allocation as a practical complement to training-phase advances for CTR prediction.

Problem

• Click-Through Rate Prediction
• Test-Time Compute Scaling
• Uncertainty
• Feature Sparsity
• Model Reliability

Innovation

• test-time compute scaling
• uncertainty quantification
• feature-path exploration
• CTR prediction
• adaptive inference