LoopCTR: Unlocking the Loop Scaling Power for Click-Through Rate Prediction

TL;DR

LoopCTR introduces a recursive loop scaling paradigm with shared layers and process supervision, enabling efficient CTR prediction models that outperform baselines with fewer parameters and loops.

Contribution

It proposes a novel loop scaling method with shared layers and supervision, decoupling computation from parameter growth for improved CTR models.

Findings

Achieves state-of-the-art performance on multiple benchmarks.

Single forward pass with no loops outperforms all baselines.

Models trained with fewer loops have higher oracle AUC ceilings.

Abstract

Scaling Transformer-based click-through rate (CTR) models by stacking more parameters brings growing computational and storage overhead, creating a widening gap between scaling ambitions and the stringent industrial deployment constraints. We propose LoopCTR, which introduces a loop scaling paradigm that increases training-time computation through recursive reuse of shared model layers, decoupling computation from parameter growth. LoopCTR adopts a sandwich architecture enhanced with Hyper-Connected Residuals and Mixture-of-Experts, and employs process supervision at every loop depth to encode multi-loop benefits into the shared parameters. This enables a train-multi-loop, infer-zero-loop strategy where a single forward pass without any loop already outperforms all baselines. Experiments on three public benchmarks and one industrial dataset demonstrate state-of-the-art performance. Oracle analysis further reveals 0.02--0.04 AUC of untapped headroom, with models trained with fewer loops exhibiting higher oracle ceilings, pointing to a promising frontier for adaptive inference.