An Incremental Learning framework for Large-scale CTR Prediction
This work addresses the need for efficient and adaptive recommendation systems in massive-scale services like Taboola, though it is incremental as it builds on existing incremental learning and distillation techniques.
The paper tackles the problem of capturing emerging trends in large-scale CTR prediction by introducing an incremental learning framework that uses warm-starting and a teacher-student paradigm to mitigate catastrophic forgetting, resulting in a 12x speedup in training and deployment cycles and consistent RPM and CTR improvements.
In this work we introduce an incremental learning framework for Click-Through-Rate (CTR) prediction and demonstrate its effectiveness for Taboola's massive-scale recommendation service. Our approach enables rapid capture of emerging trends through warm-starting from previously deployed models and fine tuning on "fresh" data only. Past knowledge is maintained via a teacher-student paradigm, where the teacher acts as a distillation technique, mitigating the catastrophic forgetting phenomenon. Our incremental learning framework enables significantly faster training and deployment cycles (x12 speedup). We demonstrate a consistent Revenue Per Mille (RPM) lift over multiple traffic segments and a significant CTR increase on newly introduced items.