理解与缓解大规模推荐系统训练不稳定：级联视角
原标题：Understanding and Mitigating Training Instability in Large-Scale Recommender Systems: A Cascade Perspective
AI 摘要
该论文针对大规模短视频推荐系统在动态工业环境中的训练不稳定问题，提出并验证了「G-W-L 级联」框架：训练发散通常从梯度爆炸开始，传播为权重爆炸，最终导致损失爆炸。作者将缓解手段按级联阶段映射为梯度控制、权重/激活稳定和损失塑形三类，并在某大型工业短视频推荐系统上进行了实证验证，表明该结构化方法能提升训练稳定性与最终性能。
正文节选
Understanding and Mitigating Training Instability in Large-Scale Recommender Systems: A Cascade Perspective Abstract. Training large-scale recommender systems on dynamic industrial platforms, such as Short-Form Video (SFV) services with billions of daily interactions, faces critical stability challenges. Rapid distribution shifts in user data and increasing model complexity frequently trigger a divergence cascade, typically initiating with gradient explosion, propagating to weight explosion, and