• Yuanlu Bai | Machine Learning Engineer II, L1 Conversion and Shopping Modeling; Yao Cheng | Sr. • Machine Learning Engineer, L1 Conversion and Shopping Modeling; Xiao Yang | Sr. • Staff Machine Learning Engineer, Ads Lightweight Ranking; Zhaohong Han | Manager II, Ads Lightweight Ranking; Jinfeng Zhuang | Sr. • Manager, Ads Ranking Introduction Lightweight ranking plays a crucial role as an intermediate stage in Pinterest’s ads recommendation system. • Its main purpose is to efficiently narrow down the set of candidate ads before passing them to downstream, more complex ranking models. • By doing so, it ensures that only the most relevant candidates move forward, improving both the efficiency and quality of our ads recommendations.
Article Summaries:
- Pinterest announced the launch of its first GPU‑served two‑tower model for lightweight ads engagement prediction. The new architecture combines Multi‑gate Mixture‑of‑Experts (MMOE) with Deep & Cross Networks (DCN), replacing the earlier Multi‑Task Multi‑Domain design. By computing Pin embeddings offline and real‑time query embeddings on GPU, the system keeps latency similar to the CPU baseline while supporting a more complex model. Offline loss dropped 5-10 % versus the previous production model, and further segmentation of standard and shopping ads reduced loss another 5-10 %, doubling iteration speed. Training efficiency was improved through GPU prefetching and optimized data loading.
Sources: