Pinterest Engineering Blog

PinterestMay 1, 2026

Optimizing ML Workload Network Efficiency (Part I): Feature Trimmer

In Pinterest's online ML serving systems, a root-leaf architecture was optimized to reduce network bandwidth usage. Initially, excessive feature transmission from the root to the leaf caused a network bottleneck, requiring system scaling based on network usage. To address this, root-leaf network bandwidth usage was reduced by 20% with lz4 compression, though it also increased CPU usage and latency. However, this did not solve the underlying problem of shipping unused data. Instead, the "Send What You Use" approach was developed, which trims unnecessary features before transmission, potentially cutting root-leaf network usage by ~50%. This approach leverages model signatures to determine required features, ensuring only necessary data is transmitted between the root and leaf.

Machine LearningData

18 min

Latest Articles

Making User-Sequence Data More Cost-Efficient, Faster, and Easier to Use

An Engineer’s Guide to Better AI Skills: Implementing a Testing Process to Optimize Agent…

Enhancing Ad Relevance: Integrating Real-Time Context into Sequential Recommender Models

Optimizing ML Workload Network Efficiency (Part I): Feature Trimmer

From Clicks to Conversions: Architecting Shopping Conversion Candidate Generation at Pinterest

Smarter URL Normalization at Scale: How MIQPS Powers Content Deduplication at Pinterest

Finding zombies in our systems: A real-world story of CPU bottlenecks

Scaling Recommendation Systems with Request-Level Deduplication

Performance for Everyone

Evolution of Multi-Objective Optimization at Pinterest Home feed

Building an MCP Ecosystem at Pinterest

Unified Context-Intent Embeddings for Scalable Text-to-SQL

Unifying Ads Engagement Modeling Across Pinterest Surfaces

Bridging the Gap: Diagnosing Online–Offline Discrepancy in Pinterest’s L1 Conversion Models

Piqama: Pinterest Quota Management Ecosystem

Drastically Reducing Out-of-Memory Errors in Apache Spark at Pinterest

GPU-Serving Two-Tower Models for Lightweight Ads Engagement Prediction

Next Generation DB Ingestion at Pinterest

Beyond Two Towers: Re-architecting the Serving Stack for Next-Gen Ads Lightweight Ranking Models…

Ads Candidate Generation using Behavioral Sequence Modeling

Pinterest Engineering Blog

Latest Articles

Making User-Sequence Data More Cost-Efficient, Faster, and Easier to Use

An Engineer’s Guide to Better AI Skills: Implementing a Testing Process to Optimize Agent…

Enhancing Ad Relevance: Integrating Real-Time Context into Sequential Recommender Models

Optimizing ML Workload Network Efficiency (Part I): Feature Trimmer

From Clicks to Conversions: Architecting Shopping Conversion Candidate Generation at Pinterest

Smarter URL Normalization at Scale: How MIQPS Powers Content Deduplication at Pinterest

Finding zombies in our systems: A real-world story of CPU bottlenecks

Scaling Recommendation Systems with Request-Level Deduplication

Performance for Everyone

Evolution of Multi-Objective Optimization at Pinterest Home feed

Building an MCP Ecosystem at Pinterest

Unified Context-Intent Embeddings for Scalable Text-to-SQL

Unifying Ads Engagement Modeling Across Pinterest Surfaces

Bridging the Gap: Diagnosing Online–Offline Discrepancy in Pinterest’s L1 Conversion Models

Piqama: Pinterest Quota Management Ecosystem

Drastically Reducing Out-of-Memory Errors in Apache Spark at Pinterest

GPU-Serving Two-Tower Models for Lightweight Ads Engagement Prediction

Next Generation DB Ingestion at Pinterest

Beyond Two Towers: Re-architecting the Serving Stack for Next-Gen Ads Lightweight Ranking Models…

Ads Candidate Generation using Behavioral Sequence Modeling