Infrastructure Storing and retrieving millions of ad impressions per second

Learn how we design a system that stores and retrieves ad impression metadata accurately and consistently in Twitter's ads ecosystem.

Open source Introducing Twitter Text Editor

Introducing Twitter Text Editor, a standalone, flexible API that provides a full-featured rich text editor for iOS applications.

Insights Deep learning on dynamic graphs

A new neural network architecture for dynamic graphs

Infrastructure Kafka as a storage system

Kafka is traditionally used to power streaming infrastructures. Learn how we used Kafka as a storage system to build the Account Activity API Replay Feature.

Infrastructure Introducing VMAF percentiles for video quality measurements

Twitter is introducing a visual quality assessment method that relies on computing VMAF percentiles. Compared to existing techniques, the method is effective and intelligible to non-video engineers.

Infrastructure How we fortified Twitter's real time ad spend architecture

Learn how our Revenue Platform team fortified Twitter's real time ad spend architecture to prevent overspend.

Insights What Twitter learned from the Recsys 2020 Challenge

In this blog post we describe the dataset that Twitter released for the RecSys 2020 Challenge and the insights we had from the winning teams.

Insights Distributed training of sparse ML models — Part 3: Observed speedups

Using our customized data and model parallel distributed training strategy provides training speed improvements of up to 60x over single-node training for sparse machine learning models at Twitter.

Insights Distributed training of sparse ML models — Part 2: Optimized strategies

We use a combination of data parallelism and model parallelism in a customized distributed training strategy to enable fast training of large sparse machine learning models at Twitter.

Insights Distributed training of sparse ML models — Part 1: Network bottlenecks

We detail the optimizations behind our custom approach to distributed training. We begin with the distribution strategies provided by TensorFlow, and the difficulties we had using them at Twitter.

See more