Scaling TensorFlow at LinkedIn with Jonathan Hung - #314
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Scaling TensorFlow at LinkedIn
This chapter explores the challenges LinkedIn faces in scaling TensorFlow training, particularly concerning fault tolerance and GPU utilization. It discusses the implementation of a Kubernetes research cluster to enhance experimentation and integration with existing frameworks while highlighting the architectural differences from Hadoop. Additionally, the chapter emphasizes efforts to refine infrastructure and improve user experience for engineers working with machine learning applications.
Transcript
Play full episode
