Deep Learning in Production Seen by a Data Engineer
summary
This talk explains how to industrialize a chosen deep learning model for production: parallelizing prediction workloads, options for deploying without Python, strategies for large-scale model retraining, and monitoring model performance. It presents concrete solutions and lessons learned by a data engineer (Romain Sagean) from Xebia.
AI-written summary · about
chapters (15)
- 0:00Deep learning in production
- 1:10Sponsored links prediction
- 2:56Deep learning frameworks
- 4:46TensorFlow selection
- 5:27Early production failures
- 6:29Deep learning basics
- 7:09Preprocessing and logging
- 8:13Scaling predictions with Spark
- 9:18Broadcast variables
- 10:01Batch predictions
- 10:41Memory issues with DL4J
- 12:29Partition balancing
- 13:13Code optimization
- 13:55Smaller jar for deployment
- 14:38Targeted compilation
related talks
33mOvercoming the Barriers to Deep Learning in Production
42mDeep Learning with the GPUs in Production
42mDeep learning beyond the learning
35mGetting Machine Learning Models Ready For Production Using Python
48mML and DL in Production: Differences and Similarities
06mMachine learning is about your data and deployment
31mBest Practices for Producing High-Quality Deep Learning Systems
47mMachine Learning in Production: An Intro to MLOps.