TensorFlow Training on the JVM | Christoph Henkelmann

Christoph Henkelmann

Explore the viability of TensorFlow training on the JVM, its compatibility with Java and Kotlin, and the benefits and drawbacks of this option for deep learning.

Key takeaways
  • TensorFlow Training on JVM is a viable option for deep learning on Java.
  • Maarten, Google’s JVM-based tensorflow API won’t be deprecated or removed.
  • In Maarten, the JVM-based tensorflow API is designed to communicate with the tensor API in Java.
  • Using Java and Kotlin for deep learning is a good choice for commercial applications.
  • TensorFlow is not smarter or faster just because it is Google.
  • The tf data pipeline doesn’t do anything essentially different from pre-processing in Java.
  • The tensorflow persistence RP is really ugly and not easy to use.
  • Java is fast and can be used for deep learning.
  • The JVM-based tensorflow API is a good choice for those who want to use tensorflow with Java.
  • The tensorflow API is not stable and can change often.
  • The tensorflow graph is a file extension.
  • The use of Kotlin for deep learning is recommended.
  • TensorFlow is not a beginner’s tool for machine learning.

More talks

Talks - Krishi Sharma: Trust Fall: Three Hidden Gems in MLFlow Krishi Sharma Java on CRaC: Superfast JVM Application Startup by Simon Ritter Simon Ritter Quarkus Unveiled: Efficiency & Green Impact • Holly Cummins & Charles Humble Holly Cummins Honey Bee Conservation using Deep Learning Dr. Sahand Hormozi Zero Waste, Radical Magic, and Italian Graft – Quarkus Efficiency Secrets by Holly Cummins Holly Cummins Project Leyden: Improving Java’s Startup Time by Per Minborg, Sébastien Deleuze Per Minborg, Sébastien Deleuze FP & Scala: The Killer Combo for Machine Learning • Marek Kolodziej • YOW! 2015 Marek Kolodziej Machine Learning in the Browser | Athan Reines | ML Conference 2018 Athan Reines