What is project Panama and Why is it important for Machine Learning on Java Platform

Panama is the new API  for interconnecting Java with native code, which provides Java the ability for faster and easier communication with native libraries required for machine learning applications. The vision is that non-java libraries should be easily accessible to Java developers
Very good and easy to follow overview given at the session at Oracle Code One conference. Some of the main points:

Some problems with current JNI approach

  • No support for off heap data (only DYI solutions Unsafe, ByteBuffer)
  • No JIT optimizations
  • Requires writing and building native code
  • Maintaining JNI bindings can be big overhead

Why is Panama better

  • No native methods like in JNI. Panama uses interfaces with annotations
  • Rich API for off-heap data
  • Allows JIT compiler optimizations
  • Support for native arrays with Array<X>
  • Callbacks for communication between Java and native code
  • Jextract tool for automaticly generating interfaces from header files for specific native library (Tested with Tensorflow, Open CL)
  • Reduces 90% of the work required for creating Java bindings for native libraries
  • 4-5 time faster then JNI with link to native (experimental)
    faster then official Java bindings for Tensorflow


  • still needs work to be done on optimizing upcalls back to java (10x more expensive than JNI at the moment )

More: https://openjdk.java.net/projects/panama/

Why Java Will Dominate the Future of Machine Learning, AI, and Big Data

Great presentation by Bernard Traversat,  Head of the Java platform development, at Oracle Code One 2018, San Francisco. Here are key points:

  • Entire Big Data stack for storing data is written in Java (Hadoop, Spark, Kafka, Elastic Search, …)
  • Java has enormous ecosystem of tools and libraries
  • Enterprises run on Java
  • Bringing new technology into enterprise (such as AI, ML) is a huge challange
    because there is a lot to learn, and that gives Java and JVM huge advantage because enterprise allready knows Java and JVM. Technology needs to be understood by the enterprise.
  • Easy access and lower cost to develop, deploy and maintain:
    • easy access to developers
    • cost efficency (cost vs ROI)
    • maintainability as the code size is growing and aging (thanks to code readability)
  • Java is in the foundations of the cloud
  • Incoming Java innovations for ML, AI and BigData (openjdk.java.net):
    •  Panama, more efficent replacement for JNI (faster and easier communication with GPU).  Also the support for vector computing natively inside the JVM and Vector API
    • Valhalla, new type system for more efficient manipulation with data in memory
    • Loom, lightweight threads for massive parallel processing
    • ZGC, order of magnitude better performance for memory management
  • To conclude:
    • Java is #1 Programming Language
    • There are 12 Milion Java Developers
    • 38 Billion active JVM-s
    • 21 Billion cloud connected VM

The Most Significant JSR Award goes to Visual Recognition API, Code One, San Francisco

The most significant JSR Award, which is given by the official organization for Java technology standardization, Java Community Process (JCP) was awarded to Visual Recognition API, proposed and led by Zoran Sevarac (Deep Netts CEO), Java Champion Frank Greco (Crossroads Technologies), and Sandhya Kapoor (IBM Watson Architect). The proposal is official supported by IBM, and it is in First Early Draft Review phase. The award was announced during the Oracle Code One in San Francisco.
This award shows great interest and potentialfor development of visual recognition and machine learning technologies, and creating standards that would make it easier for Java developers to innovate and work with these technologies.

Dukes Choice Award 2018 goes to Apache Net Beans at Code One, San Francisco

Apache Net Beans project has won the most prestigious award for the innovationon Java platform at Oracle Code One Conference in San Francisco.
Zoran Sevarac, Deep Netts CEO, as a long time NetBeans contributor was part of the group who received the award. Also, this is of great importance for Deep Netts Platform, since our tools are based on API’s from Net Beans Platform. Also this shows the dynamic development and innovation capacity of oneof the main Java IDE’s used by more than 1 000 000 Java developers (according to official statistics).

See full video from the Dukes Choice Award at

Read the official announcement from Oracle:

Also check out the great interview with Geertjan Wielenga to get the idea whats going on behind the scene with Apache Net Beans, and the challanges durng the transition from Oracle to Apache:

Machine Learning For Software Developers in 45 minutes, Code One, San Francisco

Machine Learning for Software Developers in 45 minutes at Oracle Code One conference in San Francisco, by Deep Netts CEO Zoran Sevarac and another Java champion Frank Greco,  (from Crossroads Technologies) was great success. It was full room of almost 300  Java developers looking for the fast way to get into the magic world of machine learning, but using what they allready know – Java, and skiping all the confusing parts for the beginners related to math formulas and statistics.
Judging by the comments and discussion after the session we have succeded in making fast introduction, and explaining essentials with hands on approach that typical Java developer need to get started and make next steps. The session included brief explanationof basic supervised machine learning techniques including:

  • Linear Regression
  • Logistic Regression
  • Feed Forward Neural Networks
  • Convolutional Neural Networks

This presentation will be the base for the series of blog posts on this topic.
Slides are available here


Deep Learning and Machine Learning using Deep Netts at Oracle Code One 2018

Our CEO Zoran Sevarac is speaking at two sessions at Oracle Code One 2018 in San Francisco. One about basics of machine learning, and the other about using deep learning for application perfromance tuning. Oracle Code One (ex. Java One) is the biggest and most influential conference for Java technologies in the world.
You can also find him at Oracle Developer Champions briefing on Friday 10.19. at Oracle,  Java Champions Briefing on Saturday 10.20. in Parc 55, and Java Influencers Gathering at Google Developers Launchpad Space on Thursday.  More details are bellow:

Java and Machine Learning  at Java Champions Briefing 2018

Overview of the current Java machine learning landscape, what needs to be done in order to make Java no 1 platform for machine learning development, and ongoing efforts in that domain.

2.00 p.m. - 2.30 p.m. | Parc 55 Hotel San Francisco, Market Room

Deep Learning for Application Performance Optimization

Application performance tuning usually involves periodically monitoring and adjusting several parameters that control the runtime environment, including the CPU, memory, threading, garbage collection, and more. This session presents the experience of the speaker and his team in building deep learning models for autonomous, continuous application performance tuning. The presentation includes a methodology, architecture, and best practices for building such systems.
The participants will learn how to build deep learning models for modeling application performance for various configuration settings. A case study is based on tuning a Java enterprise application but can be generalized for other types of applications or individual components.

Monday, Oct 22, 7:30 p.m. - 8:15 p.m. | Moscone West - Room 2005

If you’re comming to Oracle Code One schedule this session at: https://oracle.rainfocus.com/widget/oracle/oow18/catalogcodeone18?search=BOF4967

Machine Learning for Software Developers in 45 Minutes

Technologists know that machine learning (ML) is a humongous, long-term trend for all applications. Many companies are now applying an “AI first” philosophy in all their applications and services. This session is designed for Java developers who are new to ML and need an overview of ML and where/how to apply it.
You’ll learn about its relationship to AI and why ML is evolving at such a rapid pace.  The session also describes the types of learning styles—supervised learning, unsupervised learning, and reinforcement learning—and also discusses “deep learning.”
It covers use cases and the various ML services offerings from Oracle Cloud, AWS, Azure, and GCP. Most importantly, the focus is on how Java developers can take advantage of ML in their applications.

Monday, Oct 22, 1:30 p.m. - 2:15 p.m. | Moscone West - Room 2016

If you’re comming to Oracle Code One schedule this session at: https://oracle.rainfocus.com/widget/oracle/oow18/catalogcodeone18?search=DEV5090

Practical Machine Learning at QCon New York 2018

QCon had an amazing machine learning  track that presented parctitioners from some of the biggest players in indutry like Pay Pal and Net Flix,  but also provided some real world experiences from building data science and machine learning teams, and adding machine learning skills to existing software development teams. Did you know that without using machine learning for fraud detection, Pay Pal would not be possible? Given the currrent scale, growth and rate of attacks, machine learning is the only way to secure the business.

At Net Flix they use an interesting approach for ensuring quality of their service, which is based on using machine learning techniques to augment the work of human engineers. The basic idea is not to leave everything to the machines (which is also not possible), but to use machine learning to help human engineers to detect and resolve issues more quickly.

It was very interesting to hear how about entire machine learning pipelines used at Door Dash to predict and improve food delivery. One of the key points I liked about this preseentations was advice about evolving machine learning models. In the beggining nobody has enough data, so start with something simple and improve model over time, as th eamount of data increase.

In a very interesting talk about team building by Sally Radwan, she discovered that once the company decided to include machine learning, that resulted in 60% of the team change in a period of one year. And interesting thing was that most of the developers left, while managerial positions were stable. That doesn’t have to be rule but be aware of the posibility.

Deep Netts platform was presented during a session for application performance tuning using deep learning, whics was very well attended and we got some very usefull feedback.

I had a great honour of hosting this track, working with all these great people and I learned a lot alonf the way. So thanks QCon for the great conference!