Recommended
More Related Content
What's hot
What's hot (20)
Similar to JSR381 Visual Recognition for Java.pdf
Similar to JSR381 Visual Recognition for Java.pdf (20)
Recently uploaded
Recently uploaded (20)
JSR381 Visual Recognition for Java.pdf
- 1. Zoran Sevarac - Univ of Belgrade - @zsevarac Frank Greco - NYJavaSIG - @frankgreco JSR 381 Visual Recognition for Java Finally, A Java-Friendly ML API Gran Sasso
- 2. Frank Greco @frankgreco Chair NYJavaSIG - NY Java User Group JC, Cloud/Mobile Architect NullPointerExceptional Zoran Severac @zsevarac AI Researcher Univ of Belgrade, Serbia JC
- 3. Session Topics ● Why JSR 381 (VisRec)? ● Patterns and Machine Learning (ML) ● VisRec Architecture and Implementation ● VisRec API ● Examples
- 4. Why JSR381?
- 5. Machine Learning - Huge Impact on Everything...
- 6. Artificial Intelligence Machine Learning Deep Learning A type of algorithm(s) that allows a machine to emulate aspects of intelligent human behavior A type of AI that allows a machine to learn from experience/data A type of ML that uses powerful computing resources and advanced neural networks to more-accurately solve non-linear, highly-dimensional problems with large amounts of data (eg, visrec)
- 7. Learning a Foreign Language Understand All Rules Observe/Follow Patterns
- 8. Patterns...
- 9. 1995 1987 - Kent Beck and Ward Cunningham - OOPSLA - “Using Pattern languages for OOP”
- 10. 1977 Directly led to Ward Cunningham’s creation of the first wiki Influenced the development of Agile https://en.wikipedia.org/wiki/Christopher_Alexander
- 12. ii V I I IV V
- 13. “One of the most interesting aspects of the world is that it can be considered to be made up of patterns” Norbert Wiener - 1894-1964 - MIT Cybernetics (1948) - early steps to modern AI This was AI before it was called “AI”
- 14. The Core of Machine Learning
- 15. The Core of Machine Learning Recognizing PATTERNS in data and Making PREDICTIONS against that data.
- 16. Recognizing patterns in your data, anticipating the likely future behavior of similar data and taking steps to maximize ethical business benefit based on those predictions. The Core of Machine Learning for Enterprises
- 17. You can write your Machine Learning app in any language you want, as long as it’s Python.
- 18. You can write your Machine Learning app in any language you want, as long as it’s Python. HEY!
- 20. Java Needs to be a First Class Citizen for ML 1. Don’t want another JavaScript Doomsday Scenario occurring in our Lifetime 2. Python is a good language. There should be others to express the world’s creativity. 3. Java is a great language with a huge ecosystem with 12M developers 4. Existing ML APIs aren’t Java-friendly or AppDev-friendly
- 22. Solution: ...We need to promote Java for ML... “The best way to predict the future is to invent it.” - Alan Kay
- 23. Solution: ...We need to promote Java for ML... “The best way to predict the future is to invent it.” - Alan Kay Make ML more approachable for Java developers!
- 24. Everyone we talked was interested to collaborate
- 25. Don’t boil the ocean… [heat up some coffee first] VisRec - JSR #381 Visual Recognition for Java Spec Leads and Expert Group Zoran Sevarac (ML researcher - Univ of Belgrade) Frank Greco (NYJavaSIG, Consultant) Kevin Berendsen (OpenValue, Senior Software Eng)
- 26. Other contributors and Advisors ● Frank Liu (Amazon) ● Constantin Drabo ● Amit Nagesh ● Marissa Staller ● Eric Bruno ● Anakar Parida ● Nikita Ivanov (Gridgain) ● James Weaver (IBM) ● Werner Keil ● Jyoti Buddha ● Guillaume Laforge (Google) ● Ed Burns (fmr Oracle) ● Nishant Raut (Mumbai JUG) ● Sandhya Kapoor (fmr IBM)
- 28. JSR #381 - Visual Recognition API ● Java-friendly standard API for computer vision tasks using ML ● Also provides generic ML API ● Designed for Java app devs ● Multiple implementations ● High level abstractions ● More readable and easier to use VisRec JSR #381 https://github.com/JavaVisRec/visrec-api/wiki/Getting-Started-Guide
- 29. JSR381 API, Architecture, Implementations, and Examples
- 30. Issues with Existing Offerings ● Each has different set of APIs ● Different sets of algorithms, image classes, native dependencies ● Complex for Avg Java Developer ● Not Java-friendly (C flavor)
- 31. VisRec Design goals and principles ● Easy to use by non-experts in VisRec/ML domain ● Enable Java Developers to leverage their Java skills ● Easy to implement for existing ML and imaging libraries ● Reusable design for applied ML applicable to other domains
- 32. What Can you Do with It? 1) Classify images and data/objects classification task / differentiate between categories 2) Estimate value regression task 3) Object Detection classification + regression(position)
- 33. Generic ML Layer Visual Recognition Image Abstraction VisRec Public API Apps/Libs Visual Recognition Layer JSR381 Service Provider Interface ML Engine 1 Service Provider Impl ML Engine 2 Service Provider Impl JSR 381 Architecture
- 34. Core Classes & Design Generic ML Layer (common abstractions, interfaces for general ML tasks) Visual Rec Layer (specialization of generic ML tasks for specific objects / domain) Implementation Based on specific ML engine DataSet<E> ImageSet extends DataSet<ExampleImage> Classifier<T, R> BinaryClassifier<T> MutliClassClassifier<T, R> ImageClassifier<IMAGE_CLASS> AbstractImageClassifier<IMAGE_CLASS , MODEL> implements ImageClassifier<IMAGE_CLASS> ImageClassifierNetwork extends AbstractImageClassifier<BufferedImage, ConvolutionalNetwork> Regressor<T, R> SimpleLinearRegression< MODEL_CLASS> implements Regressor<Float, Float>, SimpleLinearRegressionNetwork extends SimpleLinearRegression<FeedForwardN etwork>
- 35. Hot Dog or Not Hot Dog Demo: Hot Dog or Not Hot Dog Scene from TV show Silicon Valley https://www.youtube.com/watch?v=vIci3C4JkL0
- 37. Building the classifier Full source at GitHub https://github.com/sevarac/HotDogNotHotDog
- 38. Training process What happens during the training: ● Example images are presented to the classifier iteratively ● Internal parameters are automatically tuned in order to minimize error ● Error should go DOWN through the epochs ● Accuracy should go UP through the epochs
- 39. How to use the trained model Actual recognition is performed here Full source at GitHub https://github.com/sevarac/HotDogNotHotDog
- 40. Under the hood: Convolutional Neural Network Automated image preprocessing and feature extraction + Classifier
- 41. Minimal Configuration for Building an Image Classifier ● Image class ● Image dimensions ● Image category labels (labels file) ● List of image files with corresponding category labels (index file) ● Architecture of convolutional neural network ● File name to export the trained model ● Optionally training parameters (max epochs, max error, learning rate)
- 42. ● Many different incompatible data formats ● Many different machine learning algorithms ● Many confusing configuration parameters ● Clear task oriented interface that hides implementation details ● Usage does not require understanding of implementation and configuration (like Collections.sort()) ● Simple and portable integration into existing Java apps and devices Pains, barriers and preferences for ML in Java SOLVED!
- 43. Reference Implementation •Reference implementation of VisRec API is based on Deep Netts Community Edition which is a Java based deep learning library provided by www.deepnetts.com More info: https://www.deepnetts.com/blog/deep-netts-community-edition • DeepNetts has released community edition of their deep learning engine, and contributing their experience through this JSR as a part of their open source strategy
- 44. How to get started github.com/JavaVisRec/visrec-api/wiki/Getting-Started-Guide
- 45. JSR381 Information • JCP.org https://jcp.org/en/jsr/detail?id=381 • JSR project on GitHub github.com/JavaVisRec • Getting Started github.com/JavaVisRec/visrec-api/wiki/Getting-Started-Guide • API github.com/JavaVisRec/visrec-api • RI github.com/JavaVisRec/visrec-ri • Examples github.com/JavaVisRec/jsr381-examples • Mailing list visrec@groups.io
- 46. JSR 381 Visual Recognition for Java Finally, A Java-Friendly ML API Please provide feedback and contribute! Java is a great language for ML!