KISS Java EE, MicroProfile, AI, (Deep) Machine Learning

An airhacks.fm conversation with Pavel Pscheidl (@PavelPscheidl)
about: Pentium 1 with 12, 75 MHz, first hello world with 17, Quake
3 friend as programming coach, starting with Java 1.6 at at the
university of Hradec Kralove, second "hello world" with Operation
Flashpoint, the third "hello world" was a Swing Java application as
introduction to object oriented programming, introduction to
enterprise Java in the 3rd year at the university, first commercial
banking Java EE 6 / WebLogic project in Prague with mobile devices,
working full time during the study, the first Java EE project was
really successful, 2 month development time, one DTO, nor
superfluous layers, using enunciate to generate the REST API, CDI
and JAX-RS are a strong foundation, the first beep, fast JSF, CDI
and JAX-RS deployments, the first beep, the War of Frameworks,
pragmatic Java EE, "no frameworks" project at telco, reverse
engineering Java EE, getting questions answered at airhacks.tv,
working on PhD and statistics, starting at h2o.ai, h2o is a
sillicon valley startup, h2o started as a distributed key-value
store with involvement of Cliff Click, machine learning algorithms
were introduced on top of distributed cache - the advent of h2o,
h2o is an opensource company - see github, Driverless AI is the
commercial product, Driverless AI automates cumbersome tasks, all
AI heavy lifting is written in Java, h2o provides a custom
java.util.Map implementation as distributed cache, random forest is
great for outlier detection, the computer vision library openCV,
Gradient Boosting Machine (GBM), the opensource airlines dataset,
monitoring Java EE request processing queues with GBM, Generalized
Linear Model (GLM), GBM vs. GLM, GBM is more explained with the
decision tree as output, XGBoost, at h2o XGBoost is written in C
and comes with JNI Java interface, XGBoost works well on GPUs,
XGBoost is like GBM but optimized for GPUs, Word2vec, Deep Learning
(Neural Networks), h2o generates a directly usable archive with the
trained model -- and is directly usable in Java, K-Means, k-means
will try to find the answer without a teacher, AI is just
predictive statistics on steroids, Isolation Random Forest, IRF was
designed for outlier detection, and K-Means was not, Naïve Bayes
Classifier is rarely used in practice - it assumes no relation
between the features, Stacking is the combination of algorithms to
improve the results, AutoML: Automatic Machine Learning, AutomML
will try to find the right combination of algorithms to match the
outcome, h2o provides a set of connectors: csv, JDBC, amazon S3,
Google Cloud Storage, applying AI to Java EE logs, the amount of
training data depends on the amount of features, for each feature
you will need approx. 30 observations, h2o world - the conference,
cancer prediction with machine learning, preserving wildlife with
AI, using AI for spider categorization

Pavel Pscheidl on twitter: @PavelPscheidl, Pavel's blog:
pavel.cool