Welcome to the JUG Oberpfalz Online Live Coding Sessions!
Our online sessions are held in English and streamed live on YouTube. The recording will be available afterwards. Join the live chat, ask questions, and follow along as we explore what the next generation of software development might look like.
This session is hosted by Johannes Rabauer with special guest Ronald Dehuysser from JobRunr.
Session Topic
Hold a Java trading card up to a webcam, and an app tells you which card it is. That is what Johannes builds live in this session, with Ronald Dehuysser supervising and steering.
For that, the app has to prepare pictures of 20 cards we made (Duke, a few frameworks, a legacy monolith), train a model on them and measure how well it did. Each step takes time, so each one runs as a JobRunr background job, with its progress in the JobRunr dashboard. The model is not built from scratch: we start from an existing pretrained image model and train it on our cards. That is called transfer learning.
The session starts with a short talk about why training is slow work and what that means for a Java app, then moves to the live build. We do not know yet how well it will work, and mistakes get named on stream.
What we'll do
* Talk about why training a model is slow work and what that means for a Java app
* Turn 20 Java trading cards into training pictures
* Run preparing the pictures, training and evaluating as JobRunr background jobs
* Follow the jobs in the JobRunr dashboard
* Train an existing pretrained image model on our cards (transfer learning)
* Hold a card up to a webcam and see what the app says, including where it is wrong
* Answer questions from the community
Topics
* JobRunr
* Background jobs
* Image classification
* Transfer learning
* Java
* Machine learning
* Job scheduling
* Live coding
Whether you are a Java developer who runs background work, someone curious about adding machine learning to a Java app, or here for the trading cards, you are welcome to join.
See you in the live chat!
