Tribuo is a Java machine learning (ML) library, developed by Oracle Labs, used inside Oracle since 2017, and open sourced in 2020. It has several features that make it suitable for enterprise deployments of ML models. It can train and deploy ML models in Java, and tracks the provenance of the models it trains. Each model is self describing; it knows its training data, how that data was processed, and all the algorithmic details that went into building it. Tribuo also can export models in ONNX format and integrates with popular libraries like TensorFlow and XGBoost.
In this presentation, we’ll cover the overall design of Tribuo, how to import and export models to/from it, and run through some demos showing model training, provenance inspection and model import.
Adam Pocock is an ML researcher in Oracle Labs’ Machine Learning Research Group. He joined Oracle Labs after finishing his PhD in 2012, and has worked on scaling up Bayesian inference using GPUs, feature selection, and Natural Language Processing among other areas. Recently he’s been working on the open source ML ecosystem in Java. He leads the development of Tribuo, co-leads the TensorFlow-Java project, and maintains the Java interface to ONNX Runtime.