TigerGraph launches workbench for ML/AI modeling of graph neural networks

TigerGraph, makers of a graph analytics platform for data scientists, today during its Graph & AI Summit event unveiled its TigerGraph ML (Machine Learning) Workbench, a new-generation toolkit that will reportedly enable analysts to measure the accuracy of ML models and to shorten development cycles.

Workbench does this using familiar tools, workflows and libraries in a single environment that plugs directly into existing data pipelines and ML infrastructure, TigerGraph VP Victor Lee told VentureBeat.

The ML Workbench is a Jupyter-based Python development framework that enables data scientists to create deep learning AI models with connected data straight from the enterprise. Graph-based ML has been shown to have more accurate predictive power and requires far less runtime than the traditional ML approach.

Traditional machine learning algorithms rely on learning systems through training sets to develop a trained model. This pre-trained model is used to classify or recognize the test data set; this can typically take days or weeks to complete for a specific use case. Graph-based ML can sometimes take minutes to create an algorithmic model.

The value of ML is high, but so is the learning curve

“Graph has been shown to accelerate and improve ML learning and performance, but the learning curve for using the APIs (Application Programming Interfaces) and libraries to achieve this has proven to be very steep for many data scientists,” Lee said in a media advisory. “That’s why we created ML Workbench to provide a new layer of functionality between the data scientists and the graph machine learning APIs and libraries to facilitate data storage and management, data preparation and ML training.

“In fact, we’ve seen early adopters increase the accuracy of their ML models by 10-50% by using ML Workbench and TigerGraph,” he said.

TigerGraph’s entire mindset revolves around defining human identity based on how you interact with others, Lee told VentureBeat.

“The same is true for graphs in data modeling, and that extends to neural networks right now,” Lee said. “Each node in a graph is connected like people. Charts are great for querying pattern recognition algorithms. Workbench helps you provide machine learning based on the information in the graph, but its real power lies in graph neural networks, which are regular graphs on steroids.

“For example, in our DGL (Deep Graph Library) there is an extension of (metas)Pytorch geometry that supports neural graph networks,” he said. “This is a great feature and shows that we’re going where the data scientists are; We’re not trying to get them to learn anything new. We use the tools they already know and are comfortable with because we’re trying to shorten the learning curve.”

Best for fraud and prediction use cases

The ML Workbench enables companies to discover enhanced insights into node prediction applications such as fraud and edge prediction applications, which include product recommendations, Lee said. The ML Workbench enables AI/ML practitioners to explore graph-based machine learning and graphing neural networks (GNNs) because it is fully integrated with TigerGraph’s database for parallelized graphing data processing/manipulation, Lee said.

The ML Workbench is designed to work with popular deep learning frameworks such as PyTorch, PyTorch Geometric, DGL, and TensorFlow, giving users the flexibility to choose a framework they are most familiar with. The ML Workbench is also plug-and-play for Amazon SageMaker, Microsoft Azure ML, and Google Vertex AI, Lee said.

The ML Workbench is designed to work with enterprise-level data. Users can train GNNs – even on very large graphs – thanks to the following built-in capabilities:

• TigerGraph DBs distributed storage and massively parallel processing;
• Graph-based partitioning generate training/validation/test graph data sets;
• Chart-based batch processing for GNN mini-batch training to improve performance and reduce HW requirements; and
• subgraph sampling to support leading GNN modeling techniques.

ML Workbench is compatible with TigerGraph 3.2 and higher, available as a fully managed cloud service and for on-premises use. Currently in preview, ML Workbench will be generally available in June 2022, Lee said.

TigerGaph competes with Neo4J, ArangoDB, MemGraph and a few others in the graph database space.

Million Dollar Challenge winners selected

At the Graph & AI Summit, TigerGraph announced the winners of the Graph for All Million Dollar Challenge – awarding $1 million in cash to groundbreaking, graph-based projects that solve many of today’s biggest global social, economic, health and community issues analyze and address climate-related concerns.

The winning projects, announced at this week’s Graph + AI Summit, were selected by the global jury from more than 1,500 entries from over 100 countries. Mental Health Hero won the grand prize of $250,000 for creating an application designed to help make mental health treatment more accessible and personal.