Google to preview new Vertex AI tools at its virtual AI Summit
Google announced a new set of product features and partnerships Thursday for its vertex AI platform designed to make deploying machine learning models into production environments easier at scale. The new tools, features and partnerships will be previewed at its virtual Applied AI summit today at noon EDT.
New Vertex AI features
Training Reduction Server
Vertex AI Training Reduction Server, which supports both Tensorflow and PyTorch, optimizes bandwidth and reduces latency of multi-node distributed training on NVIDIA GPUs. According to Google, TRS “significantly reduces the training time required for large language workloads, like BERT, and further enables cost parity across different approaches.”
TRS also simplifies the deployment of Jupyter Notebooks by reducing 12 deployment steps to a single click. This feature is designed to help eliminate routine tasks and accelerate ML deployment into production.
Tabular Workflows includes a glassbox and managed AutoML pipeline that lets users see and interpret each step in the model building and deployment process. This allows data scientists to train large datasets of over a terabyte without sacrificing accuracy. Users can choose which parts of the process they want to automate and which parts they engineer themselves.
Elements of Tabular Workflows can also be integrated into existing Vertex AI pipelines. Google also added new managed algorithms, including advanced research models like TabNet, model feature selection and model distillation. Future additions to Workflows will include Google proprietary models such as Temporal Fusion Transformers as well as open source models such as XGboost and Wide & Deep.
To fast track the deployment of ML models into production and further integrate data modeling capabilities directly into the data science environment, Google announced the Serverless Spark tool as well as partnerships with Neo4j and Labelbox to help ML model builders work with structured data, graph data and unstructured data.
For structured data, Google Serverless Spark will allow data scientists to launch a serverless spark session on their notebooks and interactively develop code.