About a year ago, Google announced the launch of Vertex AI, a managed artificial intelligence platform designed to help organizations accelerate the deployment of AI models. To celebrate the service’s anniversary and launch the Google Applied ML Summit, Google announced new features heading to Vertex this morning, including a dedicated AI training server and “example-based” explanations.
“We launched Vertex AI a year ago with the goal of enabling a new generation of AI to enable data scientists and engineers to do satisfying and creative work,” Henry Tappen, Google Cloud Group Product Manager, told TechCrunch via email. “The new Vertex AI capabilities that we are launching today will continue to accelerate the deployment of machine learning models in organizations and democratize AI so more people can deploy models in production, continuously monitor and drive business impact through AI.”
As Google has always shown, the advantage of Vertex is that it brings Google Cloud AI services together under a unified user interface and API. Customers like Ford, Seagate, Wayfair, Cashapp, Cruise and Lowe’s use the service to build, train and deploy machine learning models in a single environment, Google says, while moving models from experiment to production.
Vertex competes with managed AI platforms from cloud service providers such as Amazon Web Services and Azure. Technically, it falls under the category of platforms known as MLOps, which are a set of best practices for companies to run AI. Deloitte expects the MLOps market to be worth $4 billion in 2025, with a growth rate of nearly 12 times since 2019.
Gartner predicts that the emergence of managed services like Vertex will drive cloud market growth by 18.4% in 2021, with clouds expected to account for 14.2% of total global IT spending. “As organizations increase their investments in mobility, collaboration, and other technologies and infrastructure to work remotely, the growth of public cloud [will] to last through 2024,” Gartner wrote in a November 2020 study.
Among the new Vertex features is the AI Training Reduction server, a technology that Google says improves bandwidth and response time for distributed, multi-system training on Nvidia GPUs. In machine learning, “distributed training” refers to spreading the work of a system’s training across multiple dedicated hardware, GPUs, CPUs, or chips, thereby reducing the time and resources needed to complete the training.
Andrew Moore, Vice President and General Manager of Cloud Artificial Intelligence at Google, said in an article published today on Google Cloud Articles. “In many critical business scenarios, an abbreviated training course allows data scientists to train a model with higher predictive performance within the confines of the deployment window.”
In preview, Vertex also introduces scheduled workflows, which are intended to bring greater customization to the model building process. As Moore explained, Tabular Workflows allows the user to choose which parts of the workflow they want Google’s “AutoML” technology to deal with versus the parts they want to design themselves. AutoML, or Automated Machine Learning – which is not limited to Google Cloud or Vertex – includes any technology that automates some aspect of AI development and can manipulate the development stages from inception with an initial data set to creating a machine learning model ready for deployment. AutoML can save time, but it can’t always beat the human touch, especially when precision is required.
“Scheduled workflow elements can also be integrated into existing Vertex AI pipelines,” Moore said. “We have added new managed algorithms, including advanced search forms such as TabNet, new algorithms for feature selection, form distillation and…more.
For development pipelines, Vertex is also gaining (preview) integration with Spark Serverless, the serverless version of the Apache-powered open source analytics engine for data processing. Now Vertex users can start a serverless Spark session for interactive code development.
Elsewhere, customers can analyze the data features of the Neo4j platform and then publish models using Vertex through a new partnership with Neo4j. Thanks to the collaboration between Google and Labelbox, it is now easy to access Labelbox’s data naming services for image, text, audio and video data from the Vertex dashboard. Labels are essential for most AI models to learn how to make predictions; Models are trained to identify re-entry labels, also called annotations, and sample data (eg, “frog” caption and frog image).
In the event that the data is mislabeled, Moore offers explanations based on examples as a solution. Available in preview, the new Vertex features take advantage of “example-based” explanations to help diagnose and address data issues. Of course, there is no explainable AI technology that can detect all errors; Computational linguist Vagrant Gautam warns against overconfidence tools and techniques used to explain artificial intelligence.
“Google has documentation about the limitations and a more detailed white paper on interpretable AI, but none of that is mentioned anywhere. [today’s Vertex AI announcement]They told TechCrunch via email. The ad asserts that “skills competency should not be the criterion for participation” and that the new features it provides can “develop AI for non-software experts.” My concern is that non-experts have more confidence in AI and its ability to explain it more than they should, and now many Google clients can create and publish models more quickly without stopping to wonder if this is an issue that needs a machine learning solution in the first place, and invoke their models explainable (and thus trustworthy and good) without knowing the full scope limits around that for their particular cases. »
However, Moore suggests that example-based explanations can be a useful tool when used in conjunction with other model-checking practices.
“Data scientists shouldn’t be infrastructure or process engineers to keep models accurate, interpretable, scalable, disaster-resistant, and secure in an ever-changing environment,” Moore added. “Our customers require tools to easily manage and maintain their machine learning models.”