Google’s Announcement to levelling up the Vertex AI:
Almost a year ago, Google announced the debut of Vertex AI, a managed AI platform aimed to help businesses install AI models more quickly. On June 9, 2022, Google unveiled new features for Vertex, including a dedicated server for AI system training and "example-based" explanations, to commemorate the service's anniversary and the start of Google's Applied ML Summit. As a result of the new Vertex AI capabilities, enterprises can build machine learning models faster, monitor them continuously, and use AI to drive business results. As Google has previously stated, the value of Vertex is that it unifies Google Cloud AI services into a single UI and API. As a result, customers such as Ford, Seagate, Wayfair, Cashapp, Cruise, and Lowe's, according to Google, utilise the service to design, train, and deploy machine learning models in a single environment, allowing them to go from testing to production. Vertex competes with cloud providers such as Amazon Web Services and Microsoft Azure's managed AI platforms. In addition, it belongs to the MLOps platform category, a set of best practices for enterprises using AI. According to Deloitte, the market for MLOps will be worth $4 billion in 2025, up roughly 12 times from 2019. According to Gartner, the cloud market will expand 18.4 per cent in 2021 due to managed services like Vertex, with the cloud accounting for 14.2 per cent of worldwide IT spending. "Growth in public cloud [will] be continued through 2024 as organisations expand investments in mobility, collaboration, and other remote working technologies and infrastructure," Gartner noted in a November 2020 research.
Prediction Service by Google:
It has the following features, according to Google Vertex AI Product Manager Surbhi Jain:
- Vertex AI's Prediction Service is a new integrated component: This is where it comes into play when users have a trained machine learning model and are ready to start servicing requests from it. The objective is to make it entirely frictionless to assure safety and scalability. In addition, they seek to make deploying an ML model in production cost-effective, regardless of where the model was trained.
- A completely managed service: Because Vertex AI is fully managed, the overall service cost is minimal. The requirement for over-provisioning hardware is reduced thanks to seamless auto-scaling.
- Prediction Service supports several VM and GPU types, allowing developers to choose the most cost-effective hardware for a particular model. Furthermore, compared to open source, they have various proprietary improvements in their backend that further cut costs. They also have strong integrations with the platform's other components.
- Prebuilt components to deploy models from pipelines regularly are built-in integration for request-response logging in BigQuery in Stackdriver. A prediction service also has intelligence and assertiveness, which means it can follow the model's performance after being deployed in production and comprehend why it generates particular predictions.
- Security and compliance are built-in: Your models may be deployed within your private perimeter. In addition, their integration control tool, PCSA (pre-closure safety analysis), has access to your endpoints and ensures that your data is always secure. Finally, Prediction Service adds less than two milliseconds of extra delay with private endpoints.
More tools and capabilities of Vertex AI:
- The public preview of an optimised TensorFlow runtime was published, allowing for delivering TensorFlow models at a lower cost and latency than open-source prebuilt TensorFlow serving containers. In addition, the newly improved TensorFlow runtime enables users to take advantage of Google's proprietary technologies and model optimisation approaches.
- The AI Training Reduction Server is one of the new features of Vertex. According to Google, it improves the bandwidth and latency of multisystem distributed training on Nvidia GPUs. "Distributed training" is a term used in machine learning to describe distributing the work of training a system over numerous workstations, GPUs, CPUs, or custom chips to reduce the time and resources required to complete the training.
- In a closed preview, Google also released bespoke prediction algorithms, making pre-processing model input and post-processing model output as simple as creating a Python function. They've also linked it with Vertex SDK, allowing customers to create unique containers with their custom predictors without having to write a model server or know anything about Docker. It also makes it simple for users to test the produced images locally.
- With Vertex AI, Google claims that training a model requires 80% fewer lines of code than competing platforms, providing data scientists and machine learning engineers with the capability to build and manage ML projects throughout their lifecycles efficiently.
