February 2, 2023
Machine learning (ML) is, justifiably, receiving a lot of attention today. ML is helping companies in every industry identify performance issues, solve complicated problems, weigh resolutions and their outcomes, and make optimal decisions. From delivering the highest-quality healthcare and preventing cybercrimes to determining the best transportation route, ML is invisibly impacting virtually every part of our lives today.
Machine learning as a service (MLaaS) is helping to support the widespread adoption and application of ML. By capitalizing on a SaaS (software-as-a-service) delivery model tuned for ML workloads, organizations can quickly join the machine learning revolution. In the MLaaS model, key ML capabilities — such as data management, predictive analytics, trained algorithms, and deep learning methods — are delivered via the cloud, with complex calculations and analysis taking place on the provider’s extensive computational resources.
Because MLaaS is placing the power of sophisticated machine learning techniques within the reach of just about every company, it represents a fast-growing field. According to a recent study by Research and Markets, the global MLaaS market is expected to grow from $2.26 billion U.S. in 2021 to $16.7 billion by 2027, reflecting an annual growth rate of 39.25%.
As we’ve outlined in previous blogs, at Ansys we’re applying machine learning to make both simulation solutions and associated processes faster, smarter, and more efficient. Our forward-looking research and development efforts are leveraging ML to drive improvements in geometrical representation, surface-contact detection, and speeding up complex physics — and these are just a few examples.
Given our commitment to optimizing customers’ results via ML, it shouldn’t come as a surprise that Ansys is also exploring the potential to deliver simulation-related MLaaS in the not-too-distant future. This will enable Ansys customers to import their own simulation data, design their own ML models, and use a library of pre-built models to design a full working ML pipeline for simulation using Ansys tools.
While there are existing general-purpose business solutions that enable companies to build a data pipeline for machine learning — including AWS Sagemaker and Microsoft Azure ML — none of these platforms focus on applying ML to simulation workloads.
Both AWS Sagemaker and Azure ML provide a rich set of capabilities that enable data scientists, machine learning engineers, and developers to prepare, build, train, and deploy ML models rapidly and with ease. However, for simulation-heavy datasets, the starting point is not images or structured data — which are supported by these frameworks.
For simulation workloads, the visualization of geometries or results is a key component that Ansys MLaaS delivers. While Ansys MLaaS also supports tabular data and image-based datasets, it is unique in its ability to use simulation results as its starting point, as well as various geometry formats such as STL or IGES.
The Ansys research and development team is creating web services to accelerate building, testing, and deploying ML models for simulation. These services are loosely coupled to Ansys physics-based solvers to help users generate a large volume of training data, either offline on online.
Ansys MLaaS enables quick visualization of training data with embedded graphics, while supporting all key metrics in ML workloads such as training, validation, and testing plots. Training data can come from offline sources — such as files generated from Ansys solvers — or online sources such as Onscale or PyAnsys. MLaaS from Ansys is also interoperable with all widely used computer-aided design/computer-aided engineering (CAD/CAE) tools.
Ansys is partnering with two leading global providers of MLaaS, Microsoft Azure and AWS, to deliver these new services as a point solution that’s not tied to any specific cloud provider. While Ansys MLaaS is currently in beta testing, we hope to make this value-added offering available to customers sometime in 2023. The initial Ansys MLaaS offering will focus on ML as a point service for simulation workloads. But, in the future, it will provide connections to other Ansys solvers to provide a vehicle for training data generation.
Why should you be excited about this new, easy accessibility to ML for your simulation workloads? The answer is simple: speed, along with high fidelity. As you pose new questions to your Ansys solvers, they will no longer be starting from scratch. Instead, they can leverage the power of machine learning to look at huge volumes of existing simulation results and extract key learnings that greatly accelerate your new simulation — and, in turn, your overall product development process.
By accessing the Ansys ML model library, developed by a company with over 50 years of simulation leadership, you can achieve a meaningful edge as you launch new product designs faster than ever, without shortchanging your analysis. To maximize flexibility and relevance, you can also bring your own models and data into the MLaaS ecosystem.
The advantage is a significant one. Complex physics-based problems that used to take hours to solve will instead be solved in fractions of a second via the power of ML algorithms. These algorithms have been trained by Ansys experts, giving your product development team a powerful advantage in getting to results faster — no matter how complicated the problem is.
To get a sneak peek of the Ansys ML model library, check out the video below.