BY ALTAIR

Earlier this year we wrote that AI-powered engineering is the answer to today’s furious pace of product development and ever more competitive markets. As we said then, “Our AI-powered engineering workflows can deliver predictions that are up to 1,000x faster than traditional simulation solvers, enabling teams to evaluate more concepts and make better design decisions.”

Even though it sounds too good to be true, that stat is no joke. The all-too-often unsung hero driving this unprecedented acceleration is a technology called geometric deep learning. In a market increasingly defined by the widespread use of enterprise-level AI, geometric deep learning has received scant attention, despite its groundbreaking capabilities. Let’s explore geometric deep learning: what it is, how it works, and why it’s likely going to be one of the most powerful technologies in any organization’s AI toolkit in the coming years.

AN OVERVIEW OF GEOMETRIC DEEP LEARNING

Coined in 2016 by University of Oxford professor Michael Bronstein, geometric deep learning is a specific type of machine learning. Put simply, it helps machine learning algorithms and neural networks learn to identify 3D objects and patterns on any kind of geometric surface. More specifically, it’s an umbrella term for “emerging techniques attempting to generalize (structured) deep neural models to non-Euclidean domains such as graphs and manifolds.”

Though the technology has been around for nearly a decade, it has received surprisingly little attention outside of technical publications and academic journals. Unlike generative AI (genAI), its flashier distant relative, geometric deep learning hasn’t captured the attention of the world’s largest companies. This is because nobody has really unleashed its potential.

Only within the past few years have leading tech companies—like Altair—taken a keen interest in developing and implementing geometric deep learning capabilities within their portfolios. In fact, there’s reason to believe geometric deep learning in the coming years will become as impactful and widespread as genAI promises to be.

WHAT MAKES GEOMETRIC DEEP LEARNING SO EFFECTIVE

Seemingly niche, at least when compared to genAI, geometric deep learning nonetheless promises to transform organizational engineering workflows at a fundamental level, especially those involving simulations. “There are only a few people that are aware of this technology, how it applies to simulation, and how useful it can be,” said Altair expert Fatma Kocer, vice president of engineering data science. “But what geometric deep learning does is help optimize the way people think about and run simulations.”

Traditionally, for various reasons, engineers and designers run one simulation at a time and then tweak different key parameters in a trial-and-error manner. This work is time intensive, as these large-scale individual simulations can take multiple days to run. This method is also often incredibly computationally and financially resource intensive because of energy, compute power, and licensing costs. Geometric deep learning speeds and streamlines the entire simulation cycle, delivering true AI-powered transformation.

ADVERTISEMENT

“What geometric deep learning can do is take your existing simulations, the designs you’ve already run—even simulation results you’ve only used once—and allow you to make informed decisions in the future based on that data, even if that data is specific to a point in time,” Kocer said. “The technology lets you train a machine learning model that enables you to do as much design exploration as you like. This also increases the value of simulations you ran in the past, because now you’re using them to make design decisions for future projects.”

In short, geometric deep learning introduces a whole new dynamic to organizations’ simulation capabilities and workflows, offering unprecedented speed and ease via machine learning. “It allows you to do design exploration in a whole new way. It optimizes your simulation efforts, because now you’re able to explore and identify high potential designs much more quickly, easily, and at far less computational and financial cost,” Kocer said. “From there, you can then do simulation on those designs, as opposed to doing one-by-one, trial-and-error simulation on whatever comes to mind without really having an idea if it’s going to help you or not.”

THE FUTURE OF GEOMETRIC DEEP LEARNING IS NOW

Geometric deep learning is certainly just the beginning of a paradigm shift in the world of AI-powered engineering and simulation. “Geometric deep learning is likely going to be part of the standard AI-powered engineering process in five years for most companies,” Kocer said.

With the help of geometric deep learning, data-driven AI-powered engineering workflows are set to help companies innovate like never before by improving system and product performance, streamlining communication, breaking down silos, and ultimately, empowering them to innovate faster than their competition.

“Geometric deep learning is exciting to me because it’s truly augmenting engineers’ toolkits and speeding up design exploration in a way that was never before possible,” Kocer said. “It’s definitely going to reduce development time, reduce cost, and increase sustainability and safety for products—to say nothing of faster time to market and increased competitiveness for companies.”

To learn more about Altair’s AI-powered engineering capabilities, visit https://altair.com/ai-powered-engineering.

Apply to the Most Innovative Companies Awards and be recognized as an organization driving the world forward through innovation. Final deadline: Friday, October 4.