Latticeflow Raises $2.8M to Build Trustworthy AI Systems | ETH Spin-off

LatticeFlow, an artificial intelligence company originating from ETH Zurich in 2020, has announced the completion of a $2.8 million seed funding round. This investment was spearheaded by Swiss deep-tech investment firm btov and Global Founders Capital, who have previously invested in companies such as Revolut, Slack, and Zalando.

The core concept driving LatticeFlow is the creation of resources to assist AI development teams in constructing and implementing AI models that are secure, dependable, and trustworthy. The company’s team believes that current models often excel at identifying statistical correlations to achieve specific performance targets. However, this can lead to a lack of adaptability, as these models are primarily optimized for precision within controlled laboratory environments, rather than for consistent performance in real-world applications.

“The standard approach to evaluating machine learning models often relies on overall metrics, such as accuracy. However, this provides a very generalized view of a model’s true capabilities,” explained Pavol Bielik, the company’s Chief Technology Officer. “Our aim is to offer systematic methods for monitoring models, evaluating their reliability across diverse and relevant data segments, and subsequently providing tools to enhance these models.”

Image Credits: LatticeFlow

Developing these more versatile models while maintaining robust performance necessitates a new suite of tools, alongside a team possessing extensive expertise in these areas. The founding team demonstrably possesses the necessary background. Alongside CTO Bielik, the team includes Petar Tsankov, the company’s CEO and a former senior researcher and lecturer at ETH Zurich, as well as ETH professors Martin Vechev, who directs the Secure, Reliable and Intelligence Systems lab at ETH, and Andreas Krause, who leads ETH’s Learning & Adaptive Systems lab.

It is also noteworthy that Vechev, who previously co-founded ETH spin-off ChainSecurity, and his research group at ETH previously created ERAN, a verification system for extensive deep learning models containing millions of parameters. ERAN was awarded first place in last year’s competition for certifying deep neural networks. Vechev indicated that this victory reinforced the team’s confidence in their chosen direction, although they were already considering establishing a company.

Image Credits: LatticeFlow

“Our objective is to address the primary challenge in AI: making it practical and readily usable. This is the overarching ambition,” Vechev stated. “[…] I don’t believe a company can be solely founded on certification efforts. The collective skills within the company, my group, and Andreas [Krause]’s group are highly complementary and encompass a broad spectrum, which I believe is exceptionally unique. I am unaware of other companies that possess this range of skills addressing these critical areas and having previously achieved significant breakthroughs.”

LatticeFlow currently has several pilot customers evaluating its tools. These include Swiss Federal Railways (SBB), utilizing the technology to develop a system for automated rail inspections, Germany’s Federal Cyber Security Bureau, and the U.S. Army. The team is also collaborating with other major organizations to refine their computer vision models.

“Machine Learning (ML) is a central focus at SBB, as we recognize its substantial potential for enhancing the intelligence and automation of our railway infrastructure monitoring,” commented Dr. Ilir Fetai and Andre Roger, leaders of SBB’s AI team. “The project involving robust and reliable AI with LatticeFlow, ETH, and Siemens plays a vital role in enabling us to fully leverage the benefits of ML.”

Currently, LatticeFlow is available in early access. The company intends to utilize the new funding to expedite product development and onboard additional customers. The team also plans to establish a presence in the United States in the coming months.