Six months after its foundation, the ETH spin-off LatticeFlow has received USD 2.8 million from two venture capital investors. The aim of this move, as the new company reveals in a press release, is to support its ambitious vision of ensuring reliable and trustworthy AI. Overcoming this strategic challenge will enable the safe use of AI technology, say the company’s founders.
While there have been significant advances in the development of artificial intelligence in the past decade, many AI models are still overwhelmed when it comes to productive, real-world use outside controlled research environments. This can result in errors and unwanted behaviours. Consider a driver assistance system that misreads a stop sign in difficult light conditions – when the sun is shining on the camera, for example – or because the stop sign is combined with other signage.
Spin-off improves AI models
“Our mission is to close this gap and enable companies to develop and use trustworthy AI,” says co-founder and CEO Petar Tsankov. To achieve this, the spin-off is developing a product that assesses the robustness and reliability of AI models, while providing actionable insights into how to improve them and ensure safety. “The first step in improving the reliability of AI models is the ability to comprehensively assess the models during development and deployment,” says co-founder and CTO Pavol Bielik.
Computer science professors Martin Vechev and Andreas Krause, also co-founders of the company, focus on researching and developing principles of trustworthy AI technologies at the ETH AI Center. The ERAN system developed by Professor Vechev and his team, for example, last year won a competition that was based on the issue of safety certification of artificial intelligence, outperforming teams from MIT, Oxford, Imperial College London, UIUC and elsewhere. “We’d like to use the spin-off to help transfer the technology developed at ETH to industry,” explains Krause, Head of the ETH AI Center.
Interest from prestigious clients
Various institutions such as the Swiss Federal Railways (SBB), the US Army and the German Federal Office for Information Security (BSI) are already working with LatticeFlow. Ilir Fetai and Andre Roger, who head up the Machine Perception competence centre at SBB, comment: “Machine learning (ML) is a central issue for SBB, as we see significant potential in its use for the improved, intelligent and automated monitoring of our rail infrastructure. This project targeting robust and reliable AI with LatticeFlow, ETH and Siemens plays an important part in successfully realising the benefits of ML.” LatticeFlow now plans to press ahead with the development of its product using the new funding it has received.