Zurich-based LatticeFlow, a startup that claims to be the only artificial intelligence (AI) platform that can automatically find and fix AI data and model errors, announced on Friday, October 28, that it has raised $12M (approximately €12M) in a Series A round of funding. With this round, the startup has now raised a total of $14.8M in funding to date.
LatticeFlow says the funds will help expand the capabilities of its platform and respond to growing customer demand as more companies deploy computer vision models at scale.
Investors in this round
The round was led by Atlantic Bridge and OpenOcean, with participation from FPV Ventures and existing investors btov Partners and Global Founders Capital.
Ekaterina Almasque, General Partner of OpenOcean says, “There’s a major bottleneck in bringing AI models built in the lab into production. Despite the exponential growth of AI models, operationalising them is extremely hard. LatticeFlow is addressing this in a unique way with its unstructured data quality analysis and improvement. We’re excited to join the LatticeFlow journey to build a leading automation platform for computer vision deployments, thus accelerating the roll-out of AI.”
Automating fixes to data and model issues
In recent years, computer vision AI models have outperformed humans in several tasks in the lab, including picture categorisation and detection. However, real-world events are much more complicated and variable than lab training datasets, and models frequently do not perform as intended when put into use in production. As a result, 90 per cent of all models are never produced, costing companies billions of dollars.
Founded in 2020 by Petar Tsankov, Pavol Bielik, Martin Vechev, and Andreas Krause, LatticeFlow was built to automate the process of solving data quality and blind spot issues in computer vision AI models, critical to enabling model performance in the wild.
Tsankov says, “LatticeFlow is an enabling technology that empowers engineers and companies to deliver quality data and performant computer vision models that work in the real world. As data and models grow, delivering AI models that work in the wild becomes an unwinnable battle, so we built the first smart platform that empowers engineers to accomplish this task, addressing a major pain point.”
LatticeFlow claims it is “unique” in its ability to automatically identify and address issues with data quality at scale across datasets of millions of photos, including labelling mistakes, poor-quality samples, data biases, and others.
The platform also automates the identification of often impossible-to-identify blind spot scenarios and repairs them before real-world performance is harmed. LatticeFlow has created a novel, scalable technique for focused data augmentation in order to patch the model.
The company’s customers include a number of Fortune 500 companies such as Siemens Mobility, and AI scaleups such as Intenseye, Voxel AI, and Carscan.
01
From port to startup fort: How Lars Crama is ‘Making it Happen’ in Rotterdam