GridFlow, a startup that emerged from Zurich’s ETH in 2020, is helping machine learning teams improve their AI vision models by automatically diagnosing problems and improving both the data and the models themselves. The company announced today that it has raised a $12 million Series A funding round led by Atlantic Bridge and OpenOcean, with participation from FPV Ventures. Existing investors btov Partners and Global Founders Capital, who led the company’s $2.8 million starting round last year, also participated in this round.
As LatticeFlow co-founder and CEO Petar Tsankov told me, the company currently has more than 10 customers in both Europe and the US, including a number of major corporations such as Siemens and organizations such as the Swiss Federal Railways, and is currently piloting quite a few a few more. It’s this customer demand that led LatticeFlow to increase on this point.
“I was in the United States and met some investors in Palo Alto,” explains Tsankov. “They saw the bottleneck we have when onboarding customers. We literally had machine learning engineers supporting customers and that’s not how you should run the business. And they said, ‘Okay, take $12 million, get these people in and expand.’ That was certainly good timing, because when we talked to other investors, we saw that the market has changed.”
As Tsankov and his co-founder CTO Pavol Bielik pointed out, most companies today struggle to get their models into production and when they do, they often realize that they are not performing as well as they expected. The promise of LatticeFlow is that it can automatically diagnose the data and models to find potential blind spots. For example, in his work with a major medical company, the tools to analyze their datasets and models quickly found more than half a dozen critical blind spots in their state-of-the-art manufacturing models.
The team noted that it’s not enough to just look at the training data and make sure there is a diverse set of images — in the case of the vision models LatticeFlow specializes in — but also examine the models.
LatticeFlow founding team (left to right): Prof. Andreas Krause (scientific advisor), Dr. Petar Tsankov (CEO), Dr. Pavol Bielik (CTO) and Prof. dr. Martin Vechev (scientific advisor). Image Credits: GridFlow
“if you nothing but look Bee the data – and this is a fundamental differentiator in front of latticFlowith whom because we do not nothing but find the standard- data issues Like it label issues or arm-quality samples, but also fashion model- Blind stains, which to be the scenarios Where the models to be fail,” explained Tsankov. “Once the fashion model- is finished, we can to take the, find several data fashion model- issues and help out businesses to repair the.”
For example, he noted that models will often find hidden correlations that can confuse the model and skew the results. For example, when working with an insurance client who used an ML model to automatically detect dents, scratches, and other damage in car images, the model often labeled an image with a finger in it as a scratch. Why? Because in the training set, customers often took a close-up photo with a scratch and pointed at it with their finger. Unsurprisingly, the model then correlated “finger” with “scratch”, even if there was no scratch on the car. Those are problems, the LatticeFlow teams argue, going beyond making better labels and requiring a service that can look at both the model and the training data.
LatticeFlow reveals a data bias for training AI models for vehicle damage inspection. Because people often point to scratches, this teaches models that fingers indicate damage (a false characteristic). This issue has been resolved with a custom augmentation that removes fingers from all images. Image Credits: GridFlow
It’s worth noting that LatticeFlow itself is not in the training business. The service works with pre-trained models. For now, it is also focusing on offering its service as an on-prem tool, although it may also offer a fully managed service in the future as it uses the new funding to aggressively hire both to better serve its existing customers. as well as to expand its product portfolio.
“The painful truth is that most large-scale AI model deployments today simply don’t work reliably in the real world,” said Sunir Kapoor, operations partner at Atlantic Bridge. “This is largely due to the lack of tools to help engineers efficiently resolve critical AI data and model errors. But this is also why the Atlantic Bridge team has so unequivocally decided to invest in LatticeFlow. We believe the company is poised for massive growth as it is currently the only company to automatically diagnose and repair AI data and model defects at scale.”
