LatticeFlow raises $12M to eliminate computer vision blind spots • TechCrunch


LaticeFlow, a startup that spun out of ETH Zurich in 2020, helps machine learning teams improve their AI vision models by automatically diagnosing problems and optimizing the data and models themselves. The company announced today that it has raised $12 million in a 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 founding $2.8 million round last year, also participated in the round.

As LatticeFlow co-founder and CEO Petar Tsankov told me, the company currently has more than 10 clients in both Europe and the US, including a number of large companies such as Siemens and organizations such as Swiss Federal Railways, and currently operates pilots with a large number of clients. . a little more. It is this customer request that has increased LatticeFlow at this point.

“I was in the United States and met some investors in Palo Alto,” Tsankov explained. “They saw the bottleneck we are having with customers on the plane. We literally had machine learning engineers for customer support and that’s not how you should run a company. And they said, “Okay, take the $12 million, get these guys in and expand.” This was definitely great timing because when we talked to other investors, we saw that the market had changed.”

As Tsankov and his co-founder CTO Pavol Bielik pointed out, most companies today have difficulty bringing their models into production, and then, when they do, often realize that they are not working as well as they expected. The promise of LatticeFlow is that it can automatically diagnose data and models to find potential blind spots. Working with a major medical company, their tools for analyzing their data sets and models quickly found more than half a dozen critical blind spots in their modern production models, for example.

The team noted that it is not enough to just look at the training data and make sure there are a variety of images – in the case of the vision models that LatticeFlow specializes in – but also to check the models.

LatticeFlow founding team (from left to right): Prof. Andreas Krause (scientific advisor), Dr Petar Tsankov (CEO), Dr. Pavol Bilik (CTO) and Professor Martin Fischev (scientific advisor). Image credits: LaticeFlow

if You are Just a look in The data – and this is he is a Basic differentiation for TheAtticeFlow Because We do not Just find Basic data Issues Such as Putting signs Issues or poor-Quality samples but also Model blind splatter, Which be The Scenarios where The Models be Tsankov explained. “Once The Model he is Ready, we Can Takes He. She, FIndia Different data Model Issues And the helps comp Repair He. She.”

He noted, for example, that models often find hidden associations that may confuse the model and distort the results. When working with an insurance client, for example, who used an ML model to automatically detect dents, dents, and other damage in car photos, the model would often label the image with a finger inside as a scratch. why? Because in the training set, clients often take a close-up photo with a scratch and point at it with their finger. It is not surprising that the model then associates the “finger” with the “scratch”, even in the absence of a scratch on the car. The LatticeFlow team argues that these are issues beyond creating better labels and need a service that can look at both the model and the training data.

LatticeFlow detects data bias for training AI models to check vehicle damage. Since people often point out scratches, this makes models learn that fingers are indicating damage (a pseudo-feature). This issue is fixed with a custom augmentation that removes fingers from all images. Image credits: LaticeFlow

It should be noted that LatticeFlow itself is not in the training field. The service works with pre-trained models. For now, it is also focusing on offering its service as a tool within the company, although in the future it may offer a fully managed service as well, as it is aggressively using new funding to staff, both to better serve its existing clients and to get building its product portfolio.

“The hard truth is that most large-scale AI model deployments today simply don’t work reliably in the real world,” said Suner Kapoor, Atlantic Bridge’s operating partner. “This is largely due to the lack of tools that efficiently help engineers solve AI data and critical model errors. But it is also the reason why the Atlantic Bridge team unequivocally came to the decision to invest in LatticeFlow. We believe the company is poised for explosive growth. It is currently the only company that automatically diagnoses and fixes AI data and model flaws at scale.”

Leave A Reply

Your email address will not be published.