LGN, which helps companies deploy AI at scale, raises $2M

LGN, a consultancy specializing in distributed machine learning, deep learning, and AI technologies, today announced that it raised $2 million. The company says it’ll use the investment to bolster its product development and hiring efforts as it expands its market reach. In particular, LGN intends to pursue low-latency inference technology that can process optical data on-chip orders of magnitude faster than current-gen tech.

Businesses are in the midst of a shift in where and how they analyze data and derive actionable insights from it. Spending on AI is anticipated to break the $500 billion mark by 2024, according to IDC, and Gartner forecasts that over 50% of enterprise data will be processed outside the cloud by 2022.  But AI projects remain highly susceptible to failure. According to one study, only 25% of companies have successfully developed an “enterprise-wide” AI strategy.

LGN aims to help enterprises address data science challenges by scaling out AI deployments, improving resiliency in machine learning models, and optimizing the hardware and devices that make up edge endpoints. Borne in 2018 out of a collaboration between former Apple and BMW executive Daniel Warner, Oxbridge research fellow Luke Robinson, and Vladimir Čeperić of MIT and the University of Zagreb, LGN designs solutions for customers with the goal of minimizing transfer costs and improving training dataset quality while reducing storage and processing requirements.

“We previously worked on defense projects improving laser vision systems in less-than-ideal environments with cloud and fog. We were generating extreme amounts of data at the edge and needing to constantly retrain and improve a model,” Warner told VentureBeat via email. “These moments were the early days in us seeing how the future would unfold and plan to address them. Ceperic was at MIT making the world’s first optical convolutional neural network chip. Dan and Luke were at Harwell using AI to piece back together optical beams from quantum lasers. We immediately realized the synergy around on-chip optical signal processing and realised we shared a vision around the future of AI.”

Data scientists spend the bulk of their time cleaning and organizing data. Much of the remaining time is spent on feature engineering, or the process of using domain knowledge to extract features from input data. It’s essential to tuning AI and machine learning performance, but it’s also typically arduous and involves rewriting features before they’re deployed. Often, a missing piece is infrastructure that bridges the gap between training models and serving AI results in production environments.

LGN’s solution to this is what Warner calls “networked AI.” As Warner explains, currently, much of what AI does needs to be translated so that humans can understand it and make decisions. By contrast, networked AI is “AI-to-AI” communication, which removes the need for the human element and speeds up the action-taking process.

“AI has huge implications for the way businesses operate, yet so much of the modelling is done in carefully controlled test environments. When deployed in real world situations, anomalies always occur, which disrupts lab-grown models and undermines companies’ efforts to revolutionise how they use autonomous systems effectively,” Warner said. “By scaling edge AI, optimising the endpoints collecting the data, and making models more resilient, we are radically accelerating learning and, in doing so, giving our customers a competitive advantage in a crowded marketplace.”

To this end, LGN offers a product that automatically calibrates AI-powered sensor arrays as well as a platform — Intelligent Select — that reduces data capture and compute needs by filtering data to only record and process edge cases while transferring and storing data in a compressed form. LGN claims that Intelligent Select can boost model performance on hard classes by 180% while using just 2.5% of the original training data, and moreover cut down on data, labeling, and processing expenses.

“Networked AI will change the way the world works. All the potential benefits of AI are limited while we need to involve humans in the majority of the decisions,” Warner continued. “Edge AI is the first step towards speeding up that process, by having AI at the endpoint to rapidly collect, analyse and determine action based on data. Networked AI will take that and apply it across all aspects of an organisation, freeing up workers to focus on more valued-adding activities.”

Eight-employee LGN supports five companies operating in a range of sectors including automotive, agriculture, and manufacturing. One is investor Jaguar Land Rover, which plans to retain LGN’s services to expand its data capture efforts and analyze road travel data toward the development of autonomous cars.

“Our product[s] allow companies to operate edge AI at scale in challenging, diverse, real- world environments. For example, we’re working with agriculture companies to scale out the commercial deployment of machine vision systems on farms and in industrial chicken sheds,” Warner said. “Our model optimisation solution allows enterprises to deploy AI and machine learning models on low-cost hardware. For example, we can deploy computer vision models to run object detection or semantic segmentation on high-definition camera feeds, with a full supervision and monitoring stack on a $6 AI camera chip. This keeps the cost of the bill-of-materials down, which increases margins and unlocks commercial sales.”

The funding round announced today is London-based LGN’s first public raise. Trucks Venture Capital, Luminous Ventures, and Jaguar Land Rover participated, among others.