General Vision
Introducing Realistic AI

The global race for realistic AI increases the demand for fast, efficient, and low-power machines. The time, power, and expertise it requires to deploy a valid 'machine learning algorithm', limits providers from tapping into the full potential of AI technology. “Real-time and adaptive learning is the stepping stone for the adoption of AI in our everyday lives at home, work, public places, and more,” says Guy Paillet, President and CEO of General Vision. General Vision’s NeuroMem technology is a unique neural network architecture on silicon—trainable, scalable, and low-power inspired by the human brain—which is capable of learning and recalling patterns autonomously without being connected to a high-performance computer or the cloud. In addition to General Vision’s own CM1K chip featuring 1024 neurons, the famed Intel^® Curie module with 128 NeuroMem neurons has allowed developers from the corporate world to maker communities to deploy low-power, even battery operated, IoT devices with in-situ intelligence. A3rd NeuroMem chip, NM500, manufactured by Nepes, Korean leader in wafer-scale packaging is due for release in June.

Apart from NeuroMem^®, General Vision has been nurturing two derivative technologies—CogniSight and IntelliGlass: CogniSight^® is an image recognition engine, inspired by the visual cortex, taking advantage of NeuroMem networks to learn and recognize visual objects with highly adaptive and incremental trainability and contextual classifiers. It is an essential component to deploy edge data analytics and sensor fusion at very low power and with high scalability. IntelliGlass or Monolithic Image Perception Device (MIPD) is the combination of sensors and CogniSight engines into a glass substrate to produce low-cost, miniature monolithic vision modules.

General Vision deploys neuromorphic chips, low-power devices which can learn and react autonomously without being connected to a high-performance computer or to the cloud

General Vision’s NeuroMem technology can be leveraged across many applications from consumer and industrial IoT to big data analytics and network security. One of its flagship applications has been the inspection of fishes before fileting offshore. Thirty CogniSight cameras with each several hundred neurons have been installed since 2003 on fishing vessels in Norway and Iceland and trained by the fishermen themselves to sort the fishes as they come on the boat, saving 2 million dollars per boat per year. Other typical applications where NeuroMem can outperform conventional image processing tools are raw produce and surface inspection. The neurons based on a parallel, hard-wired architecture are capable of incremental learning and non-linear classification with deterministic latencies. They can be trained on-line and off-line and their response time remains constant and in the order of microseconds regardless of the number of models learned and stored in the neurons. This is essential for complex patterns as seen in wood, glass, textile, but also waste sorting and wafer inspection. Beyond factory automation, NeuroMem can be deployed for home and building automation, automotive, and more. Because the neurons are agnostic to the data source, they can learn and recognize signatures deriving from images, audio, bio-signals, internet packets, and semantic. Multiple NeuroMem networks can be easily cascaded for sensor fusion and robust decision making.

The technology is scalable and is commercially available today as a pioneer of realistic AI. With the continuous enhancement of its technology and products, General Vision continues to prove its real-time benefits across different sectors.