Hello, World! Please be introduced to Cogniac.
Cogniac is a young Silicon Valley startup with big plans to improve the world by making it insanely easy to automate visual observation. We’ll talk more about Cogniac the company in the future but today I would like to tell you about our precocious product, the Cogniac system.
The Cogniac system is a new type of computer: a visual computer that is programmed with visual data. It is designed to learn to observe anything and everything that people can see with their own eyes. But unlike other computers, Cogniac is not programmed with code, but rather it is programmed by its users with images and video. We call it ‘programming with data’. You’ll have to see it to believe it…
The Cogniac system enables anyone to easily extract actionable information from any source of visual data. Cogniac does things like detecting, classifying, measuring, or counting items, conditions, or actions of interest. The key is that the user can define the items, conditions, or actions in which they are interested. Then with the user’s help Cogniac learns to detect, classify, measure, or count them. To pull this off we artfully combine the latest artificial intelligence research with a system optimized for human-computer interaction in order to automate visual observation tasks using nothing more than a mobile device or web browser.
The fundamental ‘deep learning’ research that underpins Cogniac has accelerated so rapidly in recent years that some have called it the big bang of artificial intelligence. The results from the application of these techniques have been quite stunning. One variant known as deep convolutional networks now exceeds human accuracy levels on many image classification tasks while operating at much faster speeds. But for all the promise and excitement, there is a formidable barrier to applying this very powerful but also very raw technology to solve real-world problems. Successful research workflows do not translate well to commercial settings due to a number of problems including tedious data curation and preprocessing requirements, low-level tensor frameworks requiring rare expertise and extensive configuration, and of course the need to continuously refine and deploy updated prediction models as both requirements and the world-at-large invariably evolve.
Cogniac solves these problems with an elegant system built around both deep learning and human computer interaction technologies that automates all aspects of data curation, model creation, model deployment, and ongoing model search and optimization. Cogniac uses artificial intelligence to augment human visual intelligence and recursively uses human visual intelligence to improve the artificial intelligence; it is quite the machine — well, actually, a fine synthesis of man and machine!
Users are able to easily train the Cogniac system via our web and mobile applications by providing initial example images of the subjects in which they are interested. Typically a few dozen labeled and a few hundred unlabeled images are all that is required to get started. Within minutes Cogniac will discover and train state-of-the-art deep convolutional network models that perform well on the user’s custom task. Cogniac is designed to be the shortest path between visual data and actionable predictions, producing great prediction models in hours instead of days or weeks. Cogniac will periodically ask users for feedback or confirmation on a small subset of its predictions when required to enhance the system performance, and the system is constantly working behind the scenes to optimize and improve the customized models.
When it’s time to integrate Cogniac with existing systems and workflows, a comprehensive public API exposes the full system functionality. We even have an on-premises gateway option for applications with very high throughput video or image processing requirements behind a firewall.
Today we are happy to announce that we are opening a preview of Cogniac to the public so that developers, product managers, and other technologists can see for themselves the future of automated visual observation. The preview includes a subset of the full system application types, but additional application types will be added soon as we fully document them. A short video tutorial is also available. Check it out!
We’ll be telling you more about Cogniac (which incidentally is named after the legendary ENIAC, pictured above) a lot more in the coming weeks and months, as well as sharing some of our customer success stories. Cogniac is only a year old, but we have already seen quite a lot! Please get in touch at contact@cogniac.co if you want to know more.
