Start-up of NASA Administrator speeds up an applied usage of neural networks

KnuPathIntellisis company, founded ten years ago by NASA ex-administrator Daniel Goldin, came out of hiding for the first time with new name KnuEdge and two products — KnuVerse software for recognizing voice and authentication and KnuPath CPU for neural computation.

KnuEdge technologogy is affirmed to be tested in military landscape, where it allowed identifying of individuals by several words, which had been told in microphone – on any language and in real conditions with background noise.

“We don’t care what type of microphone you have – cheap or expensive. Technology works at any computing platform, and we are coming nearer to level of errors, which are manifested in units per billion”, – said Goldin.

As opposed to majority of neural networks, which operate with dense matrixes, KnuEdge works with more effective dispersed matrixes.

Jim McGregor, Chief analyst of Tirias Research, thinks that low-voltage chips KnuPath signify renunciation of CPU paradigm and moving to new computation forms. They have 256 working independently cores and 16 bidirectional interfaces of input-output. They support scaling up to 512 thousand devices and ensure only 400 ns delay between racks. KnuPath can work independently or be integrated with different devices. Potential applications include signals processing and machine learning for IoT.

At the second part of the current year, KnuEdge expects to start production of platforms and systems, which are ready for commercial use. Also it plans to offer APIs based on freemium model to all persons concerned.

Now KnuEdge has total volume of private investments of $100 million and annual revenue of $20 million.