Deep neural networks have already proved to be the most promising technology in the field of cognitive task automation: image processing, detection and recognition of objects in video, speech recognition and synthesis, etc. 

The availability of training tools, as well as the availability of "semi-finished products" - pre-trained neural networks - in the public domain have allowed companies around the world to create working AI application solutions on their own or with the help of contractors. 

Unfortunately, the cost of owning software based on neural networks includes the need to purchase expensive computing equipment, which often significantly reduces the investment attractiveness of such projects.

One of Cyber Data Control's priorities in recent years has been research and development in the field of neural network optimization. 

Unique expertise and technologies in the field of neural network optimization have been implemented as part of a specialized platform that allows us to solve such problems as significant reduction of AI project budgets, multiple reduction in the cost of ownership of AI services, embedding cognitive technologies in infrastructure with weak communication channels, launching a neural network on peripheral devices.

 unique solution in its segment:

  • High degree of optimization of neural network models with minimal loss of accuracy - up to 20 times in memory and up to 50 times in speed. 
  • No special knowledge and skills required - one button solution. Minimum time to obtain the optimized model - no more than 2 weeks.
  •  Optimization for any hardware platform. 
  • Available for use in both the cloud and on-premises.

The tool is extremely easy to use

Output = fully ready-to-use result - optimized neural network corresponding to the specified hardware limitations, with minimum possible loss of quality of work

The effect of implementation

Expected performance acceleration without loss of quality for any available neural network solutions is from 2 to 20 times

Pedestrian Detection and Car Acceleration 24x
Homer recognition 6x acceleration

Segmentation of image acceleration by 9 times

Vote Acceleration Identification by 25 times
Sots-dem definition of 12 times the speed characteristic
Event detection by sound acceleration 14 times
Utilizamos cookies
Preferencias de las cookies
A continuación, encontrará información sobre los propósitos para los cuales nosotros y nuestros socios utilizamos cookies y procesamos datos. Puede establecer sus preferencias para el procesamiento y/o consultar más información en los sitios web de nuestros socios.
Cookies analíticas Desactivar todo
Cookies funcionales
Otras cookies
Utilizamos cookies para personalizar el contenido y los anuncios, proporcionar las funciones que utilizan las redes sociales y analizar nuestro tráfico. Obtener más información sobre nuestra política de cookies.
Cambiar las preferencias Aceptar todo