ENOT Debuts Technology to Optimize Deep Neural Networks for Speed, Compression, and Energy Efficiency

ENOT says its framework for a wide range of neural network architectures and hardware can speed robots' time to market.

Getty Images

ENOT says its universal framework for a wide range of neural network architectures can deliver fast results.
ENOT.ai said its framework optimizes neural network architectures for accuracy and efficiency. The company said its framework, which has a Python API, can help robotics developers.

ENOT.ai, which stands for “Embedded Neural Network Optimization Technology,” today released its optimization technology for developers of artificial intelligence and edge AI. The Riga, Latvia-based company said its framework can make deep neural networks faster, smaller, and more energy-efficient.

“Today ... neural networks are widely used in production and applications,” stated Sergey Aliamkin, founder and CEO of ENOT. “Neural networks should be more effective in terms of consumption of computational resources and affordable.” 

Founded in 2021, ENOT said its team includes seven engineers, among whom are several Ph.D.s in computer science. They are also three-time winners of the global Low-Power Image-Recognition Challenge (LPIRC, now the Low-Power Computer Vision Challenge), outperforming teams from MIT, Qualcomm, Amazon, and others. ENOT has raised unspecified seed funding from New Nordic Venture.

ENOT intends to cut computing costs

ENOT.ai said its framework takes a trained neural network as input, after which it selects the sub-network with the lowest latency to ensure no accuracy degradation.

The company said its neural architecture selection technology can achieve optimization ratios resulting in acceleration up to 20x and compression up to 25x. This can outperform other methods and reduce total hardware computing resource costs by as much as 70%, it claimed. 

ENOT explained that its framework has a Python application programming interface (API) that users can quickly and easily integrate within various neural network training pipelines to accelerate and compress neural networks.

With ENOT's system, users can automate the search for the optimal neural network architecture, taking into account latency, RAM, and model size constraints for different hardware and software platforms, said the company. The company said its search engine allows users to automatically find the best architecture from millions of available options. It takes into account parameters such as:

  • Input resolution
  • Depth of neural network
  • Operation type
  • Activation type
  • Number of neurons at each layer
  • Bit width for target hardware platform for neural network inference

ENOT announcement

ENOT says its platform enables developers to choose the best neural network configuration for different applications. Credit: Getty Images

Accelerating robots' time to market

By helping customers improve efficiency, save costs, and launch products faster, its technology can reduce time to market, said ENOT.

“Today, robots are advancing at a very high rate, taking over more and more jobs and functions for humans, and advanced technologies like ENOT.ai's neural network compression and acceleration framework will be high in demand,” Aliamkin told Robotics 24/7. “Not only will ENOT's technology help make smart autonomous robots faster and increase battery life; it [also] will be easy to integrate and affordable, revolutionizing robotics for all.”

In addition to robotics developers, the company is aiming its framework at companies that use neural networks on edge devices, such as:

ENOT completes pilot projects

ENOT.ai claimed that it has successfully completed more than 20 pilot projects for several global technology companies, as well as a dozen midsized OEMs and AI companies. Its customers include LG, Huawei, Dscribe, Hive.aero, and PicsArt.

ENOT said it delivered 13.3x neural-network acceleration to a Top 3 smartphone manufacturer as part of the image-enhancement process. The optimization reduced the neural network depth from 16 to 11 and reduced the input resolution from 224x224 pixels to 96x96 pixels, “yet there was practically zero loss of accuracy,” the company said.

Another project with the same manufacturer delivered 5.1x acceleration for a photo de-noising neural network, even though the network had already been manually optimized, said ENOT. For end users, this translates to faster processing and a significantly extended battery life, it said.

“ENOT is at the forefront of next-level AI optimization, helping bring fast, real-time levels of AI advancement through high-throughput data into reality as science fact,” said Aliamkin. “Our journey has only just begun with examples such as the Weedbot laser weeding machine that gained a 2.7 times [improvement], thanks to the ENOT framework.”

A short introduction about WeedBot and what we have accomplished in 18 months.

Email Sign Up

Get news, papers, media and research delivered
Stay up-to-date with news and resources you need to do your job. Research industry trends, compare companies and get market intelligence every week with Robotics 24/7. Subscribe to our robotics user email newsletter and we'll keep you informed and up-to-date.

Getty Images

ENOT says its universal framework for a wide range of neural network architectures can deliver fast results.

Robot Technologies