Scientists cut AI energy consumption by 20% with new GreenOCR 2.0 model

The Russian company Smart Engines has unveiled a breakthrough in artificial intelligence: the neural network model GreenOCR 2.0, which reduces energy consumption for text recognition by 20%. The company’s press service shared this news in an interview with Gazeta.Ru. In addition to saving energy, the new development increases data processing speed and reduces errors tenfold compared to previous versions.

What is GreenOCR 2.0?

GreenOCR 2.0 is an advanced text recognition neural network that combines high performance with energy efficiency. It can run on devices without powerful graphics processors (GPUs), making it accessible for a wide range of gadgets, including smartphones and tablets.

Key features of the model:

• Energy efficiency: 20% lower energy consumption thanks to a new computational architecture.
• Speed and accuracy: faster data processing and ten times fewer errors.
• Versatility: recognition of printed and handwritten text in 103 languages, including symbols and punctuation marks.

A Technological Breakthrough

At the core of GreenOCR 2.0 lies an innovative architecture that replaces floating-point numbers with integers (int8 quantization) and combines 4.6-bit and 8-bit neural networks. This optimization allows efficient computation on mobile platforms while maintaining high accuracy. Such technology is especially crucial for devices with limited computing resources, where traditional GPU-dependent neural networks are inefficient.

“We created a model that is not only faster and more accurate, but also significantly more economical. This opens new possibilities for using AI in everyday devices,” Smart Engines emphasized.

Where is GreenOCR 2.0 Used?

The new model has already been integrated into Smart Engines’ software products, which are applied in various sectors:

• Document verification: recognition and authentication of passports, driver’s licenses, and other IDs.
• Financial operations: reading data from bank cards and QR codes.
• Data processing: automating the input of information from accounting documents, forms, and phone numbers.

The technology is particularly in demand in the banking sector, retail, and government services, where fast and accurate text processing is essential.

Why Does It Matter?

Reducing AI energy consumption is a global challenge. Modern neural networks for text or image processing require enormous computing power, driving up energy use and carbon footprints. GreenOCR 2.0 shows that it is possible to maintain high performance while minimizing strain on both devices and the environment.

Moreover, the ability to function on devices without GPUs makes AI more accessible. This is especially relevant for countries with limited infrastructure, where powerful servers or graphics processors are scarce.

What’s Next?

Smart Engines plans to expand the use of GreenOCR 2.0 by integrating it into new products and adapting it for other tasks, such as recognizing complex handwritten texts or processing images in real time. The company also intends to continue research in energy-efficient neural networks to make AI even more versatile.

In Brief…

Russian developers at Smart Engines have taken an important step in AI development with the launch of GreenOCR 2.0. The model cuts energy consumption by 20%, speeds up data processing, and reduces errors tenfold. Already in use for processing documents, QR codes, and bank cards across 103 languages, this breakthrough not only highlights the potential of Russian science but also paves the way for more accessible and eco-friendly AI solutions.