Russia Develops New Type of Neural Networks for Drones
Optical AI models promise faster performance with lower energy use
Researchers at Samara University have developed a new type of optical neural network architecture designed to enable autonomous and energy-efficient artificial intelligence onboard drones.
According to the university’s press service, the approach is based on optical transformers – neural network components capable of real-time image and speech recognition. Instead of conventional microchips, the system processes data using streams of photons and optical elements.
Optical neural networks can operate faster and consume less energy, but they currently lag behind digital systems in accuracy. The Samara team’s new approach to optical matrix multiplication could allow these models to match or surpass digital neural networks in precision.
Catching Up and Surpassing
A series of experiments using a numerical model of a modified optical scheme integrated into a conventional neural network confirmed the researchers’ findings.
The team now plans to develop a hardware prototype of the optical neural network suitable for deployment onboard drones.
