2024-07-14 20:46:11
The Max Planck Institute has developed a new optical system, offering a simpler and more energy-efficient alternative to current methods.
This system uses light transmission to perform calculations, reducing the complexity and energy requirements of traditional neural networks.
Optical neural networks
Scientists propose a new way to implement a neural network with an optical system that can be created
” data-gt-translate-attributes = “({” attribute=”” tabindex=”0″ role=”link”> Engine Learning more sustainable in the future. The researchers at the Max Planck Institute for the Science of Light published their new method i
Challenges in neuromorphic computing
Optics and photonics are particularly promising platforms for neuromorphic computing because energy consumption can be minimized. Calculations can be performed simultaneously at very high speeds limited only by the speed of light. However, until now, there have been two significant challenges: Firstly, to achieve the complex mathematical calculations required, high laser powers are needed. Second, the lack of a general effective training method for such physical neural networks.
Both challenges can be overcome with the new method proposed by Clara Vanjura and Florian Marquardt from the Max Planck Institute for Light Science in their new paper in The physics of nature.
Simplifying the training of neural networks
“Normally, the data input is embedded in the light field. However, in our new methods we propose to embed the input by changing the light transmission,” explains Florian Marquardt, director of the institute.
In this way, the input signal can be processed arbitrarily. This is true even though the light field itself behaves in the simplest way where waves interfere without otherwise interfering with each other. Therefore, complex physical interactions can be avoided to achieve the necessary mathematical functions that would otherwise require high power light fields.
This very simple physical neural network will then be evaluated and trained: “It will really be as simple as sending light through the system and observing the transmitted light. This allows us to evaluate the output of the network. At the same time , it allows us to measure all the information related to the training,” says Clara Vanjura, first author of the study.
The authors showed in simulations that their approach can be used to perform similar image classification tasks
” data-gt-translate-attributes = “({” attribute=”” tabindex=”0″ role=”link”> As digital neural networks.
In the future, the authors plan to collaborate with experimental groups to investigate the application of their method. Because their proposal significantly relaxes the experimental requirements, it can be applied to physically very different systems. This opens up new possibilities for neuromorphic devices that enable physical training across a wide range of platforms.
#Neural #Networks #Light #Photon #Power #Drives #Revolution