American researchers from the The University of California, Los Angeles, has presented the concept of an AI-based image generator, that consumes almost no energy.
Typically, AI-based image generators use a process, known as diffusion. They learn from a large set of images and repeatedly add statistical noise until the image disappears completely. After the AI receives the task of generating an image, it first recreates a screen with a lot of noise interference, and then, removing it, returns the image.
When it comes to generating a large number of images, the process becomes slow and energy-intensive. The new image generator uses a digital encoder, trained on publicly available data, sets to create a static signal. Ultimately, this signal generates an image. This requires significantly less energy.
A liquid crystal screen, known as a spatial light modulator (SLM), maps this pattern onto the laser beam. The beam, then passes through a second decoding SLM, which converts the pattern in the laser into the final image. Unlike traditional AI tools, which require millions of computer operations to calculate, in this case, light does the vast majority of the work.
“Our optical generative models are capable of synthesizing countless images with virtually no computational power, offering a scalable and energy-efficient alternative to digital AI models”, — explains the study’s lead author Shiqi Chen.
The researchers tested their system on a variety of images that used for AI training, including images of famous people, butterflies and full-color paintings Vincent van Gogh. The results were comparable to those of traditional image generators, but consumed much less energy. This breakthrough has the potential to significantly reduce the carbon footprint of AI-generated content.
This technology can also find applications in various fields. Since the system is very fast and consumes little power, it can be used, for example, to create images and videos for virtual and augmented reality displays, as well as for small devices such as smartphones or electronics such as artificial intelligence glasses.
The results of the study are published in the journal Nature
Source: TechXplore
Spelling error report
The following text will be sent to our editors: