Breakthrough in Optical Calculations and Machine Learning
The team of scientists under the leadership of Professor ChVI Vonshik from the Center for Molecular Spectroscopy and IBS (IBS CMSD) made a breakthrough in the field of optical calculations and machine learning. They have developed a method that allows the use of light phenomena in the scattering media, such as matte glass, to transmit information. [1]
Traditionally, it was believed to be impossible to view through the scattering environment due to the “confusing” nature of the transmitted light, akin to complex encryption. However, the researchers discovered that the optical input response of a nonlinear scattering medium can be determined using a third-order tensor, unlike a linear matrix.
To demonstrate their findings, the researchers utilized an environment composed of barii titanate nanoparticles that generate nonlinear signals of the second harmonic (SHG). These SHG signals create cross conditions, activating multiple input channels and violating the principle of linear superposition.
The team successfully showcased the use of nonlinear scattering media in the optical encryption process. They also demonstrated the functionality of alloptic logical elements “and,” which are only activated when two specific input channels are simultaneously activated. This approach offers advantages over silicon logic, including reduced energy consumption and the potential for parallel processing at the speed of light.
“The costs of detaching cross-country conditions from weak nonlinear signals were significant,” stated Dr. Moon Junho, the lead author of the study.
This groundbreaking study is expected to open new horizons in the realm of optical calculations and machine learning. Professor Chuvi commented, “In the developing field of alloptic machine learning, nonlinear optical layers are key to improving model performance. We are currently exploring how our study can be integrated into this area.” [2]
Sources: | |
---|---|
[1] https://www.nature.com/sarticles/s41567-023-02163-8 | [2] Interview with Professor Chuvi |