It’s unusual to see the same headline used to highlight research from two different teams released in such proximity, February 2024 and July 2024, respectively. Both of these are neuromorphic (brainlike) computing stories.
February 2024: Neural networks made of light
The first team’s work is announced in a February 21, 2024 Friedrich Schiller University press release, Note: A link has been removed,
Researchers from the Leibniz Institute of Photonic Technology (Leibniz IPHT) and the Friedrich Schiller University in Jena, along with an international team, have developed a new technology that could significantly reduce the high energy demands of future AI systems. This innovation utilizes light for neuronal computing, inspired by the neural networks of the human brain. It promises not only more efficient data processing but also speeds many times faster than current methods, all while consuming considerably less energy. Published in the prestigious journal „Advanced Science,“ their work introduces new avenues for environmentally friendly AI applications, as well as advancements in computerless diagnostics and intelligent microscopy.
Artificial intelligence (AI) is pivotal in advancing biotechnology and medical procedures, ranging from cancer diagnostics to the creation of new antibiotics. However, the ecological footprint of large-scale AI systems is substantial. For instance, training extensive language models like ChatGPT-3 requires several gigawatt-hours of energy—enough to power an average nuclear power plant at full capacity for several hours.
Prof. Mario Chemnitz, new Junior Professor of Intelligent Photonic SystemsExternal link at Friedrich Schiller University Jena, and Dr Bennet Fischer from Leibniz IPHT in Jena, in collaboration with their international team, have devised an innovative method to develop potentially energy-efficient computing systems that forego the need for extensive electronic infrastructure. They harness the unique interactions of light waves within optical fibers to forge an advanced artificial learning system.
A single fiber instead of thousands of components
Unlike traditional systems that rely on computer chips containing thousands of electronic components, their system uses a single optical fiber. This fiber is capable of performing the tasks of various neural networks—at the speed of light. “We utilize a single optical fiber to mimic the computational power of numerous neural networks,“ Mario Chemnitz, who is also leader of the “Smart Photonics“ junior research group at Leibniz IPHT, explains. “By leveraging the unique physical properties of light, this system will enable the rapid and efficient processing of vast amounts of data in the future.“
Delving into the mechanics reveals how information transmission occurs through the mixing of light frequencies: Data—whether pixel values from images or frequency components of an audio track—are encoded onto the color channels of ultrashort light pulses. These pulses carry the information through the fiber, undergoing various combinations, amplifications, or attenuations. The emergence of new color combinations at the fiber’s output enables the prediction of data types or contexts. For example, specific color channels can indicate visible objects in images or signs of illness in a voice.
A prime example of machine learning is identifying different numbers from thousands of handwritten characters. Mario Chemnitz, Bennet Fischer, and their colleagues from the Institut National de la Recherche Scientifique (INRS) in Québec utilized their technique to encode images of handwritten digits onto light signals and classify them via the optical fiber. The alteration in color composition at the fiber’s end forms a unique color spectrum—a „fingerprint“ for each digit. Following training, the system can analyze and recognize new handwriting digits with significantly reduced energy consumption.
System recognizes COVID-19 from voice samples
“In simpler terms, pixel values are converted into varying intensities of primary colors—more red or less blue, for instance,“ Mario Chemnitz details. “Within the fiber, these primary colors blend to create the full spectrum of the rainbow. The shade of our mixed purple, for example, reveals much about the data processed by our system.“
The team has also successfully applied this method in a pilot study to diagnose COVID-19 infections using voice samples, achieving a detection rate that surpasses the best digital systems to date.
“We are the first to demonstrate that such a vibrant interplay of light waves in optical fibers can directly classify complex information without any additional intelligent software,“ Mario Chemnitz states.
Since December 2023, Mario Chemnitz has held the position of Junior Professor of Intelligent Photonic Systems at Friedrich Schiller University Jena. Following his return from INRS in Canada in 2022, where he served as a postdoc, Chemnitz has been leading an international team at Leibniz IPHT in Jena. With Nexus funding support from the Carl Zeiss Foundation, their research focuses on exploring the potentials of non-linear optics. Their goal is to develop computer-free intelligent sensor systems and microscopes, as well as techniques for green computing.
Here’s a link to and a citation for the paper,
Neuromorphic Computing via Fission-based Broadband Frequency Generation by Bennet Fischer, Mario Chemnitz, Yi Zhu, Nicolas Perron, Piotr Roztocki, Benjamin MacLellan, Luigi Di Lauro, A. Aadhi, Cristina Rimoldi, Tiago H. Falk, Roberto Morandotti. Advanced Science Volume 10, Issue 35 December 15, 2023 2303835 DOI: https://doi.org/10.1002/advs.202303835. First published: 02 October 2023
This paper is open access.
July 2024: Neural networks made of light
A July 12, 2024 news item on ScienceDaily announces research from another German team,
Scientists propose a new way of implementing a neural network with an optical system which could make machine learning more sustainable in the future. The researchers at the Max Planck Institute for the Science of Light have published their new method in Nature Physics, demonstrating a method much simpler than previous approaches.
…
A July 12, 2024 Max Planck Institute for the Science of Light press release (also on EurekAlert), which originated the news item, provides more detail about their approach to neuromorphic computiing,
Machine learning and artificial intelligence are becoming increasingly widespread with applications ranging from computer vision to text generation, as demonstrated by ChatGPT. However, these complex tasks require increasingly complex neural networks; some with many billion parameters. This rapid growth of neural network size has put the technologies on an unsustainable path due to their exponentially growing energy consumption and training times. For instance, it is estimated that training GPT-3 consumed more than 1,000 MWh of energy, which amounts to the daily electrical energy consumption of a small town. This trend has created a need for faster, more energy- and cost-efficient alternatives, sparking the rapidly developing field of neuromorphic computing. The aim of this field is to replace the neural networks on our digital computers with physical neural networks. These are engineered to perform the required mathematical operations physically in a potentially faster and more energy-efficient way.
Optics and photonics are particularly promising platforms for neuromorphic computing since energy consumption can be kept to a minimum. Computations can be performed in parallel at very high speeds only limited by the speed of light. However, so far, there have been two significant challenges: Firstly, realizing the necessary complex mathematical computations requires high laser powers. Secondly, the lack of an efficient general training method for such physical neural networks.
Both challenges can be overcome with the new method proposed by Clara Wanjura and Florian Marquardt from the Max Planck Institute for the Science of Light in their new article in Nature Physics. “Normally, the data input is imprinted on the light field. However, in our new methods we propose to imprint the input by changing the light transmission,” explains Florian Marquardt, Director at the Institute. In this way, the input signal can be processed in an arbitrary fashion. This is true even though the light field itself behaves in the simplest way possible in which waves interfere without otherwise influencing each other. Therefore, their approach allows one to avoid complicated physical interactions to realize the required mathematical functions which would otherwise require high-power light fields. Evaluating and training this physical neural network would then become very straightforward: “It would really be as simple as sending light through the system and observing the transmitted light. This lets us evaluate the output of the network. At the same time, this allows one to measure all relevant information for the training”, says Clara Wanjura, the first author of the study. The authors demonstrated in simulations that their approach can be used to perform image classification tasks with the same accuracy as digital neural networks.
In the future, the authors are planning to collaborate with experimental groups to explore the implementation of their method. Since their proposal significantly relaxes the experimental requirements, it can be applied to many physically very different systems. This opens up new possibilities for neuromorphic devices allowing physical training over a broad range of platforms.
Here’s a link to and a citation for the paper,
Fully nonlinear neuromorphic computing with linear wave scattering by Clara C. Wanjura & Florian Marquardt. Nature Physics (2024) DOI: https://doi.org/10.1038/s41567-024-02534-9 Published: 09 July 2024
This paper is open access.