Researchers move closer to completely optical artificial neural network

July 19, 2018, Optical Society of America
Researchers have shown a neural network can be trained using an optical circuit (blue rectangle in the illustration). In the full network there would be several of these linked together. The laser inputs (green) encode information that is carried through the chip by optical waveguides (black). The chip performs operations crucial to the artificial neural network using tunable beam splitters, which are represented by the curved sections in the waveguides. These sections couple two adjacent waveguides together and are tuned by adjusting the settings of optical phase shifters (red and blue glowing objects), which act like 'knobs' that can be adjusted during training to perform a given task. Credit: Tyler W. Hughes, Stanford University

Researchers have shown that it is possible to train artificial neural networks directly on an optical chip. The significant breakthrough demonstrates that an optical circuit can perform a critical function of an electronics-based artificial neural network and could lead to less expensive, faster and more energy efficient ways to perform complex tasks such as speech or image recognition.

"Using an optical chip to perform neural computations more efficiently than is possible with digital computers could allow more complex problems to be solved," said research team leader Shanhui Fan of Stanford University. "This would enhance the capability of artificial neural networks to perform tasks required for self-driving cars or to formulate an appropriate response to a spoken question, for example. It could also improve our lives in ways we can't imagine now."

An artificial neural network is a type of artificial intelligence that uses connected units to process information in a manner similar to the way the brain processes information. Using these networks to perform a complex task, for instance voice recognition, requires the critical step of training the algorithms to categorize inputs, such as different words.

Although optical artificial neural networks were recently demonstrated experimentally, the training step was performed using a model on a traditional digital computer and the final settings were then imported into the optical circuit. In Optica, The Optical Society's journal for high impact research, Stanford University researchers report a method for training these networks directly in the device by implementing an optical analogue of the 'backpropagation' algorithm, which is the standard way to train conventional neural networks.

"Using a physical device rather than a computer model for training makes the process more accurate," said Tyler W. Hughes, first author of the paper. "Also, because the training step is a very computationally expensive part of the implementation of the neural network, performing this step optically is key to improving the computational efficiency, speed and power consumption of artificial networks."

A light-based network

Although neural network processing is typically performed using a traditional computer, there are significant efforts to design hardware optimized specifically for neural network computing. Optics-based devices are of great interest because they can perform computations in parallel while using less energy than electronic devices.

In the new work, the researchers overcame a significant challenge to implementing an all-optical neural network by designing an optical chip that replicates the way that conventional computers train neural networks.

Credit: Optical Society of America

An can be thought of as a black box with a number of knobs. During the training step, these knobs are each turned a little and then the system is tested to see if the performance of the algorithms improved.

"Our method not only helps predict which direction to turn the knobs but also how much you should turn each knob to get you closer to the desired performance," said Hughes. "Our approach speeds up training significantly, especially for large networks, because we get information about each knob in parallel."

On-chip training

The new training protocol operates on optical circuits with tunable beam splitters that are adjusted by changing the settings of optical phase shifters. Laser beams encoding information to be processed are fired into the optical circuit and carried by optical waveguides through the beam splitters, which are adjusted like knobs to train the .

In the new training protocol, the laser is first fed through the . Upon exiting the device, the difference from the expected outcome is calculated. This information is then used to generate a new light signal, which is sent back through the optical network in the opposite direction. By measuring the optical intensity around each beam splitter during this process, the researchers showed how to detect, in parallel, how the neural network performance will change with respect to each beam splitter's setting. The phase shifter settings can be changed based on this information, and the process may be repeated until the neural network produces the desired outcome.

The researchers tested their training technique with optical simulations by teaching an algorithm to perform complicated functions, such as picking out complex features within a set of points. They found that the optical implementation performed similarly to a conventional computer.

"Our work demonstrates that you can use the laws of physics to implement computer science algorithms," said Fan. "By these networks in the optical domain, it shows that optical neural network systems could be built to carry out certain functionalities using optics alone."

The researchers plan to further optimize the system and want to use it to implement a practical application of a neural network task. The general approach they designed could be used with various architectures and for other applications such as reconfigurable optics.

Explore further: New AI method increases the power of artificial neural networks

More information: T. W. Hughes, M. Minkov, Y. Shi, S. Fan, "Training of photonic neural networks through in situ backpropagation and gradient measurement," Optica, Volume 5,Issue , pages 864-871 (2018) DOI: 10.1364/OPTICA.5.000864

Related Stories

Training artificial intelligence with artificial X-rays

July 6, 2018

Artificial intelligence (AI) holds real potential for improving both the speed and accuracy of medical diagnostics. But before clinicians can harness the power of AI to identify conditions in images such as X-rays, they have ...

Recommended for you

Coffee-based colloids for direct solar absorption

March 22, 2019

Solar energy is one of the most promising resources to help reduce fossil fuel consumption and mitigate greenhouse gas emissions to power a sustainable future. Devices presently in use to convert solar energy into thermal ...

Physicists reveal why matter dominates universe

March 21, 2019

Physicists in the College of Arts and Sciences at Syracuse University have confirmed that matter and antimatter decay differently for elementary particles containing charmed quarks.

ATLAS experiment observes light scattering off light

March 20, 2019

Light-by-light scattering is a very rare phenomenon in which two photons interact, producing another pair of photons. This process was among the earliest predictions of quantum electrodynamics (QED), the quantum theory of ...

How heavy elements come about in the universe

March 19, 2019

Heavy elements are produced during stellar explosion or on the surfaces of neutron stars through the capture of hydrogen nuclei (protons). This occurs at extremely high temperatures, but at relatively low energies. An international ...


Adjust slider to filter visible comments by rank

Display comments: newest first

Spaced out Engineer
2.3 / 5 (3) Jul 19, 2018
To bad the brain is not all associative and some things are best left as a geometric homeostasis of a configuration space.
Unless they have low luminescent determinism for nonlinear self interaction as well, it's tough to say this solution is as substrate indepedant as cognition.
This may fulfill some of the tribes of master algorithms for an ideal enough structure to formulate, but it still looses to solutions whose basis is formulated on intractable approximations best left as a faith of reference.
This is not a defeatism. Even pre-transaction, it remains equal.
5 / 5 (2) Jul 20, 2018
#SoE: What's that in Basic English, please ?
4.2 / 5 (5) Jul 20, 2018
He's just throwing random sciency sounding words per usual.

There's a couple people on here that have the illusion that it makes them sound smart.
2.7 / 5 (7) Jul 20, 2018
This learning and on chip algorithms will have to be permanently etched as a permanent memory in case it gets absent minded
Spaced out Engineer
3 / 5 (2) Jul 20, 2018
In English. Diaphonte equations have a solution in 4 dimensions, but it is non-algorthimic. It is a geometric. The cursed curve.
Even Nature takes a guess sometimes with the homeostatic model. She unfolds and takes a guess at efficient metrics to use. She bootstraps the environment for some agency, is subtle with her hording of information. It is risky to be alive.
Cognition is known to not be entirely synchronous. Deterministic nonlinear functions underlie the chorus of voices that swarm in the brain. Its not necessarily something non-optimal. It is sufficient to not be epileptic .
The holomorphic account may not need to compute, but ought be thankful it can.
Objectivity is about the expansion of the trivial to the nontrivial, equivalence. When all is said and done it will be shared, else risk being misunderstood. Both empty and story teller need not worry.
Spaced out Engineer
1 / 5 (1) Jul 20, 2018
Some tribes, like genetic algortihms, might be improved by machines. But do you really improve an autoencoder? It just reduces the noise for some arrangement. Even in animals it takes while to train, but its not like the information is valuable, but the shared action is. The connectionalistic encoding is esoteric, but in some sense, spiritual in its self-embedded embodied updated upon revisited configuration. Walk with me.
Bayes models have to use the information they are given, but also their own phenomenalism of experiences. There are means to cheat this, but too immersive is psychotic, to real, depressive. Truth conditional, may error correct if there are priors of the uninformative, but semantic content is contextually based, if it is.
Ontology maybe conditional on the operation at hand. Function, and thus telios certainly are. Modality need not judge, just share.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.