Scientists develop next-generation reservoir computing
A relatively new type of computing that mimics the workings of the human brain has already transformed the way scientists approach some of the most difficult information processing problems.
Now, researchers have found a way to make things called “Reservoir computing.” It runs 33 to 1 million times faster, drastically reducing computing resources. Data entry is compulsory.
In fact, in a test of this next-generation reservoir computation, researchers solved complex computer problems in less than a second. Desktop computer..
Daniel Gautier, senior research author at Ohio State University and professor of physics, needs a supercomputer to solve the same problem using today’s advanced technology, and it takes a lot longer. He said it would take time.
âYou can perform very complex information processing tasks in a fraction of the time, using much less computing resources than what reservoir computing can do today,â explains Gautier.
“And the tank calculation has already made significant improvements to what was previously possible.”
The study was published in the journal today Nature Communication..
Reservoir Calculus is a machine learning algorithm developed in the early 2000s that is used to solve “the toughest” computer problems, such as predicting the evolution of dynamic systems that change over time. ..
Dynamic systems such as the weather are difficult to predict because a single small change in a condition can have a significant impact in the future, he said.
One of the famous examples is “Butterfly Effect In a metaphorical example, the changes created by the flapping of butterflies can ultimately affect the weather weeks later.”
Previous studies have shown that calculating reservoirs is ideal for learning dynamic systems and can provide accurate predictions of future behavior, according to Gautier.
It does this using a network of artificial neurons, like the human brain. Scientists send data over a dynamic network to a “reservoir” of artificial neurons randomly connected in the network. The networks generate useful results that scientists can interpret and feed back into the network, building increasingly precise predictions of how the system will evolve in the future.
The bigger and more complex the system, and the more accurate scientists want to make predictions, the bigger the system. Communication network It requires a number of artificial neurons and more computer resources and time to accomplish the task.
According to Gautier, the store of artificial neurons is a âblack box,â and the problem was that scientists weren’t sure exactly what was going on inside, only that it was working. ..
Gautier explained that the artificial neural networks at the heart of reservoir computation are built on mathematics.
âWe looked at these networks and asked the mathematicians, ‘How much do you really need from all these parts of the machine? ” “, did he declare.
In this study, Gauthier and his colleagues looked into the issue and found that it could dramatically simplify the entire reservoir IT system, drastically reducing the need for IT resources and saving considerable time. ..
They tested their concept in a forecasting task that included a weather system developed by Edward Lorenz. Butterfly Effect..
Their next-generation reservoir calculation clearly surpassed current advanced technology in this Lorentz forecasting task. In one of the relatively simple simulations performed on a desktop computer, the new system was 33 to 163 times faster than the current model.
However, the calculation of the new generation reservoirs was about a million times faster, with the aim of improving the accuracy of the forecasts. Gautier also said the new generation of calculations achieved the same precision, equivalent to just 28 neurons, compared to the 4,000 required for the current generation of models.
One big reason for speeding up is that the âbrainâ behind this next-gen tank calculation requires much less warm-up and training than current generations to produce similar results. This is what you need.
Warm-up is training data that must be inputted to the tank computer in order to prepare for the actual task.
âThe calculation of new generation reservoirs requires very little warm-up time,â explains Gautier.
âCurrently, scientists have to enter 1,000 or 10,000 or more data points to warm up. Not all lost data is necessary for the actual job. One or two or three data points. All you have to do is grab it. ” He said.
Plus, when researchers are ready to train tank computers to make predictions, next-generation systems will require significantly less data.
By testing Lorenz’s prediction task, the researchers were able to use 400 data points to achieve the same results as the current generation generated using 5,000. Data point Plus depending on the precision required.
âThe excitement is that this new generation tank calculation will adopt what was already very good and make it much more efficient,â explains Gautier.
He and his colleagues plan to expand this work to tackle even more difficult computational problems, such as hydrodynamic prediction.
âIt’s a very difficult problem to solve. We want to see if we can use a simplified model to speed up the process of solving this problem. Reservoir calculation .. “
The co-author of this study was Eric Bolt, professor of electrical and computer engineering at Clarkson University. He majored in physics at Aaron Griffith Ohio State University with a doctorate. Wendson Barbosa, postdoctoral researcher in physics at Ohio State University.
Reservoir calculation system for classification and forecasting of temporal data
New generation reservoir computing, Nature Communication (2021). DOI: 10.1038 / s41467-021-25801-2
Ohio State University
Quote: Scientists from https: //phys.org/news/2021-09-scientists-reservoir.html September 21, 2021 Next Generation Reservoir Computing (September 21, 2021) Expand
This document is subject to copyright. No part may be reproduced without written permission, except for fair dealing for the purposes of personal investigation or research. The content is provided for informational purposes only.