Scientists Develop The Next Generation Of Reservoir Computing Technology Sensors like these, of course are not intended for anything but the very lowest-stakes research applications in oceanography and marine science. Their primary purpose is to identify where there may be other potential impacts or hazards associated with large swells by monitoring those events from afar as well: “From a spatial perspective this technique allows scientists who have been studying water currents on land at one location to monitor sea level movements around that same field,” says Dr John D’Agostino, an expert in hydrography based within Carnegie Mellon University’s Department. The next generation sensors will allow researchers to use them more efficiently across much broader areas.
A relatively new type of computer that mimics the way the human brain works is changing how scientists tackle the most difficult information processing problems.
Now, researchers have found a way to make the reservoir work 33 million times faster with fewer computer resources and fewer data input requirements.
In fact, in a test of next-generation reservoir computing, researchers solved a complex computing problem in less than a second on a desktop computer.
With current state-of-the-art technology, a supercomputer is needed to solve the same problem and it will last a long time, said Daniel Gothier, a professor of physics at The Ohio State University.
“We can perform complex information processing tasks at a specific time using fewer computer resources than computing can currently do,” Gadier said.
“And mathematics is a significant improvement over what was previously possible.”
The study was published today in the journal Nature Communications.
Gott Thier stated that reservoir computing is a machine learning tool developed in the early 2000s and used to solve “hard to difficult” computer problems, such as predicting the evolution of dynamic systems over time. . .
That said, a small change in position can have a big impact on the line because dynamic systems like weather are difficult to predict.
A popular example is the “butterfly effect”, in which – a descriptive metaphor – the changes that occur when a butterfly enters its wings affect the time of week.
Previous research has shown that computing is relevant to the study of dynamical systems and can provide accurate predictions about how they will behave in the future, Gout Theor said.
This is done using artificial neural networks similar to the human brain. Scientists present data as a “reservoir” in a network of artificial neurons that are connected to an almost dynamic network. Network scientists create useful outputs that explain and return the network, thereby making more accurate predictions of how the system will change in the future.
The larger and more complex the system and the more accurate scientists want to make predictions, the larger the network of artificial neurons and the more computer resources and time are required to accomplish the task.
One problem is that the stockpile of artificial neurons is a “black box” and scientists don’t know what’s going on inside it—they know it works, says Gautam Theor.
Gout Theor explains that artificial neural networks are built on mathematics at the heart of reservoir computing.
“Given mathematicians in these networks, ‘How far do all the pieces of this machine actually go?'” he said.
In this study, Khadi and his colleagues explored that question and found that the entire reservoir system could be done very simply, significantly reducing the need for resource calculations and saving a lot of time. . .
They tested their hypothesis in forecast work involving meteorological systems developed by Edward Lawrence, whose work led to our understanding of the butterfly effect.
His next generation of reservoir computing is a clear winner in Lawrence’s forecast of the modern state. In relatively simple simulations on a desktop computer, the new system is 33 to 163 times faster than the current model.
But when targeted for good forecasting accuracy, the next generation of reservoir systems will be about 1 million times faster. Gothier said the new generation computer achieves an accuracy equivalent to 28 neurons, compared to the 4,000 needed for the current generation model.
A key factor in increasing the speed is that the “brain” behind the next generation of reservoir computing requires less warming up and training than the current generation to produce similar results.
Warmup is training data that must be input into a reservoir system in preparation for the actual task.
“For our next generation of reservoir computing, heating time is not required,” Gadier said.
“Currently, scientists have to put in 1,000 or 10,000 data points or more to heat it up. And that’s just lost data, it doesn’t require real work. We’re going to put in one or two or three data points.” are,” he said. .
Once researchers are ready to train the reservoir system to make predictions, the next generation system again requires less data.
In a test for Lawrence’s forecasting task, researchers were able to obtain similar results using 400 data points, while the current generation was produced using 5,000 data points or more, depending on the desired accuracy.
“What’s exciting is that this next generation of reservoir computing is getting so much better and making it even better,” Gadier said.
He and his colleagues plan to expand this work to solve difficult computer problems such as predicting fluid dynamics.
“This is an incredibly challenging problem to solve. We wanted to see if we could speed up the process of solving that problem by using our simplified reservoir system model.” This is an incredibly challenging problem to solve. We wanted to see if we could speed up the process of solving that problem by using our simplified reservoir system model from previous work.” and “in theory, a single crystal pool may provide sufficient storage for any number — which means no additional need or expense on your part – however it also does require some modifications: If you are working with crystals made from phosphor metal alloys (such as bismuth) then each block has 1 side being transparent while the other side has 3 sides covered in reflective material,” he continues “… I think what many users wish was another way…where they put water back into the machine; this would be at least 5% faster than existing methods used so far…”
Eric Bolt, a professor of electrical and computer engineering at Clarkson University, co-authored the study; Aaron Griffith, Ph.D. in Ohio State Physics; and Wehsen Barbosa, a doctoral researcher in physics at Ohio State.