Can a computer, like a human, learn from the past and anticipate the future? You might not be surprised to learn that advanced AI models can accomplish this, but what about a computer that looks more like a water tank?
For our exploration, presently distributed in Europhysics Letters, we have constructed a little verification-of-idea PC that utilizes running water rather than a conventional legitimate hardware processor and figures future occasions by means of a methodology called “supply registering.”
In benchmark tests, our analog computer did better than a high-performance digital computer at remembering input data and predicting future events in some cases.
What exactly does it do?
Tossing stones in the lake
Imagine two children, Alice and Weave, playing at the edge of a lake. Bob seems to throw large and small stones into the water at random.
Water waves of varying sizes are created by large and small stones. Alice learns to anticipate what the water waves will do next by watching the water waves made by the stones. As a result, she can predict which stone Bob will throw next.
The supply PCs duplicate the thinking system occurring in Alice’s cerebrum. They can learn from past contributions to foresee future occasions.
Reservoir computers can also be built with straightforward physical systems, despite the fact that neural networks—computer programs loosely based on the structure of neurons in the brain—were initially proposed as a method of construction.
While Alice watches the waves and tries to predict what will happen next, Bob tosses rocks into the pond. Credit: Yaroslav Maksymov, Creator
Supply PCs are simple PCs. In contrast to digital computers, which represent data in abruptly changing binary “zero” and “one” states, analog computers represent data continuously.
Analog computers are better than digital computers at modeling certain natural events that occur in an unpredictable sequence known as a “chaotic time series” because they can continuously represent data.
How to make predictions
To understand how a reservoir computer can be used to make predictions, imagine you have a year’s worth of daily rainfall data and a bucket of water nearby. The pail will be our “computational repository.”
We input the day-to-day precipitation record into the pail through stone. We throw a small stone on days when it only rains lightly and a large stone on days of heavy rain. For a day of no downpour, we toss no stone.
Waves are made by each stone and move around the bucket, interacting with waves made by other stones.
The state of the bucket’s water provides us with a prediction at the conclusion of this procedure. Our reservoir computer can be said to have predicted heavy rains if the interactions between waves resulted in large new waves. However, if they are small, we should only anticipate light rain.
It is likewise conceivable that the waves will drop each other, framing a still water surface. We shouldn’t expect any rain in that case.
Our repository PC utilized lone waves like those found in water fountains. Credit: Ivan Maksymov, Author provided.
The supply makes a weather conditions gauge in light of the fact that the waves in the can and precipitation designs develop over the long haul, keeping similar laws of material science.
Numerous other natural and socioeconomic processes follow suit. This indicates that a reservoir computer can also forecast human activity and financial markets.
Longer-lasting waves The reservoir computer for the “bucket of water” has its limits. First and foremost, the waves are brief. We require a reservoir with waves that are more durable in order to forecast complex processes like climate change and population growth.
“Solitons” is one option. These are waves that keep their shape over long distances and self-reinforce.
For our supply PC, we utilized minimized soliton-like waves. You frequently see such waves in a bathroom sink or a water fountain.
A metal plate with a slight inclination is covered by a thin layer of water in our computer. Solitary waves are produced, and the flow’s speed is altered by a small electric pump.
We added a fluorescent material to make the water shine under bright light and quantify the size of the waves definitively.
In Alice and Bob’s game, the pump is like falling stones, but the isolated waves are like waves on the water’s surface. Our computer can process data at a faster rate because solitary waves move much faster and last longer than bucket waves.
So, how well does it work?
We tried our PC’s memory’s capacity to recall past sources of information and make estimates for a benchmark set of tumultuous and irregular information. Not only did our computer perform all tasks exceptionally well, but it also performed better than a high-performance digital computer that was tasked with the same issue.
We also created a mathematical model with my colleague Andrey Pototsky that helped us better comprehend the solitary wave’s physical properties.
Then, we intend to scale down our PC to a microfluidic processor. Water waves ought to have the option to do calculations inside a chip that works much the same way as the silicon chips utilized in each cell phone.
Long-term forecasts for climate change, bushfires, and financial markets may be possible with our computers in the future at a lower cost and wider availability than with current supercomputers.
Because it does not make use of digital data, our computer is also naturally resistant to cyberattacks.
Our vision is that a soliton-based microfluidic supply PC will bring information science and AI to countries and distant networks around the world. But for the time being, our research is going on.
More information: Ivan Maksymov et al, Reservoir computing based on solitary-like waves dynamics of liquid film flows: A proof of concept, Europhysics Letters (2023). DOI: 10.1209/0295-5075/acd471
