Can a machine predict the future like a human and learn from the past? You might not be shocked to learn that certain state-of-the-art AI models are capable of accomplishing this feat, but what if a computer had a slightly different appearance—more resembling a tank of water?
So how does it work?
Hurling rocks into the water
Think of Alice and Bob as two little children having fun by a pond. Bob appears to throw large and small stones into the river at random, one at a time.
Different-sized water waves are produced by large and small stones. Alice learns to predict the behaviour of the waves by seeing the water waves the stones make; from this, she may guess which stone Bob will toss next.
Computers in the reservoir replicate Alice’s mental processes. They can use historical data to make predictions about the future.
Although neural networks—computer programmes loosely based on the structure of neurons in the brain—were initially envisioned as a way to create reservoir computers, they may also be made with just a few basic physical components.
Analogue computers are reservoir computers. Unlike digital computers, which represent data as abruptly shifting binary “zero” and “one” states, analogue computers represent data continually.
A “chaotic time series” is a term used to describe a set of natural events that occur in an unpredictable order. Analogue computers can model these events better than digital computers since they can represent data in a continuous form.
How to foresee things?
Imagine that you had a bucket of water nearby and a record of the daily rainfall for the previous year in order to comprehend how we can utilise a reservoir computer to make predictions. We’ll use the bucket as our “computational reservoir.”
We use a stone to manually enter the daily rainfall data into the bucket. We throw a tiny stone for a day of mild rain and a large stone for a day of heavy rain. We don’t hurl any rocks on a day without rain.
Each stone makes a wave, which interacts with waves made by other stones as they swirl around the bucket.
The condition of the water in the bucket at the conclusion of this operation allows us to make a forecast. We can state that our reservoir computer anticipates significant rainfall if the interactions between waves result in massive additional waves. However, if they are little, then just light rain should be expected.
Another possibility is that the waves will cancel one another out and create a surface of still water. Therefore, we shouldn’t anticipate any rain.
Longer-lasting waves
The reservoir computer for the “bucket of water” has its limitations. One reason is that the waves are transient. We require a reservoir with more resilient waves in order to foresee complicated processes like population growth and climate change.
One choice is “solitons.” These waves are self-reinforcing, maintain their shape, and travel long distances.
We used compact soliton-like waves in our reservoir computer. Such waves are frequently seen at drinking fountains or in bathroom sinks.
A small layer of water is flowing over a metal plate that is tilted slightly in our computer. A tiny electric pump modifies the flow’s velocity and produces isolated waves.
In order to precisely estimate the wave size, we added a fluorescent substance that made the water shine when exposed to ultraviolet light.
In the game played by Alice and Bob, the pump represents the falling stones, and the single waves represent the waves on the water’s surface.
Our computer can analyse data more quickly because solitary waves process information more quickly and sustain themselves for longer than waves in a bucket.
So, how does it perform?
We put our computer’s forecasting and memory capabilities to the test with a benchmark set of chaotic and random data. Our computer performed better than a high-performance digital computer when given the same task than it did in general.
Together with my colleague Andrey Pototsky, we also developed a mathematical model that helped us comprehend the solitary waves’ physical characteristics.
The next step is to downsize our computer and use it as a microfluidic processor. It should be possible for water waves to do calculations inside a chip that functions similarly to the silicon processors found in every smartphone.
Our computer may be able to generate trustworthy long-term projections in the future for issues like climate change, bushfires, and financial markets—at a considerably cheaper cost and with greater accessibility than present supercomputers.
Since our computer does not use digital data, it is also inherently resistant to cyberattacks.
Our goal is to make data science and machine learning accessible to remote and rural areas all around the world with a soliton-based microfluidic reservoir computer. But for now, we’ll keep working on our research.
