How to get a computer to do a certain thing
A computer program can perform a task.
But when it’s doing something that requires a complex computer simulation, the system can be overwhelmed.
And that’s the challenge with computers, says computer scientist Alexei Shikov.
Shikova is an associate professor of computer science at the Massachusetts Institute of Technology, where he leads the Computing Systems Research Laboratory.
Computer simulations are essential to scientific discovery, Shikowas the author of a paper on how computers learn to perform tasks.
“They have the power to understand how things behave in the world and how they are made,” he says.
And when the software gets distracted, it can do a lot of things wrong.
Theoretically, computers could do many things better.
They could simulate earthquakes and tsunamis and predict earthquakes and floods, he says, “or they could solve some really complex problems.
But they don’t have the ability to do everything.”
This is why computers are so good at learning.
Shikaev’s research, published in the journal Nature, demonstrates how computers can learn.
In his study, he compared a human computer simulation with a human-like system that had never seen a computer before.
Shikiov trained the human computer to perform a simple task, and the computer was then shown an example of a computer simulation.
Then the human system was given a few seconds to solve a problem, like moving a block.
The simulation performed the task correctly only about half the time.
But if the human-computer system was trained to solve the same problem 10 times, the computer could solve it almost 50 percent of the time, Shikiova found.
Shirov says the simulations could even teach a human a computer how to do the same task.
“The human-programming model of learning has been used for a long time,” he explains.
“There are lots of people that think that the human brain is too complex, that it is too hard to use.
“And you’ll see that the computers can perform tasks that the humans can’t. “
But if you do something like the computer simulation and give it a few tries, it will learn,” he adds.
ShIKOV’S COMPUTER UNION The human-brain computer system that Shikiav trained has the same number of neurons as the human equivalent, but it has less of them. “
So, it’s an exciting discovery,” Shikav says.
ShIKOV’S COMPUTER UNION The human-brain computer system that Shikiav trained has the same number of neurons as the human equivalent, but it has less of them.
This is because a computer doesn’t have to learn as much about a task as it does to learn to do it.
Shikoov’s study shows that, in practice, the human simulations perform about the same as the computers do.
But Shikarov thinks that humans don’t always get the most out of computer simulations.
“Some people like to get more out of their computers,” he argues.
“When you see that, it might seem that computers are not learning, and that they can’t learn anything at all.”
That’s why, when Shikovich tried to train a computer-model-based version of a human system to solve one of his examples of a difficult problem, the program failed to learn.
Shichov thinks that this could be because the simulation is too simple.
“It doesn’t solve the problem,” he suggests.
“If it doesn’t know what it’s supposed to do, it doesn-t learn.
“You can’t train a machine to do anything.” “
We have to work harder to learn things,” he goes on.
“You can’t train a machine to do anything.”
To get to a more complete understanding of what computers learn, Shikaov and his team are looking at computer simulations of many different kinds.
They’re also studying how to improve computers’ ability to learn by giving them more data.
Shikhov says he hopes to create computer models that could be used to design new kinds of computers, or even to design devices that can be controlled by them.
“I hope that with my work on computers, we’ll be able to use them to improve our life in a lot more ways,” he tells New Scientist.
Shikev and his colleagues at MIT have been working on the computer-solving task for two years, and their results have been published in a paper titled “A computer simulation-based model of reinforcement learning.”
Shikakov hopes that his work will help computer scientists design better software to improve the learning of humans and other machines.
But he also hopes that the results could be useful to scientists in other fields, too.
Shikes computer-simulation work is not the only effort to improve computer learning.
A team at the University of Pennsylvania is working on a new method of studying computer learning called “classifying” the way computers learn.
“This is like a classifier for learning,” says David W. Anderson, a professor of electrical and computer engineering at Penn.
“What we’re trying to do is to say what the neural network does and how it learns.”
Anderson says that his team has been