A new paper from a team of researchers at MIT, the University of Cambridge and the University Hospital of Liege has discovered a novel algorithm that uses machine learning to help clinicians make better decisions about patients and the environment.
In their paper published online Monday in Nature Communications, the researchers describe their algorithm as a neural network, a term that refers to a system that is capable of learning from data and using that knowledge to make decisions about human behavior.
The algorithm uses a neural net to identify patterns in patients’ brain activity.
It then uses these patterns to develop an algorithm that is able to predict how patients will react to a given stimulus.
The researchers used a computer simulation to develop the algorithm, which is similar to other algorithms that use neural nets to create a virtual environment for a patient to interact with, say, a computer game or a game of chess.
The paper describes how the algorithm is able and then how it can make decisions on the basis of this simulated environment.
“This is really an exploration of the human brain and how it processes emotion,” says MIT’s Dr. Benjamin Wysocki, who was not involved in the work.
“The problem is that there are many different types of emotions and they can be extremely complex.
So you need to be able to generate some sort of model that can be used in a very precise way to sort through them.”
One problem is the difficulty in using an algorithm to accurately predict the behavior of an individual patient.
This is where machine learning comes in.
“When you have an algorithm and you’re making a prediction, you need some sort the data that you have, the context in which it’s being generated,” says Dr. John McCarthy, who led the work while he was at the University and is now at the Massachusetts Institute of Technology.
“But how do you know what context the data is in?
So how do we find out what the parameters are?”
To answer that question, the team created a virtual brain environment called the MRI environment, which was a virtual space in which the patients were shown pictures of faces, which they could interact with.
“We could make these visual cues to them, so that they were able to tell what the emotion is,” says McCarthy.
The team then used the algorithm to generate images that were shown to patients and to the MRI staff.
“Our algorithm uses the image data to build an image of the patient in front of you,” explains Wysicki.
“You can see that the patients are aware of it, they are trying to figure out what they’re seeing.
We call this the emotional response to the stimulus, and this is what we call the emotional perception.
The next step in this algorithm is to generate the neural representations of the images and see what they look like when they’re presented to the patient.”
This is a common problem in AI research, says Wysicks.
“So the first thing we tried was to try to figure this out by playing a computer-generated game, where we were trying to predict what the patients’ responses would be.
We found that we could generate a bunch of images and then if we used these images to predict their reactions, we would be able predict the emotional responses of the patients,” says Wiesicki, explaining that the algorithms algorithm is very good at predicting what the patient’s reactions will be based on the image.
“It’s a very accurate prediction, and you can see it in action,” he says.
“In the MRI world, we could see that we were predicting with 99.9 percent accuracy.
So it’s not something that’s very difficult to do.”
The researchers say the algorithm can be extended to predict more complex emotional events, including those that involve social interaction or emotional contagion.
The results could also be applied to other situations that require a lot of data.
For example, the algorithm could be used to predict the reactions of children to a video game, or to predict whether a child will become suicidal or not.
“If you have a child who is depressed, this algorithm could help you predict whether that child will be suicidal or what they might be feeling about themselves,” says Gregory D. Lee, a postdoctoral fellow at MIT and the paper’s lead author.
“That’s something that we think will be useful in the future, because we don’t really know how depression works in the brain.”