For decades, researchers have relied on simple mathematical models to describe how neurons in our brains respond to what our eyes see every day in life. These old models were easy to interpret, but they failed to predict how cells in the higher regions of the visual cortex in the brain would function.
Hot Topics
About half of Libyans drink contaminated water What is the scientific reason?
Mauritania's blue treasure is in danger Will the map of fish in the Atlantic be redrawn?
The Citrus Secret Everyone's Talking About: Burn More, Crave Less & Feel Great Naturally
This deficiency has led scientists to adopt artificial intelligence networks composed of multiple layers. While these networks have succeeded in making highly accurate predictions, the price of this success has been the loss of comprehension, as tracing the path of information through millions of variables has become virtually impossible.
To solve this dilemma, a research team presented "Nature," in which the team decided to apply an old philosophical and scientific principle known as "Occam's Razor," which states that the simplest explanation is often the most accurate.
Researchers wondered whether they could compress and shrink giant computer models to reveal the actual calculations taking place inside our brains. They hypothesized that these models are riddled with digital filters and redundant connections that don't accurately represent the brain's true workings, and that removing them would reveal the true mechanism by which our brains interpret the images and situations we see every day.
He adds: “Despite their amazing achievements, large AI models are difficult to understand, just like our brains, due to their enormous size. It is difficult to identify the important filters for the system’s output, and to know which filters are redundant or unnecessary.”
When we look at an object, light enters the eye, and the signals travel to the back of the brain, passing through several processing stages. One of these stages involves a specialized area responsible for recognizing complex shapes, colors, and patterns. To understand this recognition process, scientists decided to study the brains of macaque monkeys, which are remarkably similar to our human visual system.
The team implemented a giant artificial intelligence model that predicts and recognizes patterns, then used advanced techniques to simplify this giant model.
One such technique is called "knowledge distillation," which resembles the educational process where an expert professor reads extensive and complex references, then distills this knowledge and passes it on to the student in a simplified and direct manner. In this experiment, a small network (the student) learned to mimic the direct results produced by the giant network (the expert professor) by practicing on millions of images to extract the fundamental principles.
Tags:
science
