Recent graduate Jumpei Ukita and Professor Kenichi Ohki from the Department of Physiology at the University of Tokyo School of Medicine have developed and tested a new way to improve the defense of artificial neural networks, leaving a more secure artificial intelligence (AI) system.
For those in a hurry:
Ukita and Ohki are scientists who have not only explored computer science. They also studied the human brain and this inspired them to use a phenomenon discovered in an artificial neural network (ANN).
The research, published in the journal Neural Networks and disseminated by University of Tokyohighlights the dynamic nature of the ongoing battle between executioners and defenders in the field of artificial intelligence.
AI and the brain
Neural networks, inspired by the brain’s biological neurons, form the backbone of most artificial intelligence systems. These networks process information across layers, with inputs arriving at one end and outputs emerging from the other. They are essential for making automatic decisions, such as in autonomous vehicles.
However, these networks are not infallible. They can be misled, either intentionally or accidentally, leading them to misinterpret inputs. This vulnerability is concerning, especially in crucial applications such as medical diagnostics and self-driving cars.
Typically, defense strategies focus on the initial input layer, leaving the internal layers vulnerable. But researchers have applied a phenomenon observed in the brain to artificial neural networks.
They introduced random noise into these internal layers to strengthen the resilience of the network. This approach goes beyond conventional techniques, resulting in a more robust system. By adding noise not only to the input layer but also to deeper layers, they increased adaptability without compromising normal operation.
Their approach, tested in a simulated adversarial attack, proved effective, significantly reducing network vulnerability. However, researchers are eager to refine and expand this method to protect against a wider range of attacks.
The post Scientists make AI safer by looking at the human brain appeared first on Olhar Digital.
Source: Olhar Digital
I am Joe Dow, a professional content creator and news journalist for Run Down Bulletin. I specialize in covering technological trends and advancements, with an emphasis on their real-world implications. My work has been featured in publications such as The Guardian, Wired Magazine, and The Verge.