Researchers at Johns Hopkins University found that AI systems designed to act like the human brain can recognize images without needing new data. This study in Nature Machine Intelligence questions the industry’s focus on heavy training, which is costly in time, energy, and money.
In the research, scientists changed three main types of neural networks. They found that only changes in convolutional networks led to big improvements. The new models showed brain-like reactions similar to humans and monkeys. Remarkably, even without training, these networks did as well as traditional systems trained on billions of images.
The findings suggest that having the right structure could lead to faster and more efficient learning in artificial intelligence. The research team is already working on biologically inspired algorithms that could change the future of deep learning.
