Ask Onix
Surrey researchers boost AI efficiency by mimicking human brain networks
Scientists at the University of Surrey have developed a new method to enhance artificial intelligence performance by replicating the neural structures of the human brain, according to a study published in Neurocomputing.
How the brain-inspired model works
The approach, called Topographical Sparse Mapping, organizes artificial neural networks to mirror the brain's efficiency-connecting each neuron only to nearby or functionally related neurons. This reduces redundant connections, improving both speed and sustainability without compromising accuracy.
Dr. Roman Bauer, senior lecturer at Surrey, stated:
"Our work demonstrates that intelligent systems can be built far more efficiently, drastically cutting energy demands while maintaining performance."
Energy savings and scalability concerns
Researchers highlight that training large AI models-like those powering generative tools such as ChatGPT-can consume over a million kilowatt-hours of electricity.
"That level of energy use is unsustainable as AI continues to expand," Dr. Bauer warned.
Next-generation refinements
An advanced version, Enhanced Topographical Sparse Mapping, incorporates a "pruning" mechanism inspired by biological learning. This process gradually refines neural connections during training, akin to how the human brain optimizes itself over time.
Future applications
The team is exploring broader uses, including neuromorphic computing-a field that designs hardware to emulate the brain's architecture. Such systems could enable more realistic, energy-efficient AI applications beyond current models.
