Researchers from Bar-Ilan University in Israel have solved the puzzle of why the brain, with its limited mathematical operations, can still compete with advanced artificial intelligence systems. In a recent study published in Scientific Reports, they have shown that efficient learning on an artificial tree architecture can achieve better classification success rates than deep learning. This architecture consists of highly pruned trees, which are a step towards a biologically plausible realization of efficient dendritic tree learning.
The new type of AI is based on previous research by the lead researcher, Prof. Ido Kanter, and his team that indicates evidence for sub-dendritic adaptation using neuronal cultures. The efficient implementation of this type of AI requires a new type of hardware that differs from GPUs, which are better suited for deep learning. The emergence of this new hardware will be key to efficiently imitating brain dynamics.
PhD student Yuval Meir, a contributor to the study, said, “Highly pruned tree architectures represent a step towards a biologically plausible realization of efficient dendritic tree learning by a single or several neurons, with reduced complexity and energy consumption.”
This breakthrough opens up new avenues for the development of efficient and biologically-inspired AI hardware and algorithms. The findings demonstrate that the brain’s limited mathematical operations can still compete with advanced AI systems, leading to a more efficient and biologically-realistic approach to AI.