A new study suggests that the efficiency of artificial intelligence (AI) can be enhanced not just by deepening neural networks but by choosing the most influential paths to an outcome. This research, published in *Scientific Reports*, has been spearheaded by Bar-Ilan University in Israel, offering a novel approach to improving AI classification tasks.
Deep Learning (DL), a subset of AI, classifies data through a series of interconnected layers. Traditionally, the decisions to guide this classification are made step-by-step within these layers. But what if AI systems made decisions based on a more comprehensive view of the path to an outcome? The recent research confirms the validity of this holistic approach.
Prof. Ido Kanter, from Bar-Ilan’s Department of Physics and Gonda (Goldschmied) Multidisciplinary Brain Research Center, likens the traditional method to a child trying to climb a winding mountain by always picking the nearest path. In contrast, the novel method is akin to a child using binoculars, assessing the entire route, and choosing the most direct path to the summit. Although the latter might seem slower at the onset, it proves more efficient in the long run.
Yarden Tzach, a PhD student involved in the study, noted, “This discovery can pave the way for better enhanced AI learning, by choosing the most significant route to the top.”
The research team, led experimentally by Dr. Roni Vardi, has been working to bridge the gap between the organic intricacies of the biological world and the algorithmic precision of machine learning. Their endeavors have led to the discovery of efficient dendritic adaptation in neuronal cultures and its application in machine learning. Their findings demonstrate how shallow AI networks, with the right approach, can match the performance of much deeper networks.
The study’s conclusions promise a potential revolution in AI methodologies. By making globally-informed decisions, AI systems could significantly improve their classification tasks without the complexities and computational cost of adding more layers.
