New research from Rensselaer Polytechnic Institute (RPI) could help shape the future of artificial intelligence by making AI systems less resource-intensive, higher performing, and designed to emulate the human brain.

As AI models grow ever larger, so do their costs and limitations. Researchers at RPI and City University of Hong Kong offer a potential solution: instead of expanding outward with more layers and data, they propose building upward — adding internal structure that mirrors a 3D biological neural network and incorporating recursive loops to enhance network introspection. This vertical dimension and loop allow artificial neural networks to process information more effectively and efficiently, potentially transforming AI’s ability to learn and adapt in 3D and higher dimensions.  

“This new AI framework not only boosts efficiency but also unlocks practical opportunities,” said Wang. “This research could be a crucial step toward driving advancements in next-generation artificial neural networks, closely relevant to healthcare and education, while paving the way for deeper insights into how the human brain works.”

By introducing a vertical “height” dimension and feedback loops that allow artificial neural networks to relate, reflect, and refine outputs, the new design ideas could make AI models much smarter, leading to potentially major implications. Doing more with fewer resources could help expand access to advanced AI technologies, reduce the environmental footprint of massive model training, and enable more real-time applications — from robotics to personalized medicine.  

One such application lies in neuroscience, where brain-inspired neural networks could help scientists better understand cognition and even uncover new clues about neurological disorders such as Alzheimer’s and epilepsy.

“This framework isn’t just about smarter AI — it’s about more sustainable, accessible, and explainable AI,” Wang said. “And it may help us learn more about our own brains along the way.”

The collaborative study was led by Ge Wang, Ph.D., Clark & Crossan Endowed Chair and director of the Biomedical Imaging Center at Rensselaer, and Fenglei Fan, Ph.D., Wang’s former Ph.D. student and current assistant professor at City University of Hong Kong. The research was recently published in Patterns, a journal from Cell Press, under the title Dimensionality and Dynamics for Next-Generation Neural Networks.

This work builds on RPI’s longstanding leadership in AI research. Through major initiatives such as the AI Research Collaboration with IBM and the Future of Computing Institute, RPI researchers are developing cutting-edge AI technologies that are designed not only to meet human needs, but to redefine them. From advancing human-machine collaboration to exploring brain-inspired computing, RPI is committed to shaping the future of AI.

Read the full paper here: https://www.cell.com/patterns/fulltext/S2666-3899(25)00079-0