AI Evolution: Bridging the Gap to Human-Like Information Processing

The methodology of how humans process information can serve as a framework to educate next-generation artificial intelligence (AI) systems, as highlighted in research released on Jan. 22 in Nature.

Cory Merkel, an associate professor specializing in computer engineering at the Rochester Institute of Technology, was among more than a dozen researchers globally who contributed to the research outcomes.

Merkel is proficient in utilizing brain-inspired methodologies known as neuromorphic computing to create advancements that will enhance processing capability and energy efficiency for AI technologies.

“The possibility of having efficient AI on limited devices will create opportunities in various new application fields such as brain-computer interfaces, space exploration, health monitoring technologies, and autonomous surveillance systems, for instance,” Merkel remarked.

His research in neuromorphic computing is set to meet the rising demand for AI in applications constrained by size, weight, and power, like wearable devices, mobile phones, robotics, drones, and satellites. It will markedly enhance processing and storage capabilities.

Lead author Dhireesha Kudithipudi, a professor and founding head of the Neuromorphic Artificial Intelligence Lab at the University of Texas-San Antonio, collaborated with researchers from educational institutions, national laboratories, and industry to compile a thorough review of neuromorphic computing technology.

The authors of the article state that neuromorphic designers utilize the principles of biological intelligence identified by neuroscientists to create effective computational systems. These implementations demand more powerful computational intelligence, and the human brain serves as a model for how intricate networks can operate more swiftly and efficiently.

Merkel and Suma George Cardwell, a principal member of the technical staff at Sandia National Laboratory’s Center for Computing Research, elaborated in the recent publication on the significance of emerging memory technologies, such as RRAM and Spintronics, essential for mass storage devices. Both of these new technologies are attractive for large-scale neuromorphic computing systems, with the authors offering illustrations on utilizing these devices in learning and managing or counteracting device variabilities.

As AI’s energy demands are projected to increase twofold by 2026, neuromorphic computing presents itself as a viable solution. The authors also note that neuromorphic systems are approaching a “critical point,” with scale being a vital indicator to measure the advancement of the field.

Kudithipudi and Merkel have collaborated for a long time, dating back to when she served as a professor in RIT’s Kate Gleason College of Engineering. As the director of the Brain Lab in the Kate Gleason college, Merkel maintains his emphasis on enhancing deep learning models through neuromorphic computing.