Conference: NeST: Neuromorphic Stochastic Thermodynamics Workshop at SFI

Santa Fe Institute, Santa Fe NM

Investigators

Abstract

A workshop is proposed to examine the intersection of neuromorphic computing and stochastic thermodynamics. Neuromorphic Computing is a field that started decades ago, inspired by the efficiencies of the information processing performed by biological brains. Its main goal is to replicate the complex architecture and functionality of biological neural circuits but using in-silico circuits. The field has played a key role in advancing artificial intelligence (AI) systems as evidenced by the adoption of neuromorphic concepts like compute-in-memory, attention, pooling to name a few, into the mainstream AI designs. However, a detailed comparison between in-silico neuromorphic brains and biological brains reveals a significant gap between the two systems. Even a simple insect brain with fewer than a million neurons can perform and learn a variety of complex tasks, a feat currently unachievable by state-of-the-art AI systems, let alone a neuromorphic computer. On the other end of the spectrum are large-scale AI systems, such as large language models (LLMs) and vision transformers, whose success has been characterized by a rapid increase in the complexity and the size of their flowcharts involving various abstract processes. The energy cost for training and using these AI systems is becoming unsustainable, as the field strives to achieve artificial general intelligence (AGI) or brain-scale intelligence. It is To address this above-mentioned energy challenge, the scientific community needs a fundamental understanding of the thermodynamics of distributed computational architectures, particularly how energy expenditure varies across different architectures that implement the same computational task. In this regard, the planned workshop will lead to a new body of knowledge and new research directions that will synergize the field of neuromorphic engineering/computing with the latest advances in the field of stochastic thermodynamics. It is hoped that the workshop will produce a new direction and will thus help NSF reformulate its programs according to the needs of scientific disciplines as well as emerging technologies. The workshop will produce reports to be made broadly available to the community. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.