2024. 10. 25. 23:04ㆍScience/Biology
Brain organoid reservoir computing for artificial intelligence
Brain-inspired computing hardware aims to emulate the structure and working principles of the brain and could be used to address current limitations in artificial intelligence technologies. However, brain-inspired silicon chips are still limited in their ability to fully mimic brain function as most examples are built on digital electronic principles. Here we report an artificial intelligence hardware approach that uses adaptive reservoir computation of biological neural networks in a brain organoid.
In this approach - which is termed Brainoware - computation is performed by sending and receiving information from the brain organoid using a high-density multielectrode array. (MEA) By applying spatiotemporal electrical stimulation, nonlinear dynamics and fading memory properties are achieved, as well as unsupervised learning from training data by reshaping the organoid functional connectivity. We illustrate the practical potential of this technique by using it for speech recognition and nonlinear equation prediction in a reservoir computing framework.
Introduction
The recent success of artificial intelligence (AI) has been largely driven by the development of artificial neural networks (ANNs)1, which process large datasets using silicon computing chips
. However, training ANNs on current AI computing hardware is energy intensive and time consuming.. The physical separation of data from data-processing units - known as the von Neumann bottleneck - is a key cause of these issues. The slowing of Moore’s law also places further limitations on current AI hardware. Thus, alternative approaches for the develop- ment of AI hardware are needed.
* 지금 하고 있는 PIM(Processing in Memory) 연구랑 문제 의식은 동일한데, 해결 방안이 달라서 신기했다.
Processing in Memory는 기존 폰노이만 아키텍처에서 구조적 변형을 주는 방법론이라면, Brainoware은 다른 차원의 하드웨어를 구현하는 방식인 것 같다. 둘 다 동일한 문제를 해결하고자 시작했다는 점이 신기하다.
The human brain is a complex three-dimensional biological network of about 200 billion cells, which are linked to one another via hundreds of trillions of nanometre-sized synapses. Its structure, function and efficiency could be a powerful source of inspiration for the development of AI hardware. (어떻게 여기로 흘러갈 수가 있지 . . )
In particular, a. human brain typically expends about 20 watts, whereas current AI hardware consumes about 8 million watts to drive a comparative ANN5. The brain can also effec- tively process and learn information from noisy data at minimal training cost through neuronal plasticity and neurogenesis13,14, avoiding the large energy consumption of high-precision computing approaches.
The human brain fuses data storage and processes within biological neural networks (BNNs), naturally avoiding any von Neumann bottleneck issues. Inspired by BNNs, attempts have been made to develop high-efficiency and low-cost neuromorphic chips - using memristors, for example - that store previously experienced current or/and voltages in internal states and enable short-term memory.
Such neuromorphic chips have been used for various applications, for example, in computer vision24,25 and speech recognition26,27. However, current neuromorphic chips can only partially mimic brain functions, and there is a need to improve their processing capability and accounting for real-life uncertainty and improving energy efficiency.
Brain organoids are in vitro three-dimensional aggregates that are created through the self-organization and differentiation of human pluripotent stem cells and can become brain-like tissues that can recapitulate aspects of a developing brain’s structure and function.
In this Article, we report an AI hardware that harnesses the reservoir computation and unsupervised learning ability of organoid neural networks (ONNs) in a brain organoid. The approach - termed Brainoware - processes spatiotemporal information, and achieves unsupervised learning, probably through the neuroplasticity of the brain organoid. Compared with current two-dimensional (2D) in vitro neuronal cultures and neuromorphic chips (Supplementary Table 1), Brainoware could provide additional insights for AI computing because brain organoids can provide BNNs with complexity, connectivity, neuroplasticity and neurogenesis, as well as low energy consumption and fast learning.
출처: https://www.nature.com/articles/s41928-023-01069-w