类脑智能研究现状与发展思考

Research Status and Developments of Brain-inspired Intelligence

        徐波(中国科学院自动化研究所 北京 100190;中国科学院脑科学与智能技术卓越创新中心 上海 200031)
        刘成林(中国科学院自动化研究所 北京 100190;中国科学院脑科学与智能技术卓越创新中心 上海 200031)
        曾毅(中国科学院自动化研究所 北京 100190;中国科学院脑科学与智能技术卓越创新中心 上海 200031)

         类脑智能;人工神经网络;记忆;注意和推理;脉冲神经网络;多脑区协同;自主学习

        brain-inspired intelligence;artificial neural networks;memory;attention and reasoning;spiking neural networks;multiple brain region coordination;autonomous learning

        近年来人工智能研究的许多重要进展反映了一个趋势：来自脑科学的启发，即使是局部的借鉴都能够有效地提升现有人工智能模型与系统的智能水平。然而，想要真正逼近乃至超越人类水平的人工智能，还需要对脑信息处理机制更为深入的研究和借鉴。类脑智能研究的目标就是通过借鉴脑神经结构及信息处理机制，实现机制类脑、行为类人的下一代人工智能系统。文章从受脑启发的新一代人工神经网络、基于记忆、注意和推理的认知功能模型、基于生物脉冲神经网络的多脑区协同认知计算模型等角度，并结合研究团队在类脑智能领域的研究进展，论述类脑智能的研究进展、发展方向和对未来发展的思考。

        Recent advances in Artificial Intelligence (AI) have manifested an important trend, namely, inspirations from brain science can significantly improve the level of intelligence for AI computational models. With only local and partial inspirations from the brain, great advancements have been achieved. Nevertheless, deeper investigations and inspirations from the brain are needed to realize and exceed humanlevel intelligence. The ultimate goal of brain-inspired intelligence is to bring inspirations from brain structures and information processing mechanisms to brain-inspired cognitive computational models, so as to realize next-generation artificial intelligence models and systems with general intelligence. In this article, we review recent advances and discuss trends of brain-inspired computational models, including new models of artificial neural networks, and cognitive computation models. We also briefly introduce the research of brain-inspired cognitive computation models and methods supported by the strategic priority research project of Chinese Academy of Sciences.