类脑智能与深度学习的几个问题与思考

Thoughts and prospects of brain-inspired intelligence

得益于数据规模与机器算力的增长,深度学习技术正取得空前繁荣。然而,深度学习擅长在预设的封闭环境,为特定任务找到有用的数据表示,在结构、处理和功能上距离人类智能仍有较大差距。类脑智能旨在模拟人脑神经元的运行机制、感知模式与认知机理,借助机器强大的信息整合、搜索、计算等能力,以软硬件联合的智能新形态构造接近人类水平的智能机器,是未来人工智能的发展方向。基于第216期"双清论坛",本文将分析人工智能与深度学习发展现状与局限,并从认知建模、模块装配、意识先验、自主演化、协同学习几方面分析未来类脑智能的可能发展方向。

Thanks to the growth of data scale and machine power,deep learning technology has achieved unprecedented prosperity.However,most of the available deep neural networks are good at providing useful data representation for a specific task in closed environment,which has remarkable difference with real human brain.Brain-like intelligence aims to simulate the processing,perception and cognitive mechanism of human brain neurons,and establish intelligent machines close to human level in a new form including both software and hardware,via the utilization of computers with powerful calculation capabilities.Now brain-like intelligence is guiding the direction of Artificial Intelligence(AI)research works.In this paper,we reviewed the current situation of AI and analyzed the limitations of deep neural networks.Meanwhile,some possible development directions of brain-like intelligence,including cognitive modeling,module assembly,conscious priority,autonomous evolution and collaborative learning,were also discussed.