与6位图灵奖得主和100多位专家
共同探讨人工智能的下一个十年
长按图片或点击阅读原文,内行盛会,首次免费注册
6月21-24日,第二届北京智源大会(官网:https://.)将邀请包括6位图灵奖获得者在内的上百位人工智能领袖,一起回顾过去,展望未来,深入系统探讨“人工智能的下一个十年”。本次大会将开设19个专题论坛,主题涵盖人工智能数理基础、自然语言处理、智能体系架构与芯片、人工智能伦理治理与可持续发展、机器学习、智能信息检索与挖掘、认知神经基础、机器感知、决策智能、AI医疗、AI创业、AI交通、AI+大数据+防疫、AI框架、图神经网络、知识智能、强化学习、青年科学家机器学习前沿,以及AI科技女性,遍历人工智能基础研究及创新应用,结合时局解析人工智能最新动态,探索未来发展方向。
我们将带你走进各个分论坛,领略嘉宾风采、洞悉前沿趋势。今天介绍的是将于6月22日下午举行的认知神经基础论坛。
论坛主题:认知神经基础
近年来,人工智能在解决诸如客体识别等问题上取得了令人瞩目的进展,可以与人类的表现相匹配,甚至超过人类的表现。人工智能的这些进展在很大程度上受到了神经科学对生物大脑的研究的启发,如来自生物神经网络的结构和功能的指导。但是,作为人工智能的主要载体之一的人工神经网络仍然是“黑匣子”,对网络组件的内部表征和计算过程仍知之甚少。
在这个论坛里,我们阐述认知神经科学进一步助力人工智能的潜力。具体来说,认知神经科学中可用的研究技术和方法,包括单细胞记录、脑功能成像、认知建模和损伤技术,可以作为揭示人工智能黑匣子、阐明人工智能网络内部计算和表征的工具库。此外,认知神经科学的研究成果可以通过构建先验的架构、算法和知识,为开发新一代人工智能提供启发。由此,我们提出将人工智能与认知神经科学进行深度融合,以创建认知神经智能的新的交叉领域。
认知神经基础论坛议程
论坛主席
刘嘉
北京师范大学心理学部教授,美国麻省理工学院博士。主要研究领域:客体识别和知觉学习。国家杰出青年基金获得者、中科院百人计划入选者、教育部长江学者、中青年科技创新领军人才、万人计划科技创新领军人才。任《心理科学》副主编、《中国大百科全书(心理学卷)》副主编、《最强大脑》科学总顾问。在Nature Neuroscience、Current Biology、PNAS等期刊发表SCI/SSCI论文百余篇,承担多个国家级重点项目,部分成果曾获教育部自然科学奖一等奖、中华全国归国华侨联合会中国侨界贡献奖等奖励。
演讲主题及嘉宾介绍
1.Knowledge representation in the Human Brain
议题简介:Human brain stores tremendous amount of knowledge about this world, which is the foundation of language, thought, and reasoning. What’s the neural codes of knowledge representation? Is the knowledge “roses are red” simply the memory trace of perceiving the color of roses, stored in the brain circuits within color-sensitive neurons? What about knowledge that is not directly perceived by senses, such as “freedom” or “rationality”? What’s their relationship with machine-based knowledge representation? I will present some work from my lab that addresses this issue using cognitive, neuroimaging, and neuropsychological methods with healthy subjects, individuals with sensory deprivation (blind and deaf) or with brain damage. The findings point to a highly distributed system incorporating two different types of information coding – one based on distributed sensory experiences and one based on language.
演讲嘉宾:毕彦超
北京师范大学认知神经科学与学习国家重点实验室教授、麦戈文脑科学所研究员,哈佛大学博士。主要研究领域:语义表征和物体表征。教育部长江学者特聘教授、国家自然科学基金委杰青、973计划青年项目(首席科学家)、基金委优秀青年、中组部万人计划青年拔尖人才。任Cognition、Neurobiology of Language、Language, Cognition and Neuroscience、Cognitive Neuropsychology 杂志编委。在Nature communications、Plos Biology、Brain、Trends in Cognitive Sciences等杂志发表与人类语义和语言机制的文章40余篇。连续两年被 Elsevier 评选为中国心理学领域高被引学者(、),受邀在认知神经科学领域每五年更新一版的“Bible”《Cognitive Neuroscience》最新版(第六版)中独立撰写《客体概念》相关章节。
2.Maps and Functions of Human Attention
议题简介:In everyday life, our brain is faced with the critical challenge of selecting the most relevant fraction of external inputs at the expense of less relevant information. Attention is widely acknowledged to be responsible for this selection process. In the first part of this talk, we show how to identify two key components of attention -the saliency map and the priority map in the human brain. Both the maps describe the topographic representation of attentional allocation. The saliency map is primarily based on bottom-up physical inputs, while the priority map is determined by both bottom-up and top-down signals. We demonstrate that the saliency maps from artificial and natural images are created in early visual cortex, especially in V1. Based on the properties of V1 neurons and the principle of information maximization, we propose a computational saliency map model to simulate human saccadic scanpaths on natural images, which outperforms many other models. Furthermore, we use the fMRI population receptive field (pRF) mapping technique and eye tracking technique, and show that the priority maps of natural images could be found in early visual cortex, including V1-V3. These findings provide converging evidence that the neural substrate of attention maps (including the saliency map and the priority map) could be located in human early visual cortex and significantly extend traditional attention theories that emphasize that only the parietal and frontal cortices are responsible for generating attention. In the second part, we address how the brain concurrently attends to multiple features. We found that, instead of being simultaneously and persistently enhanced, the neural representations of the attended features alternated with each other as they underwent a theta-band (~4 Hz) rhythmic fluctuation over time. This finding revealed for the first time a rhythm-based, time-multiplexing neural machinerysupporting concurrent multi-feature attention.
演讲嘉宾:方方
北京大学心理与认知科学学院教授/院长、北大-清华生命科学联合中心高级研究员、北京大学机器感知与智能教育部重点实验室教授/副主任、北京大学麦戈文脑研究所研究员/常务副所长,行为与心理健康北京市重点实验室主任,美国明尼苏达大学博士及博士后。主要研究领域:视知觉、意识、注意及其神经机制。在Nature Neuroscience、Neuron、Nature Communications、PNAS、Current Biology等顶级神经科学期刊和IEEE CVPR、IJCV、NIPS等顶级人工智能期刊和会议上发表突破性工作超过80篇。获得国家杰出青年科学基金、中国青年科技奖、长江学者特聘教授、国务院政府特殊津贴、百千万人才工程国家级人选和国家有突出贡献中青年专家、国际心理科学联合会青年科学家奖、万人计划科技创新领军人才、美国心理科学学会会士等多项荣誉。Current Biology专访的第一位亚洲脑与认知神经科学家。
3.From Representation to Computation: the Cognitive NeuroIntelligence
议题简介:Deep convolutional neural networks (DCNN) nowadays can match and even outperform human performance inchallengingcomplex tasks. However, it remains unknown whether DCNNs achieve human-like performance through human-like processes, that is, whether the DNNs use similar computations and representations to perform the task as humans. Here we investigated whether the DNNs and humans utilized similar representations to achieveface gender classificationby applying a reverse-correlation method to reconstruct the internal representations mediating task performance. We found that humans and a typical DCNN specialized for face identification, VGGface,showed a high similaritybetween their“classification images”, whichrevealed the critical information utilized to perform the gender classification task. Further analysis showed that both humans and the VGGfacerelied on low spatial frequencyinformation for gender classification andthey sharedsimilar representationsmainly atlow spatial frequencies. Finally, we found that the identification task that the VGGface performed might be a critical factor for such similarity in representation because Alexnet, a typical DCNN designated for object classification, did not use similar representation for gender classification as humans or the VGGface.Our study provides a novel way of illuminating the “black boxes”of the DNNs for the first time, and demonstrates that some degree of generalvisual intelligence can be realized independent of physical implementations.
演讲嘉宾:刘嘉
北京师范大学心理学部教授,美国麻省理工学院博士。主要研究领域:客体识别和知觉学习。国家杰出青年基金获得者、中科院百人计划入选者、教育部长江学者、中青年科技创新领军人才、万人计划科技创新领军人才。任《心理科学》副主编、《中国大百科全书(心理学卷)》副主编、《最强大脑》科学总顾问。在Nature Neuroscience、Current Biology、PNAS等期刊发表SCI/SSCI论文百余篇,承担多个国家级重点项目,部分成果曾获教育部自然科学奖一等奖、中华全国归国华侨联合会中国侨界贡献奖等奖励。
4.The Cross Talk between Biology and Computer Vision
议题简介:Object recognition is often viewed as a feedforward, bottom-up process in machine learning, but in real neural systems, object recognition is a complicated process which involves the interplay between two signal pathways. One is the parvocellular pathway (P-pathway), which is slow and extracts fine features of objects; the other is the magnocellular pathway (M-pathway), which is fast and extracts global features of objects. In this talk, I will introduce our recent modelling work on exploring the computational advantages associated with the interplay between the two pathways. Our study shows that the interplay between two pathways endows the neural system with the capacity of processing visual information rapidly, adaptively, and robustly. We hope that this study will not only give us insight into understanding the mechanisms of neural information processing, but also inspire us to develop new object recognition algorithms.
演讲嘉宾:吴思
北京大学长聘教授、麦戈文脑科学所研究员、北大-清华联合生命科学中心研究员。北京师范大学博士、日本理化学所脑科学分所博士后、英国萨斯克斯大学长聘副教授、中科院百人计划入选者。主要研究领域:类脑智能算法、神经信息处理。计算神经科学领域最大国际期刊Frontiers in Computational Neuroscience共同主编。在Nature, Nature Neuroscience, Neuron, PNAS, eLife等顶级神经科学期刊和Neural Computation,IEEE Trans. On Neural Networks, NeurIPS等顶级人工智能期刊和会议上发表大量原创性工作。中国自动化学会会士、中国神经科学学会《计算神经科学与神经工程专业委员会》主任。
5.From Brain Network to Brain-like Computation
议题简介:One of the key challenges for brain-like computation is to identify general principles from myriads of biological details of the brain. Here I discuss two examples of translating neuroscience discoveries to useful design of artificial neural networks. The first one is at the algorithmic/representational level, regarding how the self-organized critical state of the brain can inspire adaptive synapses in recurrent neural network (RNN) to achieve good performance and robustness at the same time. The second one is at the computational level, about how the prefrontal cortex in the primate brain can inspire novel network architecture that can work in more complex environments. These examples highlight the importance of revealing computational principles behind the structure and activity of brain networks to improve systems of artificial intelligence.
演讲嘉宾:余山
中科院自动化所研究员,中国科学技术大学博士、德国马普所和美国NIH博士后。主要研究领域:脑网络的结构与功能、类脑计算及脑机接口。中科院百人计划入选者。BMC Neuroscience等期刊编委。在Nature Machine Intelligence、eLife、Neural Networks、Cerebral Cortex等顶级期刊发表一系列开创性的研究成果。目前任模式识别国家重点实验室副主任。
- 点击阅读原文,内行盛会,首次免费注册-