2022年北京智源大会召开在即，5月31日至6月2日，持续三天，26场由各领域领军学者主导的专题论坛。大会将紧紧围绕这些当前学术领域迫切需要解决的问题，以及产业落地过程中存在的诸多挑战，延续一贯以来的“内行认可”品质口碑，分享真正内行认可的重大成果与真知灼见，献上一场诚意满满的AI盛宴！目前已正式开放大会线上报名渠道。大会将同步向全球线上直播。

北京智源大会倒计时：3 天

论坛议程

论坛主席

马雷，北京智源人工智能研究院生命模拟中心负责人、副主任, 北京大学国家视频与视觉技术国家工程研究中心副研究员

浙江大学学士、上海交通大学硕士、中科院软件所计算机科学国家重点实验室博士。曾于香港中文大学访问研究。在TVCG、TOG、NEURUAL NETWOR、CVPR等会议期刊发表学术论文近30篇，拥有多项国际国内专利。早年曾任国际软件企业Autodesk研究实验室高级工程师，曾联合创办增强现实企业并长期在国内多个企业担任技术顾问。最近几年致力于从事大规模精细神经元网络计算，生物医学可视化、脉冲视觉研究。

演讲主题及嘉宾介绍

1、The role of theory in understanding the degenerate mechanisms controlling circuit dynamics

议题简介：Theoretical and experimental studies on the dynamics of small circuits have revealed that similar circuit dynamics can result from appreciably different circuit mechanisms.  These different circuit mechanisms can be “cryptic”, or hidden, until they are revealed by perturbations of the circuit.  Smooth transitions between different dynamical mechanisms can occur, as well as abrupt transitions in state.  Parallel pathways and electrical coupling can confound understanding how information travels in circuits.  Small circuits can provide a set of “building blocks” that may be useful in thinking about the properties of large circuits, but new theory will be necessary to ask whether new principles arise in large circuits that are not implicit in the potential of degenerate dynamics seen in small circuits.

Eve Marder，布兰迪斯大学教授，美国国家科学院院士

Dr. Marder was President of the Society for Neuroscience in 2008. She is a member of the National Academy of Sciences, the Institute of Medicine, the American Academy of Arts and Sciences, a Fellow of the Biophysical Society and a Fellow of the American Association for the Advancement of Science. She received the Miriam Salpeter Memorial Award for Women in Neuroscience, the W.F. Gerard Prize f-rom the Society for Neuroscience, the George A. Miller Award from the Cognitive Neuroscience Society, the Karl Spencer Lashley Prize from the American Philosophical Society, an Honorary Doctorate from Bowdoin College, the Gruber Award in Neuroscience, the Distingushed Alumni Award from Brandeis University, the Society for Neuroscience Education Award, the 2016 Kavli Award in Neuroscience, an Honorary Degree in 2017 from Tel Aviv University, and the 2019 Neuroscience Award from the National Academy of Sciences.

2、Organizing Behaviors across Timescales

议题简介：Animal behaviors are complex and hierarchical spatiotemporal patterns. In the popular model organism Caenorhabditis elegans, behavioral sequences on a slow timescale emerge from ordered and flexible transitions between different motor states, such as forward movement, reversal, and turn. On a fast timescale, rhythmic bending sequences originating from the head are propagated along the body during a forward run. Fast dynamics are generated by low level motor circuitry, while slow dynamics arise from recurrent interactions between high level interneurons contributing to behavioral decision-making. By bringing together quantitative behavioral analysis, whole brain calcium imaging, and genetic approaches, I will present functional circuit motifs and computational algorithms that dictate slow and fast motor sequences, highlighting a pair of sensory/inter neurons that play a dual role in shaping slow and fast dynamics at the end of the talk. I will discuss and speculate how hierarchical behavioral patterns are organized by the neural circuit in total.

温泉，中国科学技术大学生命科学与医学系教授，博士生导师

Dr. Quan Wen is Professor in the Division of Life Sciences and Medicine, University of Science and Technology of China. Dr. Wen did his thesis research at Cold Spring Harbor Laboratory, supervised by Dmitri Chklovskii, and received his PhD degree in physics from Stony Brook University, New York in 2007. He was an associate at Janelia Research Campus, HHMI from 2007-2009, and a postdoc working with Aravi Samuel in the Department of Physics and Center for Brain Science at Harvard University from 2009-2014. Starting from 2014, he has been a principal investigator in the Life Sciences Department at USTC. Dr. Wen is interested in identifying basic principles for motor control and computational algorithms for sensorimotor transformation. His lab has been studying small animals such as C. elegans and larval zebrafish, whose nervous systems are relatively compact and transparent. These advantages allow his team to develop optical and computational methods for reading whole brain activity in unrestrained animals, holding the promise of bringing new insights into dynamic brain and naturalistic behaviors.

3、Can we reverse-engineer the simple brain of a tiny worm? 

议题简介：Reverse-engineering a biological brain is a grand challenge in neuroscience and AI research which requires a comprehensive understanding of the brain at every level of complexity. If this goal is attainable at all, it probably will first be achieved in the nematode worm Caenorhabditis elegans with only 302 neurons and a complete wiring diagram (“Connectome”). However, all previous modeling attempts were unable to generate a unifying hypothesis that explained how the worm brain works. One of the critical factors for this failure was the lack of knowledge of the intrinsic biophysical properties of most neuronal cell types in C. elegans. A compelling example is that, for over 30 years, the scientific community has incorrectly assumed that the nervous systems of C. elegans and other nematodes are strictly analog and do not have action potentials required for spike coding. Our recent work, however, identified the first neuronal action potential and spike-coding schemes in olfactory neurons AWA in C. elegans, marking a shift in our understanding of information processing in the C. elegans nervous system to inform future modeling. More recently, we further identified and characterized three additional spiking neurons including an interneuron AIA, and the GABAergic enteric motor neurons AVL and DVB. These neurons fire action potentials with characteristic features and waveforms that are distinct from those in AWA and each other, suggesting different underlying biophysical mechanisms and physiological functions. An emerging picture from these results is that instead of being the outlier organism that does not use the universal language of neuronal action potentials, C. elegans possess a plethora of diverse spiking neurons to implement specific computations. Thus, C. elegans is poised to be the ideal model system to understand how animals choose one form of coding strategy over the other for specific behavioral tasks. We are currently working toward an “electrophysiome” - the comprehensive electrophysiological characterization of every neuronal class of the entire C. elegans nervous system. It is our hope that integrating the complete electrophysiome and the complete connectome would be a crucial step towards predictive whole-brain modeling in C. elegans.

刘强，香港城市大学神经科学系助理教授

Dr. Qiang Liu was originally trained as a medical doctor in Basic Medical Science at Beijing Medical University, but later pursued an academic career in neuroscience. At the University of Toronto, Dr. Liu started his graduate training in electrophysiology in Dr. Xian-min Yu’s lab at the Center for Addiction and Mental Health (CAMH), studying phosphorylation regulation of NMDA receptors using cultured hippocampal neurons in mice. After graduate school, Dr. Liu was attracted to the simplicity and tractability of the model organism C. elegans and joined Dr. Zhao-wen Wang’s lab at the University of Connecticut Health Center and Dr. Erik Jorgensen’s lab at the University of Utah for his postdoctoral training. During this period, Dr. Liu published a series of papers on synaptic regulation, gap junction and ion channel function, and vesicle trafficking. Dr. Liu was later recruited by Dr. Cori Bargmann to the Laboratory of Neural Circuits and Behavior as a Research Assistant Professor at the Rockefeller University where he discovered the first neuronal action potential and spike-coding scheme in C. elegans. In 2022, Dr. Liu joined the Department of Neuroscience at the City University of Hong Kong as an Assistant Professor to work on the “Electrophysiome” project, which aims to systematically characterize the intrinsic biophysical properties of every neuronal cell type in the C. elegans nervous system. Dr. Liu’s lab also investigates fundamental questions in neuroscience such as how external stimuli are encoded in the olfactory circuit to dictate goal-oriented behaviors and how intrinsic neuronal oscillations are generated by the gut-brain axis underlying behavioral rhythms.

4、Dendrites lie at the crux of a customizable cortical computing architecture

议题简介：The cerebral cortex is thought to consist of a single basic architecture that is largely conserved across different cortical areas and species. Yet, how is it possible for a single computing architecture, that grows from a seed, and trains itself, to function as a state-of-the-art image processor, sound processor, tactile processor, chemical spectrum analyzer, sensory-motor controller, action planner and language processor, and seat of thought and consciousness? What are the "parameters" of the cortical circuit that allow for this remarkable degree of power and flexibility? To make forward progress on this issue, we have adopted an approach that tries to answer two questions, in constant alternation: (1) "What are the hardware components of the cortex (i.e. dendrites, neurons, local circuits, etc.) *capable* of computing?", and (2) "What real world problem(s) is a particular neuron or local circuit asked to solve?" Focusing on a particular, biologically-relevant, vision problem – object boundary detection in natural images – we discuss results of biophysical simulation studies, analyses of natural image statistics, and classical and modern neurophysiological studies, that suggest that the cortical pyramidal neuron and its elaborate dendritic tree lies at the nexus of a highly flexible nonlinear information processing architecture having little in common with the architectural motifs of state-of-the-art "deep networks". In particular, our results suggest that two core premises in the DN field -- that neurons are adequately described by simplified abstract models such as the ReLU, and that the functional interconnections between neurons are adequately described by scalar synaptic weights, are both in need of revision.

Bartlett Mel，南加州大学生物医学工程系副教授，神经科学博士项目成员

Bartlett Mel于1989年获得UIUC计算机科学博士学位。曾在加州理工学院Christof Koch实验室担任博士后，期间他将研究方向转向计算和理论神经科学领域，并开始研究涉及树突计算的问题。他的研究重点是树突、神经元及神经环路的信息处理能力，以及这些机制在视觉和记忆等领域的众多应用。

5、圆桌讨论：从生命智能迈向通用人工智能

圆桌嘉宾：

马雷 | 北京智源人工智能研究院生命模拟研究中心副主任

杜凯 | 北京大学人工智能研究院助理研究员

吴迅冬 | 北京智源人工智能研究院生命模拟中心研究员

温泉 | 中国科学技术大学教授

刘强 | 香港城市大学助理教授

陶乐天 | 北京大学研究员

臧蕴亮 | 浙江大学生物医学工程博士

圆桌嘉宾详细介绍

杜凯，北京大学人工智能研究院助理研究员

博士，北京航空航天大学飞行器动力工程系学士（2002），瑞典卡罗琳斯卡医学院神经科学系博士（2016）和博士后（2020），欧盟脑计划“大脑仿真平台” 瑞典团队主要成员(2008-2016)，于2020年加入北京大学人工智能研究院任助理研究员。他建立了领域内第一个中等树突棘神经元（medium-spiny neurons）的精细模型(Du,et.al.,PNAS,2017),并在此基础上 发展了基底核脑区的纹状体精细模型。研究工作主要发表在PNAS，J.neuroscience，PloS Comput. Bio. 等领域内一流刊物上。主要研究方向是大脑精细仿真，树突计算，以及基于大脑精细模型的新型人工智能系统和理论。

吴迅冬，北京智源人工智能研究院生命模拟中心研究员

南加州大学博士。主要研究方向为神经网络理论，计算神经科学及轻量化神经网络。他的重要工作有：使用权重隐状态避免神经网络学习中的灾难性遗忘现象，从而极大提升神经网络容量；首先提出了用于神经网络计算加速的细粒度结构化稀疏架构，这一结构被英伟达在他们的A100, H100 GPU中用于加速神经网络的推理计算。

陶乐天 ，北京大学研究员

哈佛大学学士，芝加哥大学博士，长期从事计算神经科学研究，运用数学和大规模计算的方法去模拟、分析神经生物学中神经元网络的动力学性质及其对生物个体感知和认知功能的贡献。早期注重研究哺乳动物视觉系统的网络模型，及其数学描述、简化和降维。最近几年致力于模式动物神经环路和行为的光学成像，发展跨时空尺度的计算神经解析模型。

臧蕴亮，Brandeis University助理研究员

师从美国科学院院士Eve Marder教授，计划下半年加入天津大学医学部医学工程与转化医学研究院担任英才教授。代表性工作发表在PNAS、Cell Reports、ELife和Journal of Neuroscience等神经科学领域Top期刊上。近年来一直应用计算模型探索神经元特性对脑信息处理的关键作用，具体研究兴趣包括：树突编码学习、小脑学习机制、神经系统鲁棒性机制、记忆形成和消除机制（果蝇）、类脑计算等。