第一章绪论

神经信息学
Neuroinformatics
Spring semester, 2009
LECTURE 1
Introduction
武志华
·
中科院生物物理所,
脑与认知科学国家重点实验室,
副研究员
·
Tel:
64869355
Email: [email protected]
史忠植
· 中科院计算技术研究所,
智能科学实验室,
研究员
· 人工智能 (神经计算)
教学按排
每周四: 8:50-11:40
40 学时, 2 学分
3学时/每次
地点:玉泉路园区 305
闭卷笔试
笔试内容与作业的关系
引言
- 1994年,汪云九老师在中科院研究生院开设“神经信息学”
课程。
- What is Neuroinformatics?
What is
I.
Neuroinformatics?
II. Computational Neuroscience (计算神经科学)?
III. Theoretical Neuroscience?
IV. Neurocomputing (神经计算)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Neuroinformatics is a research field including the
development of:
1. databases of neuroscience data,
2. tools for management, sharing, analyzing and
modeling of neuroscience data at all levels of
analysis,
3. computational models of the nervous system and
neural processes
Neuroscience
NeuroInformation
informatics
science
Vast amounts of diverse data about the
brain was gathered
(汪云九老师)
Human Brain Project
(A name similar to Human Genome Project)
Idea emerged in 1991: Mapping the brain and its functions. Integrating enabling
technologies into neuroscience research
1993年“人类脑计划(Human Brain Project)”的第一批项目公
布,标志着人类脑计划正式启动
Neuroinformatics uses databases, the Internet, and visualization in the
storage and analysis of the neuroscience data
SenseLab (http://senselab.med.yale.edu)
1
3
2
BrainMaps.org (http:// brainmaps.org)
Explore BrainMaps data in 3D
Neuroinformatics
=
Databases + Tools + Computational Models
What is
I.
Neuroinformatics?
II. Neurocomputing (神经计算)?
III. Theoretical Neuroscience?
IV. Computational Neuroscience (计算神经科学)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Neurocomputing is concerned with processing information:
1. It involves a learning process within an artificial neural
network architecture
2. The trained networks can be used to perform certain tasks
depending on the particular application
3. Neurocomputing can play an important role in solving
certain difficult problems in science and engineering such
as pattern recognition, optimization, event classification,
control and identification of nonlinear systems, and
statistical analysis
I. What is Neuroinformatics?
II. Neurocomputing (神经计算)?
III. Theoretical Neuroscience?
IV. Computational Neuroscience (计算神经科学)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Theoretical Neuroscience =
Computational and Mathematical Modeling of
Neural Systems =
Computational Neuroscience
What is
I.
Neuroinformatics?
II. Neurocomputing (神经计算)?
III. Theoretical Neuroscience?
IV. Computational Neuroscience (计算神经科学)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Computational neuroscience is a subfield of
neuroscience that uses mathematical methods to
simulate and understand the function of the
nervous system
Dynamical
system
Computational
neuroscience
Artifical
NeuroNeural
science
networks
(http://www.scholarpedia.org)
Computational neuroscience
• A family member of brain science. Computer simulations of neurons and
neural networks are complementary to traditional techniques in
neuroscience
• Theoretical analysis and computational modeling are important tools for
characterizing what nervous systems do, determining how they function,
and understanding why they operate in particular ways
• Neuroscience encompasses approaches ranging from molecular and cellular
studies to human psychophysics and psychology
Theoretical neuroscience encourages cross-talk among these subdisciplines
by constructing compact representations of what has been learned, building
bridges between different levels of description, and identifying unifying
concepts and principles
First neuron model - McCulloch & Pitts
model (1943)
Time: t
t+1
Range: (0, 1) or (-1, 1)
(Bulletin of Mathematical Biophysics 5:115-133)
MP model simulates a few properties
• The unit has two states depending on the threshold:
rest or activated
• Two types of synapses:
inhibitory and excitatory
• The unit receives the linear sum of all the pre-synaptic inputs
• The introduction of time, mimicking the synaptic delay
Advantage:
Be able to perform logic operations
Shortcoming: Too simple to model the real neuron
Goal
1. The first goal is to teach WHY mathematical and
computational methods are important in understanding
the structure, function and dynamics of neural
organization
2. The second goal is to explain HOW neural phenomena
occurring at different hierarchical levels can be described
by proper mathematical models
What is
I.
Neuroinformatics?
II. Neurocomputing (神经计算)?
III. Theoretical Neuroscience?
IV. Computational Neuroscience (计算神经科学)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Why not go out for a walk?
I mean the current neuroscience world is a little different
from before
For example:
The Age of Brain Science (Japan)
•
•
•
•
Understanding the Brain
Protecting the Brain
Creating the Brain
Nurturing the Brain
(October 1997--- March 2008 )
If you are a student majoring in experimental biology,
- you may figure out your experimental problem or
analyze results in a different way
If you are a student having good mathematical or
physical background,
- ??? in future
About mathematical and physical FORMULA:
-High school knowledge is enough
Caveat
Computational neuroscience is a huge and fast developing
area
This is only a very short course, and hopefully it can provide:
- The basic concepts and methods of computational
neuroscience research
- Some brief introduction to neurobiological concepts and
mathematical techniques. The techniques will be applied for
describing the behavior of several brain regions
What is
I.
Neuroinformatics?
II. Neurocomputing (神经计算)?
III. Theoretical Neuroscience?
IV. Computational Neuroscience (计算神经科学)?
V. Why we learn “Neuroinformatics” or
“Computational Neuroscience” ?
VI. Course structure
Course Structure
0. Introduction: What is computational neuroscience,
why is it urgently needed
1. Single neuron models
2. Neural network models
3. Neural Coding
4. Synaptic plasticity and learning
5. Hot points in brain modeling
6. Neurocomputing (by 史忠植老师)
Recommended Readings
1. Thomas P. Trappenberg: Fundamentals of Computational
Neuroscience. Oxford University Press, 2002
2. Peter Dayan & Larry F. Abbott: Theoretical Neuroscience.
MIT Press, Cambridge. 2001
3. Koch C & Segev I: Methods in neuronal modeling: from
ions to networks. MIT Press, Cambridge, 1998
4. 汪云九等著,神经信息学. 北京:高等教育出版社. 2006
5. 郭爱克著, 计算神经科学. 上海科技教育出版社. 2000
6. 史忠植,智能科学. 清华大学出版社,2006
7. Jeff Hawkins and Sandra Blakeslee, On Intelligence.
Times Books 2004
人工智能的未来. 贺俊杰等译, 陕西科学技术出版社,
2006
8. F. Crick(汪云九等译),<惊人的假说>,湖南科学技术
出版社,长沙,1998
作业及思考题
1 神经信息学主要包括哪些内容?
2 什么是计算神经科学?计算神经科学与神经计算的区别?
3 第一个形式人工神经元模型是什么?模拟了神经元的哪
几个性质?