Lecture 2: Basics and definitions

1
Computational Neuroscience Andy
Philippides Centre for Computational
Neuroscience and Robotics (CCNR)
COGS/BIOLS andrewop_at_cogs.susx.ac.uk Spring 2003
2

• Teaching
• 2 hours per week Monday 9.15-11.05
• Nominally 1 hour lecture, 1 hour seminar, but
  may vary
• Office hour Friday12.30-1.30, BIOLS room 3D10
• Lecture notes available online soon

3
AssessmentSeminar presentation 25

• 30 mins presentation (25 mins 5 mins questions)
  
• Papers chosen from the reading list. Papers of
  your own choosing may be acceptable but MUST be
  ratified by me
• Presentations start in week 4 (order decided by
  subject)
• Present the ideas of the paper clearly
• Give your opinion on the strengths and weaknesses
  of the ideas/research
• Be prepared to answer questions on the paper
• A little reading around the subject will usually
  be required for a good presentation
• A printout of any resources used must be given to
  me for marking and external assessment

4
Assessment Project 75

• Programming project to implement a computational
  neuroscience model
• Due in Wed 23rd April 12 noon
• Approx 3500 words
• Some suggested projects but you can choose your
  own subject to ratification by me
• More details in week 5

5
Course summary
Computational Neuroscience aims to understand the
mechanisms underlying brain function by building
quantitative models.  The course is intended to
introduce the basic concepts of this area and
give details of some of the standard models and
approaches Biological background will be given
together with the model though some basic
knowledge is assumed Some mathematics will be
required as the majority (all?) of the models are
mathematical, but I will attempt to keep it to a
minimum
6
Dont Panic about the Maths!

• You only need to know enough to understand what
  is going on not how to do it
• Maths is necessarily abstract and may not be
  clear at first, but this is to be expected
• It may take time (several viewings) to understand
  things so be patient

7
Caveat

• Computational neuroscience is a huge topic and
  this is a short course
• The lectures will not cover everything
• Will try to cover the basics of a subject and
  give pointers to more advanced topics/areas of
  interest

8
Course Structure 1
1. Course introduction What is computational
neuroscience, why is it needed, levels of
modelling, neural signalling 2/3. Single neuron
models (start small and work up) Basics of
neural signalling, membrane equation, cable
theory, action potentials, Hodgkin-Huxley model,
beyond HH model 4. Networks of neurons what
neuron models are used, how they are connected,
oscillations in networks of neurons, map formation
9
Course Structure 2
5. Learning modelling synaptic learning, how
learning shapes neural networks 6. Spiking
networks Models of spiking neurons and issues of
spike timing and coincidence detection 7. Gaseous
neurotransmission mathematical models of
diffusion, more abstract models of diffusion 8.
Systems level neuroscience. Examples of higher
level models (if we have enough time)
10
Further reading
Purves, et al eds. Neuroscience. Sinauer, 97
(many neuroscience texts at various levels pick
one thats right for you) Abbott LF Dayan P
Theoretical Neuroscience. MIT Press,
2001 Computational Neuroscience. Realistic
Modeling for Experimentalists By Erik De
Schutter. CRC Press, 2000 Koch C Segev I.
Methods in neuronal modeling from ions to
networks. MIT Press, 1998 Spikes, Rieke et al,
1996 Many others see http//home.earthlink.net/
perlewitz/books.html
11
What is computational neuroscience?

• Mathematical modelling the construction of
  quantitative models to understand observable
  phenomena. Explaining phenomena in terms of
  underlying mechanisms.
• Computational modelling Modelling what the brain
  does in terms of computations. Crudely, trying to
  understand the brain as a computing device a
  rather newer idea (eg Info theory etc)
• Relation to AI/ALife These try to understand
  computing devices in general and how best to
  solve computational problems. CN studies a
  particular computing device and how it solves
  problems. Sometimes uses same tools (eg neural
  nets) but one must be careful. Also biology
  provides great deal of inspiration for Alife
  techniques

12
Why do we need models?

• When we have enough data about the brain, won't
  we understand how it works? Analogy with
  astronomy.
• Common misuderstanding Modelling is a form of
  hypothesis testing.
• Force one to make assumptions explicit. Can only
  get so far with hypotheses expressed in intuitive
  terms. E.g. visual experience affects visual
  development''.
• Enables many virtual'' experiments to be done,
  can pinpoint the one that is most crucial.
• Can lead to unexpected predictions.
• Often much quicker/easier to try out ideas eg
  lesioning studies in silica rather than in
  vitro/vivo so can guide potential experiments

13
What makes a good model?

• However, its easy to make a bad Alifey model
• Good to have close contact with neuroscientists
• (Maybe) Model should not only replicate existing
  data but must also make predictions about the
  biological system Editor of biological journal
  (??! Discuss)
• Roger Quinn cockroach models normally only
  informative when they dont work

14
Levels of modelling

• Many different types of models continuum from
  very realistic to very abstract. All models must
  make simplifications to be useful.
• E.g. model of single neuron.
• Binary threshold unit
• Continuous unit
• Integrate-and-fire (continuous in time)
• Spiking
• A few compartments
• Many compartments
• Individual channels
• Detailed model of channel dynamics
• etc

15
Which to use?

• Depends on purpose of the model! Different types
  appropriate for different sorts of questions.
  Should become clearer as we learn about
  particular types of models in the course.
• Eg Top-down vs bottom up models.
• Top-down Start with idea about abstract task /
  problem, figure out good way to solve it, see if
  that's what the nervous system does.
• Bottom-up Look closely at nervous system, try
  and figure out what its doing, derive the task /
  problem from there.

16
Models can be categorised in other ways

• Biological perspective (Molecular, neuronal etc)
• Marr (82)
• Computational (what computation is to be
  performed, in terms of optimality, modularity
  etc),
• algorithmic (looks at nature of computations
  performed)
• implementational (how algorithm is implemented)
• Abbot and Dayan (2001)
• Mechanistic based on known anatomy and
  physiology
• Descriptive summarizes large amounts of
  experimental data
• Interpretive explores behavioural and cognitive
  significance of nervous system function