Welcome to CNV! Computational Neuroscience CNV is about using computational simulations to understand how biological visual systems work. of Vision The course is geared towards people who are interested in and may be considering doing further research in Dr. James A. Bednar computational neuroscience. jbednar@inf.ed.ac.uk It may also be valuable as background for work in http://homepages.inf.ed.ac.uk/jbednar computer vision or machine learning, but those more–practical areas are not the focus. CNV Spring 2006 1 CNV Spring 2006 2 Why study vision? What does a visual system do? Claim: Vision provides an animal with potentially useful • Early stages are relatively well understood information about its environment. • Easy to control stimuli A visual system is a physical implementation of vision, and • Large percentage of brain is part of a nervous system. • Standard test case for understanding the brain Thus, a visual system provides information about the external environment to other parts of the nervous system. We will focus on animals whose visual systems are similar to humans’ (e.g. monkey, ferret, and cat). CNV Spring 2006 3 CNV Spring 2006 4
Questions about the visual system Studying the visual system (1) The visual system can be (and is) studied using many • How does it work? different techniques, including: I.e., what algorithm(s) does it perform? Psychophysics What is the level of human visual • What physical parts implement the algorithm(s)? performance under various different conditions? • What does each part do? Anatomy Where are the visual system parts located, and I.e., what do they contribute to the overall algorithm(s)? what do they look like? • How are the parts and their connections constructed? Gross anatomy What do the visual system organs (From a blueprint? By learning?) and tissues look like, and how are they connected? • How specific is the algorithm to vision in particular? Histology What cellular and subcellular structures can be seen under a microscope? CNV Spring 2006 5 CNV Spring 2006 6 Studying the visual system (2) Computational models A computational model represents a concrete Physiology What is the behavior of the component parts implementation of a theory of how (part of) the visual of the visual system? system works. Electrophysiology What is the electrical behavior of neurons, measured with an electrode? Computational models integrate information from many or all of these techniques. Imaging What is the behavior of a large area of the nervous system? Ideally, computational models will be designed to answer specific questions that arise from either existing Genetics Which genes control visual system experiments or theoretical considerations. development and function, and what do they do? CNV Spring 2006 7 CNV Spring 2006 8
Ideal model? (1) Ideal model? (2) One might imagine that the ideal computational model And also: would include all of the properties of each part of the • quantum effects visual system and its environment, in full detail. This • developing from a single cell would require simulating: • sensory environment during development • many billions of nerve cells and connections • all previous visual experience • morphology of each cell • sensory environment during testing • chemical context surrounding cells • behavioral task being performed • ionic currents • all parts of the brain that connect to the visual system • specific ion channels CNV Spring 2006 9 CNV Spring 2006 10 Ideal model? (3) Ideal model? (4) And more! Obviously, not all of these can be included in Most importantly, such an “ideal” model would be entirely any single model: incomprehensible. One could just as well say that the animal and its • We don’t have sufficient computing power environment are their own ideal model, because nothing • No one has measured the data necessary to model all has been gained or lost by using the model rather than the these things animal. • We don’t understand how the components that are Useful models will instead have to leave out most of the connected to the visual system work (i.e., the rest of properties of the “ideal” model, simplifying and abstracting the nervous system) them to focus on a few important properties. CNV Spring 2006 11 CNV Spring 2006 12
What to include in models Suitable vision models Models should be constructed to answer specific scientific Vision in mammals requires large areas of brain tissue, questions. composed of many billions of neurons and many trillions of connections. The models should then include the features of the Representations of the visual fi eld cover a large fraction of nervous system and its context that are most relevant for the brain (nearly half in monkeys). the questions being asked, modeled at an appropriate level of abstraction and detail. Disabling any single neuron, and probably a very large fraction of scattered neurons, has no significant effect. Everything else should be omitted. Thus, for understanding vision I believe it is crucial to model large numbers of neurons, rather than modeling single neurons or parts of neurons extremely well. CNV Spring 2006 13 CNV Spring 2006 14 Types of models in this course Course topics This course will thus focus on modeling the visual system Biological background: Visual systems in animals and at a level large enough to see interesting behavior, humans simulating regions at least several square mm in size Computational modeling levels: Single-unit models, (large enough to represent recognizable visual features). topographic map models I will generally draw examples from my own research and Modeling early visual system development: How brain related research, because those are the ones that I can areas organize cover most authoritatively (and provide working code for!). Modeling adult low-level processing: Edge detection, At the same time, the tools we will use and the overall orientation processing, etc. approaches are very general, and can accomodate a very large range of other possible models. Higher-level processing: Face and object processing CNV Spring 2006 15 CNV Spring 2006 16
Course components Topographica simulator The course will consist of lectures, readings, coursework, Course exercises will use the very-recently-released and exams. Topographica simulator, currently under development here in my laboratory. The coursework and exams assume a basic level of mathematical and programming competence, but the The simulator still has plenty of rough edges, but it should focus is on the biology throughout. be usable for the tasks we’ll be doing in this course. Prior experience with Python and CVS is helpful, but not Anyone who wants to extend the simulator or make required. improvements is welcome to join the Topographica developers group by sending me your SourceForge.net username. CNV Spring 2006 17 CNV Spring 2006 18 Text Summary The text for the course is our recently published book • Vision is an important test case for the brain Computational Maps in the Visual Cortex (Springer, • Computational models at the right level can be used to September 2005), obtainable by: understand vision • Ordering from Springer or Amazon (60-70 GBP) • We will study the basic architecture and function of • George Sq. library (including 1 on reserve) early vision, and how they can be investigated using • DTC student offices (2 copies) simulations The book is a synthesis of research from four authors • Few of the techniques are specific to vision, and thus using closely related models to study the visual system. they can also be applied to other systems All of the simulations in the book are my own except for those in chapters 3 (SOM) and 11-14 (perceptual grouping). CNV Spring 2006 19 CNV Spring 2006 20
Recommend
More recommend
