require './support.php'; ?>
G89.3405 - Introduction to Computational ModelingMeeting Time/Place: MW 4-5:50pm, Meyer 851
Instructor: Todd M. Gureckis
Office: Meyers 280
Office Hours: By appointment
Computational modeling plays an increasingly important role in the social and behavioral sciences. This introductory course provides a broad survey of computational approaches to human behavior. Topics will be organized around interests of students in class, however, the core concepts we will cover are the goals and philosophy behind developing models and basic issues in model evaluation, testing, and fitting. Readings and lectures will survey a broad set of approaches to modeling cognitive processes with an emphasis on what are traditionally considered "higher-level" cognitive processes. Example topics include reviews of the basic properties (and limitations) of artificial neural networks/parallel distributed processing, contemporary approaches to modeling memory, learning, and decision making processes, modeling of reaction time data, developmental approaches (i.e., dynamical field theory, etc...), models of categorization, reasoning, problem solving, analogy, etc..., approaches to integrating models and findings from cognitive neuroscience (i.e., what can they tell each other), the relative merits of bayesian/rational approaches and mechanic models (a bit of modeling philosophy), other topics might include a segment on agent-based models of socio-behavioral processes (i.e., models based on interactive, distributed processing by independent components).
In other words, we'll aim to cover a relatively broad set of topics in formal modeling. In an ideal world, everyone would leave the course with a richer understanding of the role that computational model plays in contemporary cognitive science, understand how to fit/evaluate models, and how to read a modeling paper, think about the predictions it makes, and perhaps even implement it yourself.
This is not a quantitative course. For the most part, the focus will be developing an understanding and intuition for concepts as they relate to human behavior relative to a math skills course. If you have taken Math Tools in the psych department, or had linear algebra or calculus as an undergrad you will be in the best position for approach the material. However, we will, when needed, review some of the basic concepts needed to understand the assigned papers. For the (infrequent) homework/assignments, I will generally assume some basic familiarity with programming in something like Matlab or Python (if you know what a for loop is you'll be fine), however, if you are interested in the above topics and the programming is the main hang up, please consider enrolling. An effort will be made to adjust the assignments to people's individual (and hopefully diverse) backgrounds as much as possible.
There is no textbook to purchase for this course. Readings will be selected research articles and book chapters
provided via this website (password to access the files will be provided in class).
Graduate standing or permission of instructor.Grading:
Grading will be based on attendance/participation in lectures, a small number of homework assignments, and a final project. The final project will be a modeling project of the student's design or a proposal for such a project written as a NRSA grant proposal. The projects will be presented at the end of class and will be evaluated by the peers in the class (akin to a mini grant panel). The expectation is that the final project will apply some idea from the course to the student's own research background/project.
Attendance/participation: 40%, Final Project: 20%, 2-3 Homeworks 30%Python tips:
By popular demand, here are a collection of helpful resources. Note please stick with python 2.5 (i'm using 2.5.4 on mac). Python 2.6 are in a bit of a transition and I'm unsure of the compatibility status of scipy/numpy (libraries we will want to use).
|Jan 21, 2009||
Welcome, Course Policies, General Overview,
Pre-Course Survey - Please fill out and email to me
|Jan 26, 2009||
Week 1: Introduction to Cognitive Modeling
Lecture 1: What are the goals of modeling?In the first week, we take a first step at thinking about the role of models in cognitive science. We'll read some classic papers from the start of the cognitive revolution that argued for why computational/mathematical models are necessary in behavioral sciences. We'll discuss some of the primary goals of modeling (i.e., accounting for past data, formalizing theories of cognitive function, and making novel predictions).
Some Fun and some Beautiful Examples
|Jan 28, 2009||
Week 1 (cont): Introduction to Cognitive Modeling - Types of Models, and "Thinking in Levels"
Today we'll take an overview of types of models. In addition, I'll be walking through a bit of what is required for the first homework. Please read the following papers before class (all are interesting but you might focus on the ones that match you background/research interests)
Homework 1: Hands on with a complex, interactive model
This homework we will play with a couple hands-on modeling simulators I developed for a undergraduate class. However, I think they will be useful in getting people to think about how the goal of cognitive modeling is to understand complex processes which are often hard to understand/predict without actually simulating them on a computer.
- homework 1 description *note: some parts of step III are just copied from my undergraduate course notes because i lost the original word file i used to make the PDF. Skip the steps about writing your own lesion code, and the stuff about turning in a screen shot of a input pattern. There is nothing to turn in for this really, just experiment on your own (maybe see how to run a python script).
|Feb 2, 2009||
Week 2: Evaluating Models
The next two classes, we will start to get our hands dirty and focus on some pragmatic issues of model evaluation and testing. Basically, how do we assess if a model is a "good" account for our data? There are both qualitative (i.e., does the model make reasonable psychological assumptions?) and quantitative answers to this kind of question (i.e., if model A fits better than model B we should prefer model A). However, model evaluation is a complex statistical question. The readings focus on currently accepted model evaluation technique in journals catering to psychology/social science.
Code for Exemplar/Prototype Model!
Python script for exemplar/prototype modesl
|Feb 4, 2009||
Week 2: What should we compare our models to? The group or the individuals?
How do these choices influence the
inferences we make about cognitive processes?
An example (and recent debate) about inferences to group or individual data:
A solution!!! (advanced reading, but breeze through for basic idea)
Homework 2: Model fitting methods
Code for Exemplar/Prototype Model PLUS code for doing Nelder-Mead Simplex Parameter Searches
Python script for fitting exemplar model using RMSE.
|Feb 9, 2009||
Week 3: Connectionist/PDP Approaches - Lecture
This week begins the first segment focusing on particular types of models in more detail and the psychological literature which they address. In particular, we begin somewhat in the middle of the cognitive revolution with the advent of parallel distributed processing (PDP), also known as connectionist modeling, neural networks, etc.. In a couple lectures we'll review some of the basic properties of neural network, and discuss some early and important papers.
|Feb 11, 2009||Week 3 (cont): Connectionist/PDP Approaches (cont) - Lecture|
|Feb 16, 2009||No Class, President's Day!|
|Feb 18, 2009||
Week 4: Derivation of the Back prop algorithm (including how to implement it in python)
*note about mini homework*: it is true that you need a bias unit or step function in your neurons to solve the task.
|Feb 23, 2009||
Week 5: Models of Category Learning
Fresh off the debates last week about the relative merits of the PDP approach, we take a look at a particular domain in which a variety of different modeling approaches have been developed. In particular we are going to read three papers. Each account for the same data, but each make substantially different claims about the nature of cognitive representation
Medin, D. L. and Schaffer, M. M. (1978). Context theory of classification learning. Psychological Review, 85, 207-238.
Nosofsky, R. M., and Palmeri, T. J., and Mckinley, S. C. (1994). Rule-plus-exception model of classification learning. Psychological Review, 104, 266-300.
Love, B.C., Medin, D.L, & Gureckis, T.M (2004). SUSTAIN: A Network Model of Category Learning. Psychological Review, 111, 309-332.
|Feb 25, 2009||
Week 5 (continued): Models of Category Learning
|Mar 2, 2009||
Week 6: Bayesian Approaches
Griffiths, T. L., & Yuille, A. (2008). A primer on probabilistic inference. In M. Oaksford and N. Chater (Eds.). The probabilistic mind: Prospects for rational models of cognition. Oxford: Oxford University Press. (manuscript pdf)
Hemmer, P. & Steyvers, M. (2008). A Bayesian Account of Reconstructive Memory. In V. Sloutsky, B. Love, and K. McRae (Eds.) Proceedings of the 30th Annual Conference of the Cognitive Science Society. Mahwah, NJ: Lawrence Erlbaum
|Mar 4, 2009||
Week 6: Bayesian Approaches Continued (approximation methods, MCMC)
|Mar 9, 2009||
Week 7: Rational vs. Mechanistic Approaches
|Mar 11, 2009||Week 7 (cont): Models of Memory|
|Mar 16, 2009||No Class, Spring Break!|
|Mar 18, 2009||No Class, Spring Break!|
|Mar 23, 2009||
Week 9: Science Day
Instead of class, please attend the CNS Science day on models and experiments: CNS ScienceDay
A note about presenting a paper. Some of these papers can be complex and have tons of simulations and data (As nick now knows!). More than we can have time to talk about, and more than anyone could process in 20 minutes or so. The best hope is that the paper gives someone some pointers into a literature they were interested in, and the basic background for further reading. To make things most effective, the best thing would be to present some basic background. What is the model developed for? What are the major assumptions going into it? Then talk about the data it is intended to account for. Why is this data the best place to test out a computational model? What hole exists in the literature that this model fills? Next, describe the basics of the operation of the model. Equations are useful when they contribute to understanding but not strictly necessary. We just want to have a feel for why that model predicts what it does. Then, choose a couple of the key behavior effects the model explains, and try to explain why this naturally comes out of the model. Finally, what is wrong with the model (definitely something). What assumptions are unreasonable? How might the model be applied in some other domain? What are the best things about the model? Hopefully these final questions can hope provoke some discussion. Slides are not necessary but could be helpful in some cases.
|Mar 25, 2009||
Week 9 (cont): Models and Cognitive Neuroscience (processes models as data analysis)
Busemeyer, J. R. & Stout, J. C. (2002) A Contribution of Cognitive Decision Models to Clinical Assessment: Decomposing Performance on the Bechara Gambling Task. Psychological Assessment, 14, 253-262 - Todd
Nathaniel D. Daw, John P. O'Doherty, Peter Dayan, Ben Seymour & Raymond J. Dolan (2006). Cortical substrates for exploratory decisions in humans. Nature, 441, 876-879. - Bob
Anderson, J. R., Albert, M. V., & Fincham, J.M. (2005) Tracing Problem Solving in Real Time: fMRI Analysis of the Subject-Paced Tower of Hanoi. Journal of Cognitive Neuroscience, 17 1261-1274. - Craig
|Mar 30, 2009||
Week 10: Models and Cognitive Neuroscience (cont.) - implications of models for cog. neuro findings
Nosofsky, R. M., & Zaki, S. R. (1998). Dissociations between categorization and recognition in amnesic and normal individuals: An exemplar-based interpretation. Psychological Science, 9(4), 247-255. - John
Ashby, F.G. and Alfonso-Reese, L.A. and Turken, A.U. and Waldron, E.M. (1998) A Neuropsychological Theory of Multiple System in Category Learning. Psychological Review, 105 (5), 442-481. - John
Busemeyer, J. R., Jessup, R. K., Johnson, J. G., & Townsend, J. T. (2006). Building bridges between neural models and complex decision making behavior. Neural Networks, 19, 1047-1058. - Doug
|Apr 1, 2009||
Week 10: Models of Object Detection and Perception
Ullman, S.(2006). Object recognition and segmentation by a fragment-based hierarchy. Trends in Cognitive Science, 11(2), 58-64. - Mordechai
Ullman, S. and Soloviev (1999). Computation of pattern invariance in brain-like structures. Neural Networks, 12, 1021-1036. - Mordechai
Hummel, J. E., & Biederman, I. (1992). Dynamic binding in a neural network for shape recognition. Psychological Review, 99, 480-517. - Todd
|Apr 6, 2009||
Week 11: Learning and Perception
Goldstone, R. L. (2003). Learning to perceive while perceiving to learn. in R. Kimchi, M. Behrmann, and C. Olson (Eds.) Perceptual Organization in Vision: Behavioral and Neural Perspectives. Mahwah, New Jersey. Lawrence Erlbaum Associates. (pp. 233-278) - Keyong Jin Tark
|Apr 8, 2009||
Week 11: Modeling Higher Level Cognition (semantic memory)
Steyvers, M and Tenenbaum, J. B. (2005). The Large-Scale Structure of Semantic Networks: Statistical Analyses and a Model of Semantic Growth, Cognitive Science, 29, 41-78. - Youssef
Griffiths, T.L., Steyvers, M., & Tenenbaum, J.B.T. (2007). Topics in Semantic Representation. Psychological Review, 114(2), 211-244. - Todd
|Apr 13, 2009||
Week 12: Decision Making and Learning
This week, we will look a a variety of approaches to modeling human decision making. In fact, decision making processes is probably one of the most heavily modeled behaviors.
Gigerenzer and Goldstein (1996) Reasoning the Fast and Frugal Way: Models of Bounded Rationality. Psychological Review, 103 (4), 650-669. - Jim
Gureckis, T.M. and Love, B.C. (in press) Short Term Gains, Long Term Pains: Reinforcement Learning in Dynamic Environments. Cognition - Jim
|Apr 15, 2009||
Week 12: Decision Making and Learning
Busemeyer, J. R., and Myung, I. J. (1992). An adaptive approach to human decision making: learning theory, decision theory, and human performance. Journal of Experimental Psychology: General, 121(2) , 177-194. - Todd
|Apr 20, 2009||
Week 13: Developmental Approaches
|Apr 22, 2009||
Week 13: Complex Adaptive Systems
We conclude with a final segment on complex adaptive systems.
Goldsone and Janssen (2005) Computational models of collective behavior. Trends in Cognitive Science, 9(9), 424-430.
Todd, P.M. (1997). Searching for the next best mate. In R. Conte, R. Hegselmann, and P. Terna (Eds.), Simulating social phenomena (pp. 419-436). Berlin: Springer-Verlag.
|Apr 27, 2009||
Week 14: Limitations of modeling
Luce, R.D. (1995). Four tensions concerning mathematical modeling in psychology. Annual Review of Psychology, 46, 1-26.
Uttal, W.R. (1990). On some two-way barriers between models and mechanisms. Perception & Psychophysics, 48, 188-203.
|Apr 29, 2009||Week 14: Part 2 Grant panel for final project|
|May 4, 2009||Week 15: Final project presentations|