NESC 4177/CSCI 6508 Neurocomputing 2011

◇◇◇

... Instructor: Dr. Thomas Trappenberg ... Office:Room 4216 (in Mona Campbell Building on Coburg RD) .... email: tt@cs.dal.ca .... Office hour: most times I'm in my office or send email ...

Syllabus

Basic slides for the lectures

Notes:

There are currently now tutorials scheduled for Wednesdays at 3:30pm.

The grading scheme was changed with general concent that the lowest assignment is removed and assignemts and project is each taken with 50% into account.

Assignments:

  1. a: Give Three examples of models and briefly describe what makes them models.

    b1: Give two exmaples of neuroscientific issues at different levels of analysis. (NESC 4177 only)

    b2: What is the relation between theoretical and experimental studies in neuroscience? (CSCI 6508 only.

    Submit answers to prof6508@cs.dal.ca with subject line A1 before Thursday, January 13.
  2. a. Given is the following ODE: dx/dt = -x with initial condition x(0)=1. Plot the difference between the analytic solution and the numerical solutions using the Euler method and the Runge-Kutta method (function ode45 in Matlab).

    b. What is the difference between an EPSP and an action potential?

    c. Modify the simulation program for a synapse (program EPSP.m) to show the time course of the EPSP when the synapse is stimulated with neuro- transmitters every 20 ms.

    d. Use the Hodgkin–Huxley program to plot the current-response (activation) function.

    Answers must be submitted to prof6508@cs.dal.ca with subject line A2 before the class on Monday, January 24. The answers can be prepared by groups of two.
  3. a. In the lectures on conductance-based models, we distinguished synaptic models and spike-generation models. To which type does the leaky integrate-and-fire model belong?

    b. Simuate a regular-spiking Izhikevich neuron and plot the corresponding activation function (the average firing rate versus the input current)

    c. What is the value of the threshold in the activation function of the Izhikevich neuron?

    d. Plot the time course of a dynamic population node that is driven by input which is switched on and off at regular intervals.

    Answers must be submitted to prof6508@cs.dal.ca with subject line A3 before the class on Monday, January 31. The answers can be prepared by groups of two.
  4. a. Write a program that implements STDP for one synapse and show how the synaptic value changes with repeated synaptic events for regular firing patterns and for stochastic firing patterns. Explain briefly your results.

    b. The program weightDistribution.m uses exponential distributed rate values of presynaptic neurons and the postsynaptic neurons. What is the resulting weight distribution if these rate values are chosen from a Poisson distribution? Explain.

    Answers must be submitted to prof6508@cs.dal.ca with subject line A4 by Friday, February 11. The answers can be prepared by groups of two.
  5. a. Implement a single-layer perceptron and train it to translate the digital letters given in file pattern1 into the corresponding ASCII representation. Plot a training curve and interpret your results.

    b. Implement an MLP and train it to translate the digital letters given in file pattern1 into the corresponding ASCII representation. Plot a training curve and interpret your results.

    c. Investigate how much noise the different perceptrons (with and without hidden nodes) can tolerate in the pattern before being unable to recognize a letter.

    d. Which letter is represented in file pattern2? Explain

    Answers must be submitted to prof6508@cs.dal.ca with subject line A5 by Friday, February 25. The answers can be prepared by groups of two.

Course Project:

Find a partner for you course project and at topic and submit a brief project proposal by March 2. The aim of the course project is to investigate a topic in neuroscience by simulating some corresponding models. You need to choose a target article that you will discuss. You sould implement the corresponding model, possible in a reduced or simplified form and run some simualtion that demonstrate some arguments of the target article. You should then try to extend the research by investigating the behavior of the model in a novel way. Your instructor and your TA, Paul Hollensen, are there to discuss the topics with you.

Pleae submit a brief project proposal by March 2 to prof6508@cs.dal.ca with subject line Project Proposal. The proposal shoudl include the names of the group members, an initial title for the project, and an abtract of the proposed investigation . The formulation of a good project requires some background studies, so allow yourself enough time before the proposal deadline. The project must be written up as a scietific paper at the end of this course.

THE DUE DATE FOR THE SUBMISSION OF THE COURSE PROJECT PAPER IS FRIDAY APRIL 15. PLEASE SUBMIT TO pro6508 with subject line project.

Below are some suggestions for a target article:

1. Temporal sequence learning and the hippocampus.

The first paper below is an example of work that originated in a previous course project. The paper is about sequence processing in a architecture that was inspired by hippocampal models. The second paper is a recent review of replay activity in the hippocampus.

M. Lawrence , T. Trappenberg, A. Fine (2006) Rapid learning and robust recall of long sequences in modular associator networks, Neurocomputing , 69(7-9): 634-641.

Margaret Carr, Shantanu P Jadhav and Loren M Frank (2011),Hippocampal replay in the awake state: a potential physiological substrate of memory consolidation and retrieval, Nature Neuroscience 14, pp147 - 153.

2. Restricted Boltzmann machines:

Unsupervised learning with hirarchical cortical models is another great area of research with interesting applications. The following two papers are some reviews by Geoff Hinton who is the main inventor, followed by two examples of applications in computer science.

Hinton, G. E. (2010) Learning to represent visual input. Philosophical Transactions of the Royal Society, B. Vol 365, pp 177-184.

Hinton, G. E. (2007) Learning multiple layers of representation. Trends in Cognitive Sciences, Vol. 11, pp 428-434.

Salakhutpapers\dinov R. R, and Hinton, G. E. (2007) Semantic Hashing. Proceedings of the SIGIR Workshop on Information Retrieval and Applications of Graphical Models, Amsterdam.

Hinton, G. E. and Salakhutdinov, R. R (2006) Reducing the dimensionality of data with neural networks. Science, Vol. 313. no. 5786, pp. 504 - 507, 28 July 2006.

3. Sparse coding

The paper below is already a classic. It is about sparse coding that we believe is essential in learning cortical representations.

Olshausen, Field (1996), Emergence of Simple-Cell Receptive Field Properties by Learning a Sparse Code for Natural Images, Nature 381

4. Reinforcement learning

Learning from reward is an active and interesting area of research. I included three papers in this category. The first two are some good modeling papers, and the last is a nice recent review that relates some of the computational work to neurological disorders.

Suri and Shultz (2001), Temporal Difference Model Reproduces Anticipatory Neural Activity, Neural Computation 13


Izhikevich (2007), Solving the Distal Reward Problem through Linkage of STDP and Dopamine Signaling, Cerebral Cortex

Tiago V Maia and Michael J Frank, From reinforcement learning models to psychiatric and neurological disorders, Nature Neuroscience, Feb 2011, pp154 - 162

5. Calcium modeling of STDP

The model below decribes how the form of STDP could be explained through an interactions of BAP and NMDA through Ca influx. The model could be much simplified.

Shouval, HZ, Bear, MF, Cooper, LN. (2002) A unified model of NMDA receptor-dependent bidirectional synaptic plasticity. Proc Natl Acad Sci USA, 99(16): p. 10831-6.;

Resources:

Programs used in the course

The TED video of Ramachandran

My book chapter paper with examples of dynamic neural field theory in decition making

A very much recommended popular science book and brain plasticity that I mentioned in class is: Norman Doidge, The Brain that Changes Itself

Functions creature_state.m and main.m for Matlab demo

A brief guide to writing a scientific paper.

Simplified RBM letter example

Plegarism:

Please familiarize yourself with the university policy on Intellectual Honesty. Every suspected case will be reported.