Introduction to Computational Cognition Course

Spring, 2012

Prof. Larry Manevitz,  Instructor

 

 

EXAMINATION MOED A GUIDELINES

Here are sample questions for MOED A.    The questions may differ in form; some may be “American”; some may be
open ended.   However, the information you will need to reply is the same as for these questions.

The exam will be closed (i.e. no help pages).

 

1)  You have to know the following networks

 

i)  Hopfield Network

ii) SDM 

iii)  EE and FM Human Declarative Memory

iv) Kohonen Network

 

2) You will need to know the abilities and differences between:

 i) Perceptron (McCullough-Pitts Neuron)

ii) Network of such neurons

iii) Liquid Integrate and Fire (LIF) neurons

 

3)  You need to know well what are the following methods for machine learning:

i)  one –class learning

ii) two class learning   3) You need to know how Kohonen algorithm works

 

4) You need to know how feed forward neural networks work

 

5)  You need to know the following model neurons:  i) McCullough-Pitts  ii) Sigmoid iii) Leaky Integrate and Fire Neuron

 

6) Know that one McCullough Pitts Neuron can only classify linearly separated sets neurons while a network can in principle separate arbitrary classifications

 

 

 

 

Sample Questions (exam questions will require the same knowledge as these but may be expressed differently):

Questions for Course on Computational Cognition

1.   Explain why fMRI scans can be used to identify cognitive states.     What classifying systems can be used?

2.  Why should feature selection be useful?

3.   Describe the difference between one-class and two class classification.

4.   Describe the system used for one-class classification of fMRI from the Boehm-Hardoon-Manevitz paper?

5.  Describe and compare the SDM and the Hopfield associative memory systems

6.  What is an “attractor”?

7.   How is a memory associated with an “attractor” in a Hopfield network.

8.   What purpose does the hamming distance serve in SDM.

9.   Describe how a Kohonen map could result in a somatosensory mapping.    Make sure you describe the function of the topology.

10.   What is meant by “equiprobable  topological map” in the context of Kohonen mapping.

11.  Describe the set up of the “Liquid State Machine”.

12.  According to the paper by Hananel Hazan, what is the importance of  "topologies" on the liquid?.    What is the result of using, e.g. a "small world" topology.

13. Describe how “randomness” is used in (i) SDM  (ii) Liquid State Machines  

14.   Describe the set up in NET Talk.   What does the network look like?   How  is the information placed in the network.

15.   Describe the capabilities and limitations of

(i) Perceptron  (ii) Network of McCullough Pitts neurons   (iii) Liquid Integrate and Fire neuron

16.   1)  What is a somatosensory map?   Explain how this might be formed by a Kohonen algorithm.

17)  i) what is the difference in the human declarative memory system EE and FM

      ii)  explain how machine learning could help establish their separate existence from fMRI signals

 

18)  Describe and compare Hopfield Associative Memory with Sparse Distributed Memory.

 

Below are materials from the course  (Some may be repeated more than once and you are not responsible for all the material here.)

You can also look in Wikipedia or in a standard Neural Network Book (Such as Faussett,  Introduction to Neural Networks) for background on the computational neural networks.

0.Students Notes (accuracy not guaranteed)

1. Background

2.Background in Kinds of Neurons and Networks

3. NetTalk and Feedforward net

4. FEM and FeedForward net

5.Reading the Mind, one class and feature representation

6. Hopfield

7. Sparse Distributed Memory Associative Memory and TSDM

8.EE and FM

9. Kohonen Algorithm and Self Organizing Maps

10. Liquid State Machines

11. Virtual Reality

12. Disambiguifying in Silent Reading

 

This site is under construction

1.  Here are slides from the first lecture

2. Here are the slides of the general overview of “Classical Neural Networks”

NetTalk Original Paper

NetTalk Data Set

Interview with Mitchell and Just

Hardoon-Manevitz-Boehm Lecture on One Class Reading the Brain

For Work on Hopfield Associative Memory Consult any text on Neural Networks

Sparse Distributed Memory

Temporal Sparse Distributed Memory

Talk on Using fMRI to discover alternative declarative associative memory