Difference between revisions of "CoSMo 2016"

Line 54: Line 54:
 
[[Media:CoSMo2015-vOpstal.pdf | van Opstal syllabus - linear systems theory]]: a great syllabus developed by John van Opstal for CoSMo on using linear systems to model gaze control with theory, exercises and answers to exercises  
 
[[Media:CoSMo2015-vOpstal.pdf | van Opstal syllabus - linear systems theory]]: a great syllabus developed by John van Opstal for CoSMo on using linear systems to model gaze control with theory, exercises and answers to exercises  
  
[https://www.cl.cam.ac.uk/~rmf25/papers/Understanding%20the%20Basis%20of%20the%20Kalman%20Filter.pdf Understanding the Basis of the Kalman Filter Via a Simple and Intuitive Derivation ]
+
[https://www.cl.cam.ac.uk/~rmf25/papers/Understanding%20the%20Basis%20of%20the%20Kalman%20Filter.pdf Understanding the Basis of the Kalman Filter Via a Simple and Intuitive Derivation] <br>
 +
 
  
 
=== Day 4 - Optimality and data analyses ===
 
=== Day 4 - Optimality and data analyses ===

Revision as of 16:53, 2 August 2016

This page contains course materials for the CoSMo 2016 summer school.

CoSMo logo


Introduction

Aug 1-4
Lecturers: Gunnar Blohm, Paul Schrater, Konrad Kording

Day 1 - Overview of sensory-motor computations

Organization slides
Philosophy of modelling slides
Sensorimotor overview slides


Afternoon tutorial 1 - plotting neural data

Here is the file [Stevenson Data Set] As part of the tuning curve exercise we will understand it.


Afternoon tutorial 2: gain modulation for reference frame transformations

The goal of this tutorial is to understand how gain modulation can be used for reference frame transformations and how gain modulation can emerge from training a simple artificial neural network carrying out reference frame transformations.
There are 2 different approaches to solving this:

  • exact determination of read-out weights from eye-position gain-modulated neurons as in this seminal paper. Here the solution can be found by computing the least-square optimal set of weights mapping the gain-modulated neurons (population code) to head-centered output neuron(s). For this to work, population code neurons need to be of the exponential function family.
  • training a neural network to perform reference frame transformations using this code. For this you can plot each individual neuron's receptive field for different eye positions and analyze how the receptive field changes with eye position in each network layer.


Day 2 - Bayesian approaches

Bayesian perception - an introduction: a tremendous book written by Wei Ji Ma, Konrad Kording, Daniel Goldreich


Morning lectures and tutorial (Konrad)

Conditional Probabilities slides

Bayesian models slides

Cuecumber nation slides

Tutorial and additional steps for extra points


Afternoon lectures and tutorial (Paul)

Here are tutorial instructions. Please also read the papers by Lappe and Seno.


Day 3 - Linear systems

van Opstal syllabus - linear systems theory: a great syllabus developed by John van Opstal for CoSMo on using linear systems to model gaze control with theory, exercises and answers to exercises

Understanding the Basis of the Kalman Filter Via a Simple and Intuitive Derivation


Day 4 - Optimality and data analyses

TBD


DREAM database - Shared data and models for CoSMo projects

Jul 31 - Aug 14

Curtesy: Konrad Kording

You can get the DREAM project from Gunnar on a USB drive. DREAM can also be downloaded piece-wise (data sets, models, tools, and documentation) from CRCNS: http://crcns.org/data-sets/movements/dream/downloading-dream. You will need to create an account on CRCNS to be able to download the project files. If you want "all" of DREAM (models, tools, and documentation), click here: AllDream.zip

If you're familiar with svn and would like info/credentials for code in the repository, contact Ben Walker


Here's the latest version of LoadDreamPaths.m. (This script should work for all OSes.)


Here is a description of data sets currently in Dream. Dream is growing, but this list is accurate as of the time of the summer school (click on the link to access the related publication).

  • Burns -- reaching with head tilt and left/right visual perturbations
  • Corbett -- reach trajectory predictions based on EMG and gaze movements
  • Fernandes -- reaching with uncertain and rotated midpoint feedback
  • Flint -- decoding of reaching movements from local field potentials
  • Kording -- reaching with uncertain midpoint feedback
  • Mattar 07 -- generalizing from one, two or multi targets to another direction
  • Mattar 10 -- reaching to a distance (short/long), generalizing to the other one (long/short)
  • Ostry -- move in force field, get an estimation of where the hand is
  • Scott -- monkey (no spike), center out: even and not evenly distributed targets, also a forward/back
  • Stevenson -- center out, monkey with neural time stamps
  • Thoroughman -- reach adaptation to perturbations with different complexity
  • Vahdat -- movement in force field with FMRI scans pre/post learning
  • Wei 08 -- visual perturbations, cursor shown only at target
  • Wei 10 -- movement in differing force fields
  • Young -- movement time stayed the same, but distance changed; fast, medium, slow reaches.


Group projects

Jul 31 - Aug 14

Here are some ideas for 2-week project topics.