Homepage for Dylan Richard Muir

Disclaimer

This homepage is complete, and is not under construction.

When injections of dye are made into the upper layers of cortex, a remarkable structure of neuronal connectivity is revealed. Dye is absorbed by neurons and transported along axons and dendrites to fill entire cells. When a large population of neurons is stained in this way, slices examined in tangential section show a semi-regular array of denser staining.

This pattern is remarkable simply because it appears when any area of cortex is examined, over a wide range of animals. The pattern, and the yet unknown connectivity system which underlies it, is a general feature of cortex. Since spatially segregated areas of cortex perform radically different functions (from visual processing to movement control to higher cognition), this connectivity system could form part of a general, adaptable computation engine in cortex.

See the Daisy Project website for more information.

A simulated injection of tracer

Investigating the structure of the superficial patch system

We are using simulations to investigate the geometric connectivity structures that form the patch projection system in superficial layers of cortex. Through simulation and experimental verification, we hope to describe possible structures which could underly the patch system, and rule out structures which cannot. We can then decide whether computational modules could exist within this system, and what form those modules could take.

The response of visual cortex to a set of horizontal black bars takes the form of peaks of elevated activity

Analysis of the structure of OI response maps

Neurons in the visual cortex respond to the visual world with a intricate pattern of activity. Different areas across the surface of the cortex respond strongly to some visual inputs, and weakly to others. This can be seen in the figure, which shows a "map" of how a small area of visual cortex responds to a picture of horizontal black bars. These maps seem to show that neurons that are active for a particular visual input are grouped into clusters (several of which can be seen in the figure).

There is another clustered system which exists in cortex, known as the superficial patch system. This system consists of a lattice of long-range connections between excitatory neurons, sometimes spanning up to seven millimetres across the cortical surface.

In this project we are trying to determine to what extent the structure of this anatomical system (the patch system) constrains the functional system (the maps of function, as shown in the figure). We have developed techniques for automatically locating the clusters of active neurons in a functional map, and we are comparing their spatial arrangement with that of the superficial patch system.

More information can be found on the project page at the Institute website.

Other Projects

Continuous optical imaging to form functional maps of cortex

We are implementing an intrinsic optical imaging protocol involving continuous stimulation. This technique results in a much improved signal-to-noise ratio, and so produces better maps faster than previous techniques.

Spike Toolbox for Matlab

As part of my work with the aVLSI group at the Institut für Neuroinformatik, I am developing a toolbox for Matlab for generation, manipulation and analysis of digital spike trains. The toolbox supports supports stimulation of spike-based communications devices such as analog neuron simulation chips, via the PCI-AER system also under development at the INI.

More information about the toolbox is available on the Spike Toolbox page.

Shared research links

Using the magic of del.icio.us, you can access my database of links to both journals and important researchers in the fields of developmental neuroscience, the cortical microcircuit and computational neuroscience.

Publications

Muir, D; Indiveri, G and Douglas, RJ. 2005, 'Form specifies function: Robust spike-based computation in analog VLSI without precise synaptic weights', Proceedings of the IEEE International Symposium on Circuits and Systems, May 2005, Kobe.
Download unfinalised paper

Muir, D and Sitte, J. 2003, 'Seeing cheaply: flexible vision for small devices', Proceedings of the 2nd International Symposium on Autonomous Minirobots for Research and Edutainment (AMiRE 2003), February 2003, Brisbane.
Download unfinalised paper

Muir, D and Towsey, M. 2001, Better FSAs Through Clustering, Technical Report FIT-TR-2002-01, Queensland University of Technology.
Download technical report

Muir, D; Towsey, M and Diederich, J. 1999, 'Clustering to enhance FSA extraction from recurrent networks', Proceedings of the Australian Machine Learning Workshop, p. 11, November 1999, Canberra.
Download workshop abstracts

Software

All software available here is written and copyright by Dylan Muir unless stated otherwise in the code headers. Please feel free to use it, but please also acknowledge the code's origins.

Language Processing / Neural Network utilities

These programs were written for the Language Processing Group project, and are described in detail in the Technical Report written for it.

`tlearn` reverse engineering excerpt

download excerpt
tlearn is a neural network simulator written by Jeff Elman and others. In the course of the Language Processing Group project, I reverse-engineered and modified tlearn to perform on-line clustering while training. This modification is called tlavq (for Adaptive Vector Quantisation).

The code contained in this excerpt from the LPG project Technical Report is copyright Jeff Elman and the authors of tlearn. Please see the tlearn software page for information on re-distribution.

dstat

download description
download code
dstat will extract the uni-gram and bi-gram statistics for a data set, based on a {name}.pattern file.

SymStrip

download description
download code
SymStrip takes a transcribed language corpus with the words tagged with their word type (vowel, noun, etc) and separates the tags and words into separate text files. Symstrip can mark superfluous tags and insert reset markers at sentence boundaries.

MakeFSA

download description
download code
MakeFSA constructs an FSA definition from the probed hidden unit activations from a recurrent Elman network simulated with tlearn. A cluster analysis program is required to extract the locations of the FSA's states within the hidden unit space. The resulting clusters are loaded into MakeFSA. Transition tables are generated for the hidden unit activation data, and used to construct a deterministic FSA.

pattern

download description
download code
pattern generates .pattern files from tlearn .teach and .data files. .pattern files are used in several other applications, such as dstat and MakeFSA.

vector

download description
download code
vector takes a data set of known tags and sentence boundaries, and writes the corresponding tlearn .teach and .data files. The output can be written in both localist and distributed representations, and the input and output lines can be either binary or normalised together.

tlbe

download description
download code
tlbe stands for tlearn bit error. It can extract the true error (not the averaged error generated by tlearn) for a distributed output and target. It will give the number of incorrect predictions over a tlearn run for a one-step-lookahead task.

sclust

download description
download code
sclust performs adaptive spherical cluster analysis on a data set. The data can be of any dimensionality. sclust uses a modified adaptive Forgy's algorithm, and is deterministic (i.e. the analysis only needs to be performed once, and will always return the best result for the algorithm used).

o2clust

download description
download code
o2clust performs cluster analysis on a set of data. Although the correlation-matrix cluster representation implemented in the corrmatrix module is complete, the algorithm does not work at present. The problem seems to lie with forming clusters that have too few points, and therefore are unnaturally biased along an arbitrary axis. When forming clusters by adding points (as opposed to splitting larger clusters into progressively smaller clusters) the clusters grow from a few points to encompass (hopefully) a natural cluster. However, when a cluster contains one or two points, the correlation matrix is either singular or very close, and the resulting "shape"; of the cluster is merely the axis through the two points. This severe skew persists until a greater number of points are used to form the matrix.

For more information, read the Technical Report written for the Language Processing Group project.

`tlavq`

download description
download code
tlavq is an extended implementation of tlearn. tlavq performs learning on a user-defined neural network, much in the same way as tlearn, except that tlavq can also perform on-line cluster analysis of specified neurons, with the intention of using this analysis to either partially or wholly classify the specified neuron's activations into another set of nodes.

The purpose of this was to implement the online clustering architecture outlined in Das and Moser [1998], but tlavq retains tlearn's inherent flexibility. The network can be configured to any architecture possible in tlearn, and clustering can be turned off entirely. With this feature disabled, the program behaves identically to tlearn.

The source code also serves as an example to aid in the further extension of tlearn.

Software modules

These are general-purpose modules covering a range of applications.

StdDefs - Provides a set of standard definitions for all modules.
2darray - Creates and destroys easy-to-handle 2D arrays in C.
gauss - Inverts square 2d matrices using the Gauss-Jordan elimination method. Matrices must be created via the 2darray module.
RunAvg - Computes a "running" average, taken over a fixed number of double-precision samples. All samples are initialised to zero. The sampling interval is handled by the user.
htable - A brief hash-table implementation.
smdarray - Enables the use of infinitely-dimensioned arrays in C. When collecting n-gram transition data for a (say) 29-class problem, you've got a big transition array. You won't be able to allocate it. This module manages sparse any-dimensioned arrays.
TokenLst - Separates whitespace-separated words into separate tokens.
TokScan - Uses the token list data structure to retrieve tokens from a file.
vector_utils - A series of functions for manipulating n-dimensional vectors.
vector_read - Provides utilities for reading and writing vectors from file streams.
cluster - A series of functions designed to represent and manipulate clusters of vectors.
corr_matrix - Implementation of the second-order (correlation matrix) distance measure.

Bio | Contact Information

Dylan Muir is a Doctoral Student at the Institute for Neuroinformatics at the ETH | Universität | Zürich.

Dylan has a Bachelor of Engineering (Electronics) (First Class Honours) and a Bachelor of Information Technology (with Distinction) from the Queensland University of Technology (2002).

Dylan is interested in cortical development, spiking neural networks, and speaking in the third person.

To obtain a full resume, please send me an email, or phone me.

Contact Information

Email:

My apologies for the absence of a link; this is to beat spam-bots. Other bots are generally fine by me.

Phone: +41 1 635 30 46
Fax: +41 1 635 30 53
Mail: Dylan Muir, Institute for Neuroinformatics, Winterthurerstrasse 190, 8057 Zürich, SWITZERLAND

This page designed and written by Dylan Muir using Notepad. Page last updated on the 24th August, 2007.
Link to: http://www.ini.uzh.ch/~dylan/

Visceral Delights

Dylan Richard Muir