Contact Details Email: crispin@cantab.net
A photo of me not working

Crispin's Research Page

This page exists to answer the following obvious question:

What do you do?

I used to research bio-inspired electronic engineering and computer science. I've written a fair bit about this below.

However, I've just started a PhD in social science. Specifically, with the School of City and Regional Planning at Cardiff University. This may seem like it bears no relation to my background, but bear with me: it's interdisciplinary research. Once I've learned something about social science, economics and city planning, we'll see where I can apply my existing skills. Depending on your technical background, I might summarize my existing skills either as "genetic algorithms, systems that model biological growth, and learning" or more simply as "artficial intelligence". Unfortunately the latter label has acquired a plethora of meanings, most of which have nothing to do with me, but it suffices for a quick summary I suppose.

Quite what I'll do is anybody's guess. The research is very likely to involve agent-based modelling. Ideas to date have included simulation of sustainable urban growth, disaster management and evolutionary economics.

What did you do before?

I worked for the Intelligent Systems Group, University of York, UK.

I initially worked on the POEtic project. This is a hardware platform which seeks to unite Phylogenesis, Ontogenesis and Epigenesis - three long words which mean evolution, growth, self-repair and learning. For a few years now people have been playing with hardware that can do all this - mimicking evolution to create circuits which design themselves; mimicking growth to create complex structures and mimicking learning, usually with Artificial Neural Networks (in our case, Spiking Neural Networks). The POEtic project attempted to put everything together.

Oddly enough I was working on some audio applications for this platform. The reasons for this was that (1) they showed off the chip's features quite well, (2) the world of audio technology could do with more of this kind of thing and (3) the interactive aspect of these applications appealed to the funding body.

As a progression from this, I proposed a study of unconstrained evolution combined with growth, and hardware artificial life experiments. This has recently been funded although I am now employed elsewhere.

I've also had a brief stint as an outdoor pursuits instructor, a maths tutor, a private detective and a software engineer with Cambridge Silicon Radio.

Papers:

Some audio clips can be found here.

Shelved Projects

Music Technology

I thought a lot about creating a neural agent based drum machine. When it comes around to it though, I'd rather get on with actually writing music.

The Disposable Synthesiser

This project was going to demonstrate the growth. It's a physical modelling synth, in hardware, that simulates a plucked string. Every time a note is played, a string of the appropriate length "grows" onto the chip's "ontogenetic tissue" and is played. This should happen within a few milliseconds. When the key is released, the string dies.

It's also a VST plugin.

There's a video of a basic growth process here. What you're seeing is (mostly the routing plane of) POEtic's hardware simultor, developed by Yann Thoma over at EPFL. Each block of 2x2 routing units represents a cell, and all cells (apart from the bottom left one) are identical. Each cell "decides" to connect to another, and does so until a chain of the requisite length is formed (along with backward connections - each cell downloads it's address to the next to allow this). During the growth process, the view is changed to show the underlying molecules for a couple of growth cycles - then switched back again.

Crazy Multidimensional Meshes

So far, Digital Waveguide Meshes have mainly been used to simulate real objects. This isn't a limitation of the capabilities of meshes - meshes are perfectly capable of simulating the acoustics of multidimensional objects. However, people have a hard time visualising these things and designing a 64-dimensional user interface would be pretty daunting. The parameter search space, when looking for a particular sound, is massive.

I was having a go at evolving some multidimensional meshes, representing the connectivity with a varient of Lindenmayer Systems - a bit like this. Selection was interactive, (although it would be an idea to try automated selection by spectral shape or maybe fractal depth one day).

This is still in early stages, but here are some interesting sounds made from finite unbounded spaces. As an analogy: when Pacman leaves one side of the screen, he re-emerges on the other side. That's because his maze is finite and unbounded. It's as if he lived on the surface of a torus.

• Sheet.ogg is what Pacman would hear if he clicked his fingers and the maze had no walls. It's not just a click noise because the sound leaves one side of the room and enters the other again... and again...
• Room.ogg is the same thing for a 3-d room.
• A friend asked me what you'd see in a room like that. Well, this is what you'd see if you (the observer) were a black cardboard cutout oval.
• Wormholes.ogg is the sound of a 2d space rapidly being filled up with wormholes in space. Italic script is used here to denote a deep voice with lots of delay and reverberance. By wormholes, I mean random connections between points which aren't next to one another.

Last Updated 26.02.2007