It's been too long

since my last blog post. For various reasons I haven't been able to settle down to write one.

But more important I'd been out of work for too long. At last I'm back at work, doing something I find interesting and getting enough money to have a reasonable set of choices of things to do outside of work.

The downside is that I had to move interstate to get this job. For now I'm living in pretty basic accommodation in a hostel mostly catering to overseas students. Ah well, it cuts costs. And I don't mind living in Brisbane.

I'm working as a research assistant for the University of Queensland in the Institute for Molecular Bioscience and the Department of Mathematics. I'm working on model-based clustering, in particular the use of finite mixture models for clustering.

You see ,clustering is answering the questions does this data fall into groups, if so how many and which observations are in which groups. The thing about cluster analysis is that you only have the observations, not the groups that they belong to. For any of them. You have to create the groups, not put the observations into known or predefined groups. That is classification.

Not surprisingly it is a less well defined problem than classification. Usually you create a matrix of distances between observations. (And one of your first decisions is how to define these distances.) Then you put the observations through some algorithm for joining them up into groups. A problem with this approach is that it is difficult to come up with an objective measure of how good the classification that you come up with is.

I'm working on testing improvements in one alternative approach. This is to assume that the results come from a mixture of distributions of specified but flexible forms. You put the observations into an initial clustering and then improve it by iterative methods. In principle this should put the clustering on a better mathematical foundation.

One of the nice things about the job is that I get to go to some very interesting seminars at the IMB. And since I started out in Biology I find some of them very interesting indeed.

Let's see there was the talk on larval development in sponges. Fascinating. The larvae are more structured than the adults. They are more closely related to the rest of the animals than we thought. You see at one time we thought Porifera (sponges) might have developed from protists independently of the rest of the animals.

But here you see a sponge larva. And it looks like nothing so much as a planula, the larval form of such things as hydras and jellyfish. With a couple of interesting differences. One is in the arrangement of the flagella. But the big one is that it is radially symmetrical where a planula is bilaterally symmetrical even though the adult Cnidaria (Hydras, jellyfish, corals, sea anemones etc.) are radially symmetrical. There are a couple of common factors producing developmental gradients both in Poriferan and Cnidarian larvae. The thing is in the Cnidaria one of them runs longitudinally and one runs dorso-ventrally. In the Porifera they both run longitudinally. This suggests some interesting questions about the development of radial and bilateral symmetry in animals.

Things are looking up. Finally!

Coming up soon, the nervous system we didn't know we had.