Home

I am Alexandre Devert, (evil) computer scientist, engineer and stuntman. This is my homepage. You will find here:

A blog is born

For various reasons (giving back a contribution, because I owe so much from others work, exercise in writing, fuel for motivation to do things, etc.) I started a blog, and it's here Don't expect more than a post every 2 weeks. It will be mostly on programming things with a computer, although I hope to tinker with hardware some days.

Experimental setup uploaded

I uploaded part of the experimental setup used for the paper A Study on Scalable Representations for Evolutionary Optimization of Ground Structures. It's in the code section of this website, in the research category. That part covers the NEAT based experiments only. The other setup, based on CMA-ES and Multi-Layer Perceptrons, will appear later. The documentation is included in the archive.

New job, new journal paper

This page is becoming the Happy News Journal ! I am now lecturer at the SSE, Software Engineering School of the USTC, in Suzhou, China. Thanks to all my colleagues, for the great environment. Teaching materials will blossom over the year.

And I also glad to announce that we, Thomas Weise , Ke Tang and myself, have an article accepted in Evolutionary Computation Journal, titled A Study on Scalable Representations for Evolutionary Optimization of Ground Structures. This article shows new, surprising, though-provoking results, comparing different encodings for a classical mechanic benchmark problem. We show that an ontogenic encoding is the least computationally expensive to find close to optimun solutions. The ontogenic encodings shows a great scalability, whereas the generative one is not, which is was not so evident.

ESKit ? Hum... Python bindings. Later. Not now.

Journal paper accepted

I am very happy to announce the publication of an article in the IEEE Transactions on Evolutionary Computation journal, Robustness and the Halting Problem for Multi-Cellular Artificial Ontogeny. This article is a heavily reworked chapter of my PhD thesis, adding more experiments and an important work on improving the redaction. The article introduces an extensive set of experiments on a simple model of Artificial Ontogeny. We show the influence of the stopping criterion of a development process, when optimizing development processes. The stopping criterion is a key element to obtain stable development processes that stabilize on a consistent, stable solution. Big thanks to my co-authors Nicolas Bredeche and Marc Schoenauer.

On others totally unrelated topics... I am not looking for a job anymore, being employed since January 2011. As a consequence, work on ESKit is paused for now. In the other hand, my research work is not paused, albeit temporarily quite reduced.

Job hunting

I moved to Ho Chi Minh City, and I am looking for a job. I would like to stay there on a long term basis. Something that match my feature set is of course what I wish. Optimization, machine learning guy, scientific software guy, data mining guy, software tool maker guy, AI guy, computer graphics guy, computer science teacher, pick the one you like !

You can also fund my research, if you are interested in shape optimization, I have some nice results and awesome ideas. From a more general point of view, my research goal is to provide tools for quick design and conception of solutions to large-scale, complex engineering problems. I have a fair experience in robotics, signal processing, and machine learning as well. I would be also very happy to teach about all those topics.

Meanwhile, the Python binding for ESKit did some nice progress, but requires additional work. A lot of testing, making examples with genuine chunks of eye-candy inside... Let's aim end of November for a release.

ESKit 1.2.2 released

This release brings an important bug fix : on 64 bits architectures, the random number generator was not random at all. As a result, the optimizers were not as good as they used to be. It was due to some implicit conversion to 32 bits made by the C compiler.

Although not yet ready for the day light, the Python binding is already in a usable state, likely to be included the 1.3.x series.

ESKit 1.2.1 released

Some compilations issues have been fixed, discovered while compiling ESKit for using it on a cluster. Check it out here

A big thanks to Andrew Miller, who did a binding allowing the use of ESKit in Haskell . You can get it on github and hackage

ESKit 1.1.0 released

I added the possibility to set a custom covariance matrix for CMA and SepCMA. This little change allows to do nice tricks like bootstrapping CMA by running SepCMA first. I also found a bug and fixed it, a typo than can accidentaly the whole optimization... Check it out here

Hello, world !

So here it is, I got my own hosting for my own website. The motivation behind that is a strong will for... Naaaaaah, it's just that I had no other options : my former lab, which gently hosted my previous homepage for free, have to close my account (I'm not working there since more than a year, makes sense), making any update impossible. Since I tend to be rather nomadic those days, renting some webspace seemed the best solution, thanks Thomas for the useful tips.

No Joomla, PHP, Django, Ruby On Ion Drive or some of the numerous hype ways for the Web 2.0... A 154 lines Python script, Mako, and Docutils turns a bunch of reStructured text files into this fully static website. In plain english: this website is resource friendly, and is easy to maintain and feed on my side.