Eldan Goldenberg's lab notebook

ALife X braindump

Last week I attended ALife X. I spoke to many people, took many notes, and flagged a large number of papers to read. Because of the difficulty of writing a worthwhile summary of a conference like this, what follows behind the cut is a disorganised 'braindump' from my notes. That way at least I'll cognitively process the information a little more, and it's within the realm of possibility that the notes will be of some use to someone else.

1. In his keynote, Norman Packard mentioned a measure his research group uses to analyse the persistence of genome parts through evolution, and it sounds like it would be useful to me as well: simply attach a counter to each element in the genome, increment the counter every generation, and reset it each time the component is mutated. This way, each component's counter tells us the number of times it has been passed on since last changing.
2. Hod Lipson's group had some quite impressive results for crossing the "reality gap" to control robots, and a really cool set of robots that self-organise to form a variety of structures in a fluid substrate. I do wish he'd stop describing things like the Molecube as self-replicating though: all they do is assemble a larger structure from pre-made modules, which strikes me as rather a long way from what biologists mean by self-replication.
3. Lipson also mentioned something that I wrote down as "Fab@Home", which I think the Dorkbot crowd would love: a relatively small (as in, it would fit in my study) 3-d fabricating machine that costs less than $1000 and can extrude/3d print components to order. Unfortunately I can't find anything about it online, so either I wrote the name down wrong or it's not on the market yet.
4. Yet another xyz@Home distributed computing project: FightAIDSAtHome. It seemed worth mentioning because it has a more direct beneficial application than most.
5. Information Theory—specifically using it to understand evolved controllers and robot architectures—has definitely hit the big time this year, and it sounds like something I need to know more about, because it should help me describe my experimental conditions in a more precise way. A good starting point would probably be some papers from the Adaptive Systems Research Group at Hertfordshire University.
6. Several of the people I consider my peers (because they did their MSc at the same time as me) have recently defended PhD dissertations. Of course, this is because they're doing their doctorates in Britain, where the process is much quicker because there's usually no classroom requirement.
7. There's a new[ish] and growing ALife-related research group at Southampton University: Science and Engineering of Natural Systems.
8. There was an art exhibition associated with the conference, which I thought was a great idea. The exhibition was interesting in itself, but the part that I thought most engaging was an evening symposium. It ended up mostly being a panel discussion among the artists, mostly about the interaction between technology-inspired art and society.
9. Dirk Helbing's keynote was full of interesting applications of dynamical simulations to traffic flow problems (both pedestrian and vehicular). A case in point: the self-organising traffic light control system.
10. I wish I could remember what prompted me to write this down, but anyway: I think it used to be biologists' understanding that queen ants gave orders, whereas these days ant colonies are understood in terms of self-organising processes mediated by purely local signalling, without any central issuer of orders. This strikes me as a good example of the ease with which an observer can see a central controller even where none exists - an inviting trap when analysing my own results.
11. Melanie Mitchell's group presented a paper on spatialised co-evolution. This could be a useful technique for me to apply, because I run into problems with each agent only being evaluated on a subset of possible trials, which can either introduce excessive evaluation noise (if I keep randomising the trials) or overspecialisation (if I introduce new trials gradually).
12. Online psychology experiments
13. Larry Yeager drew a useful distinction between task-based and task-free fitness functions. Biological evolution is task-free, in the sense that organisms differentially get to reproduce due to their overall success in their environment, whereas most artificial evolution is task-based because individuals are explicitly evaluated on a specific task. Nothing revolutionary in this idea; it's just a good distinction to bear in mind when trying to generalise artificial evolution results.
14. It strikes me that one of the biggest gaps in our understanding of biological evolution is right at the beginning: how did we get from primordial soup to basic replicating proteins? From there, many of the other steps are relatively explained in terms of the constraints that the basic protein toolkit puts on the search space, but how we got to that point remains an open question. In turn, it begs other questions, particularly whether the evolution of life was inevitable on this planet, and whether ours is the only set of building blocks that could have led to complex life forms.
15. I need more microphone experience, and in general need to talk more slowly and more quietly. Empirically, people who did so held my attention a lot better than people who actually talked like me. This is going to take some work.
16. Janet Wiles described a project to collect and share design patterns for complex systems experiments, so as to reduce the frequency with which researchers re-invent the wheel. So far, it's too small to be all that useful, but it's open to submissions and is definitely a good idea: Complex Systems Patterns Project.
17. I need to start open-sourcing my simulator code. This also means I need to tidy some parts up and improve the commenting, but I should be doing that anyway for my own sake.