020091215 - Deadline stacks I don't know why, but deadlines - academic, professional, personal, domestic - always seem to stack up. Some brief initial thoughts, with a “part 2” to follow at an unspecified later date. Part 1 here.

020091124 - Writing a thesis

It seems that the key to finishing a thesis is having an already-written thesis to revise. It's too hard to get it right first time, so don't even try. Write something, anything, and then iterate. It doesn't matter if the first version is rubbish. What matters is that there is a first version.

020090726 - Syntilects Resurrected

Syntilects. True, general human intelligence in software. The “hard problem” (it is hard!) of AI and machine learning research. I don't think we can design syntilects, but I do think we can evolve them. Here is an overview and a roadmap, and it's time to put some rubber to the road.

020090601 - Dodging Sleep

Sleep is wonderful, but there is so much else (jitsu, Mandarin, student politics, investing, writing code, drawing, my research, reading, dancing, thinking, writing fiction, writing for my website, etc.) and it's all so wonderful too, so sleeping's actually quite a low priority for me. A bit like house cleaning actually - it has to be done, but I don't want to spend more time on it than I have to. That's the goal then: To sleep as little as is sustainably possible.

More here.

020090314.2 - Geneng: The Moral Imperative

Because I was there, it shouldn't be any surprise that a recent (and strictly hypothetical!) conversation about parenting and kids turned rather naturally to genetic engineering. Parents, even hypothetical parents, are obliged to do the best possible for their children. This applies to both care (healthy food, lots of fresh air) and upbringing (do your homework). But your obligations don't end there. If you can genetically engineer your child so that they are smarter, healthier, stronger or have a greater life expectancy then it is clear that you are morally compelled to do so. Doing otherwise would be the same as letting strangers smoke in your home, encouraging delinquency, not buying healthy snacks or not cooking healthy meals. Yes, there are a whole range of social issues at play in this debate, both subtle and obvious, and maybe I'll venture there in a longer essay some day, but the starting principle is clear. Good parents do everything they can for their children, and whether or not they are born or even conceived at the time of the action is utterly irrelevant.

(Discussion with friends on facebook shows that this is indeed a very complex and subtle question, but the moral imperative to do what's best for your children is undeniable, and from that perspective GE is very hard to argue against.)

020090314.1 - Understanding Academic Papers 101: Problems

Sometimes it can be hard to understand what some random professor has written in an academic article, let alone what they mean. I found the following table of translations useful, and I hope that it can help laypeople, other postgraduate students and undergraduates understand the research they read.

020090304 - The Goal

A recent interview made clear that the long term goal I vaunt in About me was only clear to me. A new essay here will hopefully solve that problem, so now anyone who wants to can know where I'm planning on going.

020090212 - An Idea

Graphical models like Bayesian networks are one of the greatest things since sliced bread in modern artificial intelligence work. Inference of a BN is fiercely difficult though, especially inference of the graphical structure. Parallelisation is a huge boon, but it's difficult to parallelise a very small task because of the overheads (communications cost, synchronisation) between nodes, whether those nodes are CPUs, threads or something else.

One idea is to apply an evolutionary algorithm to the structural search, and then perform parameter inference for different models in parallel. Map-reduce strikes again…