Fast Self-Healing Gradients

We took the Constraint and Restoring Force memo, restructured it took take our language out of the story, explained the problem it solves more clearly, and gathered some more conclusive experimental data. The result is a paper, Fast Self-Healing Gradients, to appear at the ACM Symposium on Applied Computing in March.

Constraint and Restoring Force memo

I've posted a memo on Constraint and Restoring Force, a framework for self-healing systems that we're using in amorphous computing. Primarily developed for calculating gradients (a gradient in this sense is a field of distance estimates to a source region) this technique can also be used for other calculations like cumulative probability fields.

I've also updated my CV for the first time in two years and fixed a lot of little bugs and stale bits on the page.

Thesis!

My thesis is complete, turned in, receipt obtained, and now it's up on my web page as well. Weighing in at a slim 218 pages (144 when front matter and appendices are excluded), this document officially concludes my passage through graduate school.

Now, on to post-doc and beyond!

Successful Defense

My defense went well yesterday. There were a *lot* of people there---significantly more than the room could hold, including 11 professors besides my committee. The talk was good, and held the attention of the audience, though there is definitely more polishing to be done: I don't yet have enough exposition of the meaty technical core of the work. Still, that is about looking forward to job talks.

The slides are up on my web page.

Countdown to Defense...

I'm in countdown to my thesis defense right now (it's coming up on August 8th), yet the rest of life hasn't stopped rolling along. I've been to two conferences in the last month, and their associated publications and talks are now up on my web page.

On the Spatial Computing/Amorphous Computing side, there was the first edition of IEEE SASO, where we presented a short paper on continuous time semantics in a 15 minute talk.

On the AI side, I have now posted the talk on Developmental Cost that goes along with the paper I posted back in May.

Amorphous Computing Reviews

I'm posting a draft article reviewing Amorphous Computing written by me, Hal, and Gerry, and an informal talk I gave to the Friday bull session a couple months back. These were both really fun chances to look back over where things started and how far along they've come, though the review article was fairly stressful given my looming thesis.

Also, I decided that the "Engineered Emergence" section of my web-page ought to be below the sections that have been there longer and have more formal stuff.

Thesis fragment appearing at AAAI

One of the things I've had to wrestle with in my work is how to explore architectural ideas freely while still keeping things connected to human intelligence. I finally managed to write down a clear explanation of my approach for my thesis, and that fragment is going to be appearing this summer at a AAAI workshop on "Evaluating Architectures for Intelligence."

Here it is: Developmental Cost for Models of Intelligence

Slides, and a new section

In March, I gave an invited talk at a workshop on Unconventional Computation, in which for the first time I drew together the scattered threads of my research into a single theme: "engineered emergence." I'm not really happy with the name, since the word "emergence" carries a lot of flakiness baggage. For the moment, though, it's the best way I know to describe engineering a bunch of unruly parts so their interactions will produce coherent behavior of the whole.

The slides from the talk are here, and I've reorganized my page to give engineered emergence its own section, pulling in the Assassin's Guild talks which push me towards thinking this way.

Tech Report: "Building Spatial Computers"

Lots of stuff in flight, waiting to see whether it gets accepted or not. Also, I'm deep in the thesis-writing process.
Anyway, we've memoized an amorphous computing paper that got rejected, since it might be a while before its next turn around the wheel of submission:

Building Spatial Computers, Jonathan Bachrach and Jacob Beal, MIT CSAIL Tech Report 2007-017, March 2007.

This was a submission to HotOS, talking about the operating system level problems of building Proto and making the amorphous medium abstraction actually work. Half the reviewers thought we'd already published the paper---no! This is the one where we actually talk about implementation, rather than sweeping it under the rug to focus on why somebody should care about our abstraction. A third complained that we didn't focus on the programming abstraction. The fourth had good, helpful feedback that will actually help us on the next turn of the wheel.

It seems to me like there's a fundamental problem in trying to present a non-incremental change in a computer science paper. If you try to present it all in five to ten pages, you can't give enough details to satisfy anybody. If you slice off a bite-sized chunk to talk about, then either you've already published on the idea or the reviewers complain that they wanted the other parts. I don't know whether other fields are like this too, or if this is just to be expected when you're working on problems that don't come with a pre-existing focussed sub-community dedicated to their study.

If I were a better writer, I might know how to present things to get the readers to reliably take that leap in five pages. For now, I'll keep struggling and getting frustrated every time I get a review that says something like "But why aren't you programming in C?"