Tuesday, November 27, 2012

Keynote on Engineered Self-Organization

Just a brief note today, as I squeeze a post between proposal, paper, and parenting: earlier this month, I gave a keynote talk at the Through-Life Engineering Services Conference, a new conference put together by folks in England who are involved in a large mixed academia/industry project to tackle complexity in large-scale engineered systems like aerospace vehicles, the power grid, and trains.

Attending was fascinating for me, getting to see how people who are right in the middle of these manufacturing and management problems are actually thinking about things, and what applied research looks like in the area.  Actually, it helped clarify for me some ways of thinking and talking about my own research, particularly my work on energy demand management. Lots of interesting people too, and hopefully some of the possibly collaborations will come to pass.

The other nice thing about giving a keynote (and writing an invited paper to go along with it), is that it gave a good chance to put together a review on my work in engineered self-organization, and to pull together a unified view for myself of how all the pieces fit together. The talk, "Engineered Self-Organization Approaches to Adaptive Design," is on my webpage now (also in PDF), as is the paper---though unusually, I recommend reading my slides rather than the paper, since they were finished much later, and I think I understood the story my better by the time I wrote them.

Thursday, November 15, 2012

Silent Communion

Tonight, I looked down into the growing dusk over Montana and saw a single tiny light burning amidst a vast expanse of snow.  It sat in the middle of dormant fields, wrapped around by the darker tendrils of a rough-hewn river system.  Five minutes later, another slides by, a fiercely orange pinprick of civilization alone in the wilderness of Western America.

Who are these lonely sentinels of the wilderness?  I hope they sit warm and content within their domains, no matter the frozen lands around, and I think how lovely silent it could be, alone in the snow and nothing to see but the land, the stars, and the planes passing by above.

Presentations and networking done, I am homeward bound through the night, pulled by the stream of pictures from home that trickled into my phone across the morning, images of my smiling daughter playing, laughing, sleeping, happy in Ananya's arms.  Tonight will be late and hard, gliding into Boston well past midnight, with an internal proposal deadline still to hit tomorrow.  But I wouldn't give it up for the world.  Just sometimes, looking down, I think how nice it would be to spend a month in a cabin in the wilderness, and just let everything stop for a while.

Monday, November 12, 2012

Swarm Presentation & Paper available

I've now posted the presentation and paper from my talk at the AAAI Fall Symposia online.  This is a case where I actually recommend the presentation, "From Spatial Computing to Tactical Command of Swarms,"and its accompanying bundle of live Proto demos over the paper.  The reason is simply that by the time I wrote the presentation, I understood much more clearly how to enunciate the contribution I am making in the area of swarm control.

It comes down to one of the core problems that I hit on again and again in all of these different areas: composability.  There are lots of clever ideas for how to make a swarm of robots do something together as a group.  Many of them come from natural inspiration (e.g., flocking like birds, foraging like ants or bees, flowing like water).  The problem however, is that robots are neither birds, nor insects, nor water. For any realistically complex application, there are a lot of different aspects that have to all be gotten right, and inevitably it is the cast that not all of those will be identical to any particular natural source.  For example, if you want your robots to flock together like birds, well, they probably don't steer like birds, and the consequences of hitting one another may be more severe than for birds, and their sensors pick up different sorts of information, and they communicate with different ranges, and so on and so forth.  So we need to take the basic natural behavior (e.g., bird-like flocking), and modulate it to fit the requirements of our actual platform and application.  Moreover, you're probably going to need to put a bunch of these different pieces together in order to get anything complicated done---and our ambitions for engineered systems are usually pretty complicated, even when the core ideas or main "normal mode" behavior is simple.

So what we get from a continuous abstraction like the amorphous medium, and composition models like Proto uses, is a clean model for how to put the pieces together to get complicated behavior.  Dataflow composition gives us a clean separation of different computations, state-through-feedback means we don't have to deal with weird interactions through persistent variables, and restriction---ah restriction, the most subtle spatial operation---lets us modulate behaviors by changing where they are being computed.

If you're interested, the talk lays it out pretty well, and the demos illustrate it really beautifully...

Friday, November 09, 2012

The Joy of Scientific Airline Travel

I'm writing this now in a plane, flying back from England, where I just gave a keynote on Engineered Self-Organization and spent a couple of days after the conference working out possible collaborations with colleagues.  I'll talk about all that sometime in the near future---right now, though, what I want to talk about is the joy of scientific air travel.

I never really flew much as a kid---my family tended to drive into the nearby wilderness for our vacations, so I never got exposed enough to become comfortable with flying.  When I started flying professionally in dribs and drabs during grad school, I was always completely afraid on takeoff and landing, willing the plane up into the air or safely down to the ground as I stared intensely out the window.  It didn't help either that I was coming from Boston, since all the landing paths at Logan Airport come in over the water, and you never have land below you until just moments before the wheels touch the ground.

These days, though, I rather look forward to it.  Somehow, flying transformed from a frightening necessity into a comfortable routine. Now I sit by the window just because I enjoy the view, and also because it gives me minimal interference from my fellow passengers. Once we're airborne, out comes my laptop or the papers I need to read, and there I am with nowhere to run and no Internet to find me (no, I have never paid for Gogo, and I pretend it doesn't exist).  It's a calm, focused time, tapping away getting things done, and with my MacBook Air these days, I can eke out around eight hours of battery if I'm just writing papers with my screen brightness turned down.

Sometimes I've got something in particular I need to do, other times I just open up my machine and take stock of the state of my intellectual world.  Some of my best thinking gets done while doing that (and you get some quality blog posts too).  It just seems rather ironic to me that one of the places I am most grounded is when I am 10,000 meters in the air.

I still sit by the window, whenever I can, so that I can look out at the world going by, see the intricacy of the land and the settlements of people upon it.  Clouds too, though I'll admit I find them boring after a while.  When it's clear down below, I love to watch my progress against the map and try to identify the landmarks as they go by.  Chicago is one of my favorite cities from the air, as is New York, and on a good day flying into Boston from the West, I can mark every major city, river, and highway from Utica on in.  Once, flying out of San Francisco, we passed right by Half-Dome in Yosemite, and it practically hovered there right outside my window, turning in three dimensions.

But why am I just talking about it?  This is a blog, and I can show you pictures just as easily.  Here are a few of my favorite memories from the air: flat and two dimensional, faded compared to how they looked in person, but maybe still enough to give you a feel.

Hindu-Kush Mountains
Michigan Shore

If I ever stop caring to look out the airplane window, I'll know that I've lost an important part of my soul.

Monday, November 05, 2012

An Accidental Investigation of Publication Metrics

Dear reader, welcome once again to one of my more philosophical posts.  I've been working on reorganizing my webpage---something long in need of doing.  It used to make sense, when I was a grad student or a young postdoc, to have a simple list of all my publications.   Over the past few years, though, as both the number and variety of my publications has grown, I think this has become less sensible.  Now the list is rather long, and all silted up with the detritus of scientific publication---dead ends, early work, incremental reports, and important-but-boring filling in the gaps.

One of the things that makes my webpage such a mess is that my current list does not discriminate between types of publication: journals, book chapters, conferences, workshops, tech reports, and unpublished white-papers are all jumbled together in chronological order (possibly the worst reasonable ordering tiebreaker).

I used to solve the density problem by segregating the publications by subject area.  Subdividing further would be unsatisfactory to me these days, however, since there are so many connections between different pieces of work---do I put the first "functional blueprints" paper into morphogenetic engineering or spatial computing, since it was much more focused on spatial/cellular approaches than what came after?  How about my energy work, which started out as an application of Proto, but has evolved to shed both Proto and spatial computing in general?  There are far too many such boundary cases, and I don't want a reader to miss a publication because they're looking in the wrong section.

I suppose I could resolve the density problem by segregating them into type: put the Respectable Journals up front, followed by the High-Impact Factor Conferences, and so on.  Problem is, I've got tech reports and workshop papers that I think are more important than some of my journal papers.

Which leads to a general comment on scientific publication, I think.  So far, in my career at least, I find there to be a minimal correlation between importance of publication and "significance" of venue. An idea put forth first in a workshop (the amorphous medium abstraction), has become the most central element of my whole line of research, and I still cite that workshop paper.  Maybe someday it will be replaced with a Reputable Journal paper updating and expanding the results, but that hasn't happened yet, and isn't likely to happen soon, what with my jam-packed publication queue and parenthood.

So, let's see how my intuitions hold up against data (ah, the scientific lifestyle), and try plotting "venue" vs. "importance" .  First, I've gone through all the publications on my website and pulled out those that I think are "important," further coding some of them as "foundational"---meaning they are something whose importance I think is broad and durable, generally leading meaning it's at the root of a significant ongoing research program.  Now let's group them into publication classes using my CV, which lists 91 non-thesis publications (Google scholar finds more, but we'll ignore that whole can of worms for the moment).  In my CV, where publications are broken up into six classes, which we'll order by typical ferocity of peer review (a proxy for venue quality), in decreasing order: Journal, Conference, Book chapter, Workshop, Abstract, Informal (tech report, white-paper, etc.).  Plotting the numbers of each type as a stacked graph, we have:

Huh... my publication profile actually looks a lot more conventional than I expected.

It's completely unsurprising that the abstracts are barren of value, since they're typically just too short for anything significant---no more than two pages.  The big surprise, looking at this, is how barren the conferences are.  My guess is that a lot of those "unimportant" conference articles are steps on the way to a more complete result---and looking more deeply into them, it seems like about half of them are exactly that.  That workshop articles are largely barren is less of a surprise, since so many of them are position papers, dead ends, or roads not taken---and a deeper inspection confirms that completely.  Workshops are apparently where I toss ideas against the wall, and some of them stick (with massive importance), while most of them just fade away.

Digging into those journal articles further, I find that six of the eight journal articles started life as a "lesser" publication, and then were extended and upgraded into a full journal publication---which then supersedes the prior publication in importance, hogging all of the spotlight.  That's appropriate, I suppose.

Does this mean that I should expect the foundational workshop and informal publications to migrate into journals as well over time?  Perhaps they will---and in fact, I know that one of them is trying to already.

So what we have here in many ways is a "revisionist" picture of science, where the material that turns out to be important ends up migrating over time upwards in venue quality.  If that's the case, then "journal papers are more important" is only true for people who aren't the author: it's a selection process that retroactively highlights the important work, rather than a leading indicator.  Perhaps we should instead think of publications as some sort of an exploratory tree process.  Here's a notional diagram of what that might look like:
Color to match bar graph above.  Arrows indicate dependency, pointing from a dependent work to its source.  Size: large=journal, medium=conference/chapter, small=workshop/abstract/informal. Concentric publications indicate "venue promotions" that supersede a prior citation.
Let's say a research program started at the large bottom node with a workshop publication.  As it goes up and out, it grows and branches.  Importance tends to relate to how much research is running back through a publication.  Also, as publications become more important, they sometimes upgrade into more "quality" venues---which renders the prior version (shown as concentric) unimportant.  Sometimes a big step can be taken directly, sometimes it needs to go through bridging stages on the way.  And of course there are lots of things that end up staying unimportant, either because the initial idea was wrong, hit a dead end, or just plain got triaged by the 24-hours-per-day limit.

I suspect that I may have a somewhat higher than average branching factor, given the nature of my research and personality.  I don't know though---this may instead be an impression that I've gotten due to the operation of just such a process.  After all, the informal publications tend to fall away from visibility if they are not deliberately preserved and archived online by a researcher, and it's hard to see anything besides the mature work of another researcher.  It would be fascinating to study this over a number of scientists, but really hard to do effective coding on publications.

Coming back to the root problem that started me down this intriguing rathole: when it comes to laying out my webpage, since I'm going to be showing people a snapshot of time, I think it's only right to classify things by current perceived importance, and not by category.  And now, dear reader, an exercise for you: let's see just how long it takes between this post and an actual restructuring of my website.  If it happens very quickly, it probably means I'm engaged in proscrastination; if it takes more than six months, well, you have my permission to point and laugh.  And if you're a scientist reading this, would you be willing to contribute a coding of your own publications?