Million, Billion, Trillion, and Visceralization

Been busy all day so here's a quick catch-up. A lot of things that happened today are not suitable for discussion in a blog - this thing is "on the internet", right?

In any case, the presentation by famous illustrator Nigel Holmes was thoroughly enjoyable. One thing he demonstrated was to how to show people the difference among Million, Billion, and Trillion because they look so disarmingly similar in spelling and word-length. Volunteers were lined up in the front; the "billion" guy literally competed for the same spot with the "million" guy, but the "trillion" guy... well, he was far, far away. The smart ones would think, wait, you should use logarithmic scale. But logarithmic scale has exactly the same effect as the look of the words: by logarithmic measures they'd look comparable while they're not.

Cool as this "visualization" is, it doesn't differentiate well between million and billion - they're actually at one spot. You could try to rescale the experiment (picture the "trillion" guy crashing through the walls of the Hilton Hotel like a projectile)...

In the San Francisco bay area, people have developed, knowingly or not, another way of understanding the difference between million and billion. I would even venture to claim that we Californians have a better understanding of this difference than any visualization can ever provide. We have a more primal, gut-feel, visceral understanding of the difference of million and billion. We have acquired this instinct by associating our life with millionaires and billionaires.

Before I continue with the analysis, I think I may just go ahead and establish a new branch of study, right here. Visceralization. This is more than visualization and targeted at the gut. While visualization stimulates the neocortex, visceralization excites the thalamus. For example, with visceralization you can experience a vector field, one from e.g. “high energy physics” simulation, as euphoria or, as an Alex special, an ancient kind of arousal. You can be physically “sick to the stomach” due to some anomalies in the simulation, and "a sinking feeling" tells you to change the parameters. Of course, a lab where visceralization technology is adopted would probably look like the institution in Flying over Cuckoo’s Nest, but that’s in the name of furthering science.

OK, now come back to million and billion. I’ll further qualify million as one million, and billion and one billion; let’s leave the gray area to economists. In the Valley, everybody knows a millionaire is nothing. To avoid using cliché, I’ll only remind you of the saying, “shoot in any direction and …”. Many millionaires don’t even have a house. When they want to buy a house, they can’t get a good one. When they decide to buy a not-so-good house, they stop being millionaires. Well, they can hold up a bold face and speak of “net-worth”, but we all know that they’ve got no stinking liquidity.

The billionaires, though, are a different story. There aren’t any really constraints on their life. I was to list the kind of things they can do or get, but I’d be sitting here all night. (By the way, listing such things with full abandon is a good way to visceralize the billion.) So I turn to things they can’t do - resorting to the dual space, as more academically oriented friends say - which we can count with one collective thumb. It’s not good for them to get divorced.

Data arrives

FedEx delivered the disk ahead of schedule. Then George arrived from Chippewa Falls with another disk. It turned out that the transfer from Mountain View finished 7:30am this morning.

Meet the neighbor

I went to the dinner reception tonight and walked around the see the posters.

The most interesting conversation I had was with Leonid Zhukov, of Yahoo! Research. A guy from Stanford and Leonid guarded this poster about the visualization of correlations between music from bands and singers. There are visualization and data gathering aspects to it, but what I’m most interested in is data processing, e.g. the computation of the correlations and other measures on a giant database. Can large, shared-memory systems be of help here? Leonid said they have a very large database and the clusters were horrific to use since, for one, their matrices are very uneven (although still in general sparse). This creates data distribution problems and failure could follow.

This is really music to my ears, and Leonid was very receptive to shared memory architectures. I think we were mutually enthusiastic. The enthusiasm may face some obstacles because, as Leonid said, Yahoo! won’t easily let the data out of the door to commercial entities (though they’re open to academia). A large hardware vendor was refused access because lawyers couldn’t agree. But I hope we (SGI and Yahoo!) can work something out. Leonid was certainly interested when I talked of daily access to 64p, 256Gb systems, if not bigger. (Rocky, we should try to hook up with these guys when I return. Everything else aside, it would be delicious to understand their problems. It would be fun as well to borrow from our tremendous in-house expertise with large matrices.)

The most goofy demo at the reception was this jello sculptor (which currently visualizes height fields only – I guess otherwise it’s got a problem supporting it own weight.) The one I liked best is a tracked two-hand-held, semi-rigid screen. You’d move it in front of a projector and the picture is shown as if you were slicing through a volume. All position, angle and flex (!) are tracked. The real trick, according to the guy demo’ing it (they have a paper for Friday which I’ll miss), is the lens. It’s screwed onto an Epson projector and is claimed to have a sharp focus from 0 to infinity. I blurred out something like, “Oh it’s a pinhole”, but then I realized that you couldn’t pump this much light out of a pinhole.

So there goes my day. Blogging is actually fun. To write is kind of weird; you are not in total control. I think much of the time I watch my fingers dramatize and exaggerate my sentiments and thoughts. It’s funny and it’s good to be out of control.

The Network is the Anti-Computer

With this statement I may have rubbed various people in various ways, so I have to justify my position. I had not had this position until this afternoon. But let me explain…

I spent the afternoon in the SGI booth doing something I’d never done before: installing software for a demo. The software is an open-source ray tracer co-developed by SGI and the University of Utah, with much of my work in it. But the software is not really the issue, the data is. The dataset involved in this demo is 34-Gigbyte, representing the whole Boeing 777 airliner in all its glorious and gory details. You see all these wild pipes and wires and... stuff. Screws, rivets, and buttons are not exactly wild objects, but it's wild to have so many of them; it's wild to have all of them represented in detail.

In any case, the plane is digitally bulky at 34 Gigbytes and 350 million triangles. Triangles? Sorry for the geek speak, but these were what Jim Clark founded SGI to draw, and SGI is still doing it, so many years later, with Jim watching (or not) from his sailboat. Yeah, the drawing of triangles has a tradition in the Silicon Valley that far predates nVidia and ATI... I still have not explained why we draw triangles, but you get the idea – it's the starting point of a great adventure, the first masterpiece by a man revered as a national treasure, and a stepping stone to the mighty heights of Netscape.

Did they tell you that it was easy to digress in a blog? I’m going back to installation of the demo. The first sign of trouble came when I noticed that the data file was appallingly small – about 2 gig. My heart sank further when I found that, in fact, it was the wrong data – a much smaller piece that was usually used only for testing.

In case you have not realized, we were in trouble. The good thing was that there was no confusion; the problem was well defined: ship 34Gb of data in the fastest possible way from 1500 Crittenden Lane, Mountain View, CA to 1001 Marquette Ave., Minneapolis, MN, i.e. from SGI headquarters to a Hilton Hotel in Minneapolis. (I have the full addresses here because they will become relevant later.)

The first thought, of course, was to transfer it over the network; the immediate second thought was, don't even try. Despite widespread gossip about Paris Hilton and her Sidekick handheld, this hotel isn’t exactly a high-tech place and nowhere near the backbone of the Internet. For a while Alan was able to pull only a few K’s per second. Something else had to be done.

George came to the rescue. The SGI campus at Chippewa Falls is about an hour away, so George came up with an idea. His good friend Brian would transfer the data to the Chippewa Falls office, and George would go and fetch the hard drive via the public road system. I felt relieved… George got into his car and went on his way, speeding towards Chippewa Falls. In the meantime, swarms of data packets of delicious bits flowed relentlessly from California to the same destination. They’d rendezvous and make me whole. It all sounded good.

Then I got a phone call and it was George. No, his car didn’t break down. Despite the tens of thousands of explosions required to get him home (uhr, they call these “internal combustion”, but you know what I mean), George arrived without an issue. But the bandwidth from Mountain View to the Chippewa Falls office wasn’t good. I forgot the actual number, but it was not good at all – the ETA, or expected time of arrival, would be many hours; not good to the extent that George suggested that we have a back-up plan.

The backup plan was… drums rolling… FedEx. This has been a joke in the large data processing community, but it was now unfolding in front of my eyes. I called home and the good man and woman of SGI, Mountain View, (you know who you are – thank you!) managed to get a disk loaded and mailed out at sundown. It will be delivered before 10:30 tomorrow morning. After all the trouble, I can easily settle with that.

So if the network isn’t exactly anti-computer, it is easily the weakest link. (Imagery begins to arise, "you see Mr. Anderson, what is the use of a phone call if you can't spea..k?", but I refuse to digress.) The bandwidth will not be there when you need it the most.

Ultimately, to move anything around physically is an expensive deal. Moving around large collections of atoms (e.g. the bodies that comprise this conference) is the most troublesome. Moving electrons is relatively easy, but as we have learned, it’s not so easy to move e.g. 35Gb’s worth of them. I guess spintronics is to avoid moving the electrons… ok, I don’t know what I’m talking about, and so I stop.

InfoVis

I caught up with John and Alan in the hotel's cafe and bastards left me alone at the table in the name of work (setting up the machines). To be fair, I arrived when they were already done, but still... For those who don't know, in California we sacrifice everything in the name of work. Marriage comes to mind. But I digress...

I stayed with the InfoVis session this morning. This feels more interesting than "hardcore" visualization because here visualization isn't just rendering. It's about data processing and the transfer functions to visual elements. There are some fundamental questions, though. First, infovis implies that human beings are in the loop. The human beings, as we all know, are most often bottlenecks. So, infovis already got organic bottlenecks built in before things even get started. Of course, one could argue that InfoVis is not for the masses; it's for decision makers. But we already know how capable decision makers tend to be. In any case, I do think InfoVis is useful; but in VC speak, it is a Vitamin, not a painkiller.

There is a second and more important weakness that hampers current infovis studies. It has been argued that InfoVis is a must because the human brain processes patterns much better than current computers. Very well, and very true. However, here also lies the problem. Our understanding of the brain is so limited that it is not clear what it means to present information in a form catered to the brain (which is the premises of infovis). This difficulty is similar to those in other branches of study where researchers try to involve cognative and perceptual measures of quality. Visual quality is hard to measure because we don't know what it is (why is a woman beautiful? - this is for Alex). The same is true with artificial intelligence; come to think about it, it's pretty optimistic to try to make it artificial without knowing what it is. Of course, the pioneers didn't know that they didn't know, but still.

The situation is partly reflected in the quality of the papers. There are good ones that I'll mention below, but I heard quite a bit of rah rah about the grand possibilities (to be fair, part of this is probably a left-over from the keynote), the kind that is usual associated with neuroscience, robotics, and computer vision. (I'm actually a big fan of all these disciplines, but let's admit that they're the types that most easily get people's hopes up.) A lot gets said about a few little pictures. An extreme case is this presentation that installed a lot of totally irrelevant photos in the slides, those of dams, white-water rafting, etc., as decorations. Visualization of your information visualization ideas? Gimme a break.

Two papers are of particular interest. One deals with highlighting conflicting dynamics in "event data", in this case news reports of world events. You can see in an animation friends become foe and vise verse (friends on the same side of the axis, the more friendly the larger the cooridinates.) It's pretty cool. A thought I had was that if you mine the Iraq related news with this method, it would probably still show the two country as on opposite sides at a great distance (indicating great hostility), because marines get killed regularly in Iraq. Imagine some red-faced high-ranking official rush to pull the plug on the projector, or perform a fast draw to shoot the computer - whichever is faster. There goes a third cautionary note about infovis: it is considered useful only when presenting what the boss want to see. I predict that being useful in this way is a make-it-or-break-it deal for future infovis systems.

That, ladies and gentleman, is high-tech. That capability is probably from a part of the brain the computer would never have enough sophistication to mimic. We're really looking into the future here.

Another paper I liked is about the "low-level components of analytic activity". The title made me deliver a high-pressure aerodynamic event from the nasal passage, i.e. to sneeze, but it's about what kinds of queries the user would like to do in front of data. I like this paper because the approach is well-grounded in experiments.

Picture

Here's my feeble attempt at including a picture (drag-and-dropped from my cell phone, the now ancient Sony T610).

A Grand Challenge in Science

Jim Thomas of PNNL has just started delivering a keynote. I believe they're an IBM shop, the first user/co-developer of the SGE, or Scalable Graphics Engine. That's the 4 GigE input framebuffer thingy with inputs from Mesa renderers on an IBM multiprocessor.

Subtitle of the keynote is "Turning information overload into the opportunity of the decade". I would think that the kind of high-cost "opportunities" he would likely talk about is "of the century"; but balancing that with the possibility of non-opportunities, "decade" seems a reasonable compromise. But I promise this was the last manifestation of my morning cynicism.

Stuard Card "Information Visualization: Wings of the Mind" - very inspiring, according to the speaker. Somehow this reminds me of "Ender's Game", but that must be a different Card.

Book: "Rembrandts in the Attic", "Illuminating the Path".

What's Changed? Scale; Time to decision: secs, minutes, hours, as compared to days and months before. Nice. Bombs get dropped much more quickly these days. Sorry for breaking my promise (already). Next gen distinguished by Data-to-Decision in complex systems (color duplicated from the slide).

Visualization of Ports of Entry - the slide is terrifying. You've gotta see it to believe it - they're everywhere. (I'm sorry for the total relapse. What was the quote in Office Space about Monday?)

Discourse and discovery. Relationship visualization: tree, mesh, star structures for determination of intent (related to the Washington DC sniper).

Information synthesis - high dimensional discovery.

Walk-up usable; information into my context space; old info from new perspectives with new experimental data (sniper again - another guy shot, and the Petri-Net changes);

Systems biology: data sources... (I had a senior moment and the speaker flipped to next slide)

Share Lessons Learned: Most researchers don't "experience" the need - I think this is illustrated by the term of the "perfect storm"; for the guy who really experienced it, there's nothing perfect about it.

A movie clip was shown in which a guy successfully performs the stunt of avoiding a collision by running on top of the car that was the threat. He was promptly collected by another car traveling in the opposite direction, just as he began to be complacent. The morale of the story is that if you don't get killed by some data, you will be by others.

Don't understand how the human mind performs abstractions so effectively on so many levels. Reminds me of the scene in "Oh Brother": "Who elected you the leader of this outfit?" - "I'm the one capable of abstract thought".

Critical thinking: concepts and tools; Psychology of Intelligence Analysis; Mega-creativity; more books recommended, along with mentioning of many disciplines of study.

Emerging fields of study: visual analytics, something the speaker's very much into.

I thought he was done, but I was wrong. Apparently, stuff being talked about here has much to do with the DHS - the Department of Homeland Security; a movie clip by the departmental head of DHS was played. The guy's nice, and the message predictable. There's collaboration between PNNL and Stanford - Pat H. is involved. More people thanked.

Top 10 challenges with Vsual Analytics (from a book the speaker's written, which he referrs to profusely): human info discourse and discovery based on cognitive aspects of critical thinking; new visual paradigms; blend computational, sensor, and visual analytics for hypothesis; blah.

The speech ended with: The best is yet to come. (Again I duplicated the actual color on the slide.) I agree with the spirit, however it's an odd mix with the DHS.

Now Q&A: a director of some agency with a long name ("subject to change at any minutes") stood up and kissed ass (i.e. there was no Q and the redundant A was "thank you"); guy's asking why not many universities are involved; how to redirect compiler people and parallel processing people in this direction. Redirect? These people are not to be directed, much less redirected, and much less redirected by the government. Now comes an interesting one: how does such research work on an international basis? - Q by a guy who is "not a U.S. citizen". Darn it, such people do exist. A: working through it, takes energy and time, etc.

Now it's officially over.