Tamara's notes, VRC05 workshop These will be hard to make sense of without also looking at slides, I mostly only wrote down what they said in addition to the slide text. So the most extensive notes are of the discussions etc. ---------- beyond moore's law, moorhead turner whitted bodi, smith, rhymes log/semilog, freq domain hanrahan: 1880 lalonde invented this "from the time i learned to use a slide rule, i wanted to visualize" tried to redefine graphics, not just vis, in 87 meeting. domains came into graphics. application: data/models tie-in between viz and tools smith charts. what we need are more of these. [me: plea for visual metaphor design] necessity, convenience, commodity ---------- chuck hanson need to define audience. impact? excel biggest infovis tool. integrated sw environment. not a solved problem, but might seem like it from the outside. report: collaborate, infovis. wrong emphasis. visual thinking, how it can solve problems that cannot be solved today. scenarios need help. kill google for data one. science vs hacks - for community, not put in report. ---------- mike halle gabriel lippman nobel prize for color photography. breakthroughs don't happen with dumping money or more hacking at keyboard. happen in shower, due to personal experiences. [me: google as exploratory tool. not as "search" per se] VG not going to come from us, won't be directly funded by govt. [me: but DL funding *did* fund google!] can't open flower buds. traditional govt services - security and safety. traditional govt roles in research: basic research: industry won't touch, too risky standards: after research is over. level the playing field. keep from reinventing without stifling ip. can't do AI until write Common Lisp. don't want everybody to write themselves how do you give confidence to the people that will have these ahas. visual thinking not from vis alone. fun - learn from game industry. building worlds. want kids to aspire to. telescope/mag glass - science tools that are fun. vis is tool for next two generations that will be their microscope. ---------- steve feiner other modalities, and their coordination more sources of data, computational power, displays and little not big. not just moore's law to big honking things. put together lots of little probes (traffic example) dumb, low bandwidth. and see all over. prosaic not profound. siphon big money from science. democratization - not only science/lab/highend, but all/anywhere [in favor of umbrella approach] ---------- moorhead: is vis a CS discipline chuck: vis has been neglected. other funding sources. [me: where are the boundaries of field. torsten workshop: expansionists vs isolationists] halle: purists not isolationists. ebert: scepticism from other fields if too expansionist. credibility problem. ambassadors. bullit: if you're a purist, need clear prongs to other fields. like good database management. need to be new ways of evaluating so we know if it's useful. halle: previous report like that: not just building a supercompouter for vis. break between CS and vis to split. interaction and devices. all hw and interface smooshed into too small of a space in the report. foundation for that, just like itk foundation for segementation and registration. narrowly focus is vis interface toolkit. always reinvented, takes forever, cna't compare different interaction techniques. paper could compare techniques head to head. terry: theme of vis for mom and pop. (turner - that's an afterthought, my head still in engineering/financial). rosenblum - don't think it's an afterthought. sensorized world. report shouldn't just say transition, need exciting new things that nobody has thought of. turner: info processing and sensor networks symposium last week - not automatically process, but aggregate and present to people. vis as afterthought, info processing is real key. models for process - sensors don't tell you anything unless they feed model. how do you translate that into visual representation. for mom and pop it's red light or green light. terry: pda now great, newton was horrible. same idea, better tech/implementation. govt bad at bringing things to marketplace. what is nurturing role that we're supposed to ask for. ebert: we still aren't meeting the needs of scientists. much less masses. severe weather - plot on paper and draw with markers. moorhead: mom and pop can understand tornado box. ebert: but *forecasting itself* is still 2D. lives have veen lost because 3D structure not understood. kwanliu: mission to help scientists not mom and pop. bullit: neurosurgeon, crying need for visualizations that don't exist. can give us explicit examples. don't know how to solve this with vis alone. part of it is that as vis changes, can do new stuff, that can drive back and change needs for visualzation [me: research cycles!] complex and rich fields. me: can better collaborate if define own fields. halle: neurosurgery vis lab, not vis neurosurgery lab. chris: still havne't gotten our research into the hands of people who can use it. still go to app domain talks and think "my god, i can help". we haven't done a good job of engaging application researchers. at Vis conference, downloaded dataset of CFD, no idea of provenance and task. long way to go to engage applications in all fields. [me: specialist vs masses, not science vs mom/pop. or clear need vs vague need. task-oriented, can judge/evaluat success if crisp] terry: chuck said excel. also matlab. now way beyond math. we didn't invent it. never see matlab papers at vis conference. most of vis community. lowercase v vs. uppercase V as tmm said. lorensen: purism means knowing what you can do right. all the vision people do segmentation. interact with other communities. we don't have good definition of field. totally diverged. papers on everything at Vis. with collaboration, govt can make a difference. big stick. it's hard to do that in multidiciplinary fields. 70% social, 30% technology. govt can pay for that social atmosphere. can't say we're going in with new segmentation technique, when computer vision guys did it years ago and we don't know about it. moorhead: deborah silver was doing both seg/vis, even tho old example. lorensen: sure, need overlap. all top techniques funded by industry. ray tracing (bell), isosurface extraction (ge), volume rendering (pixar), brian cabral lic (sgi). no cabral was at livermore. came out of customer-driven need. didn't just make triangles because was neat. need customers to make you innovate. tech push doesn't work. [me: place for some push, but also need pull] keim; already multidisiplinary in infovis. hci guys only want user studies. graph drawing algs and complexities. defining what we really want as a paper. we do integrate. different flavors of papers, not everybody likes all papers. hci, GD, algs, lots of ingredients. border to applications. kwanliu: in next few years, how do we do multidisciplinary work while keeping our discipline defined. ebert: report to last 15 years. or say make vis solved/ubiquitous in 5 years. solve and move on. colleagues who say aren't doing fundmanetal if can be used in next 10 years. if we make dramatic impact in next 5 year different goal than 15. today with apps - trusting results and knowing they're right. would be very concerned to rely on vis package if something being done. bullit: crying need for vis of expected error. crj: practical level - in terms of funding agencies. for a while in NSF, vis was lumped in with CG. there is no vis entity within NIH. one-off projects. nsf large-data vis call. nih understands importance of vis, but no program. do we want to have an identity within CISE? do we need identity within nih or are we just facilitator. terry: is vis a CS discipline? if not, then CISE isn't best home. has been home so far, for last 15 years. interdiscplinary office at NSF, not tied to particular directorate. is it better placed interstitially. [me: no!] retain identity while reaching across aisle. if it's a rec, we should know that before we leave. rosenblum: other approach, people on .edu list saw call that math/cs call. math has spent a lot of budget on math/bio, math/phys, math/cs, etc. they spent lots of their budget on collab. could have cs budget with matching between others. from funding point of view, nih nsf very different. nih budget just doubled. some real differences. nsf boot camp. time-series of funding levels for research. bars of who's who. now big huge nih bar, more than half. nsf holding its own [me: get this chart]. terry: change in emphasis not necessarily request for more money. rosenblum - thought that was good. peter sees this as played-out field. if you say "have been doing this for 15 years, now need to transition". somebody else might kill, new priorities might be more important. it's been there for 20 years, not succeeding in real world, why fund? lorensen: can't stand on our own. vis and X. [me: for all X]. vis on its own, that time is gone. contrast first page of this report to first page of nvac report. style, impact transition should be handed off to whoever thinks they can make money doing it. scalar volvis solved problem. [me: transition used in two very different ways.] halle: example. before egyptians, no blue. aquamarine came from process for making glass. aquamarine is tiny little space. all the different styles of expression in art. small and not so many innovations. but impact is huge. can't paint lots of stuff without blue. is there a reason to have new blue, new advances in photography? yes. feature overlaps. future of weather may well be about new vis techniques. nothing wrong with that, as long as i know where the line is. funding to paint with watercolor, not for watercolor. [me: this is way turning into research cycles] hauser: are we externalizing ourselves to the point where nobody needs vis expert anymore. will this be true for all of vis in the end? we're not there yet, all problems are not yet solved. time-dependence, uncertainty. but in 15 years, will nobody need vis experts anymore. [me: work ourselves out of a job] ward: one of the problems we've been facing is that app areas have people who fancy themselves as vis people, but they're hackers. we're hackers in their domain who don't understand data. they think they're doing vis with rainbows. they don't see why it's a visualization. he goes to sgtatistical graphis meetings regularly. mathematicians who are plotting formulas. don't think about visula complexity, colors, perception. don't think about interaction. they live in static environment because of how they publish. hacking aspect both ways. terry: umbrella/role discussion in feb. microscope/telescope discussion. tool that everybody learned how to use, even if don't use every day. new idea for improving resolution of light microscope by order of magnitude. would have huge impact. is vis done? absolutely not. harder thing to sell, no funding [me: but microscopy is huge area with lots of funding! even if not optical stuff] multivariate infovis problem, not just a vector. macroscope. tmm said fooscope. let's get away from optical to radiotelescope. basic beyond moore's law helwig: react to matt with question - if somebody has a problem, like how to improve mobile design. might need vis and sim. for sure would search for sim expert but do vis themselves. matt: same reason why html makes people think the could be web page designers. joy: on nvac. we're beating around bush. grand challenges. very easy to define in homeland security. drove whole document. we need to define grand challenges that drive field. bill has been trying to say. clients not individuals, they're problems. define the driving applications that drive for next 15 years. 87 report had great thing, we could see inside something for the first time. all of a sudden medecine etc. missing from this report: definition of vis, grand challenges. if can't define, can't be purists. if we don't have grand challenges, then we are played-out field. every other vibrant field, here's the grand challenge. [google: exploration and explanation] ------------------------------------------------------------------------ research cycles [my panel: rosenblum, hanrahan, schroeder, maceachren] not: shneiderman, nowell ---------- hanrahan: [must get slides] 11:15 start science of analytical reasoning how people actually reason with data factually and analytically presentation and communication considered equally important as exploration. defensible, can show reasoning behind discovery. lots of HCI types. people presenting problems they're trying to solve. exercises from app domain people. analytical reasoning - human side of the equation vis - us data rep - CS analytical reasoning - ribarsky and card mainly. collaborative - groups vs individuals. wikis, etc. scale is huge issue - many sources, many people. not just big size (n). vis rep: hanrahan, eick, chuck, maceachren interaction: stasko and robertson much bigger issue than vis. lagging in many ways. expensive to build systems now. shouldn't have to write lots of opengl. new visual paradigms information spaces. (visual google) photocollections. abstraction - pat and jim. infovis - discrete not continuous 5 that survived. initial list of 10. data rep probabilistic reasoning, text processing, data mining, comp statistics - all well funded how to couple with interactive systems makes everything harder, needs to be redone. vis is not so good at generating really complex displays. run lots of algs to make simpler displays. google - 100K computers to make list, good example. dissemination big issues - DHS as first responders. doctors, firemen, people who run into buildings. need info, stuff that works. will throw hmd on ground before running in. interest in rhetoric and how people present information. intelligence agencies - do analysis and make 10 different presentations: general public, boss, president. time-consuming for them to do this. all their time making visual materials, not doing analysis. real problem. move into practice eval: laskowski, plaisant, etc. good summary of state of the art. released within a month. bunch of recommendations. funding committed. lots of interest. convincing argument of critical mission need requires basic research. lorensen: first chapter is grand challenges. hanrahan: hard to do - distill all challenges into a series of grand challenges. lots of effort. ken joy - had 10, distilled down into 2 or 3. hanspeter: how many meetings? two big ones. pat had 50. aug/oct. report more or less finished in january. kris cook did a huge amount of writing, fulltime person. ward: fewer grand challenges - constraints on bullet items. hanrahan: everybody has pet grand challenge. finidng repressentative ones. reasoning time-critical, etc. do challenges capture the more fundamental problems. narrowing list from 10 to 3. quite different kinds of challenges. rosenblum: collaboration with other agencies. what do you envision with nsf collab? pat: not it to answer on that one. nasa, doe (heavily involved). don't know about nsf. turner - organizational. huge amount of overlap. that panel was very focused, this is less focused. hanrahan: this panel has luxury of unconstrained report. that one was goal-directed. ebert: even though specific tasks are related to homeland security, process is science - hypoth, test, defend, present. crj - doe corridors report, not first time ideas. cite the other reports. jim very much interested in having larger group saying the same thing to govt. more basic science will get done if speak in one voice. tell senators the same thing. hanrahan: no problem to convince biologists etc of goal. analysis and vis. human reasoning and analytical techniques. new-ish. coupling those two things into process. computers do what good at, people too. ---------- rosenblum. 6.1/2/3 cycles long time. ONR straight science until 1990. success = nobel laureates. 90s + now: help military not basic science. hard problem, not stovepiped. different people running. beyond scope for us? just say not stovepipe. liked eval topic. multiagency practical - leverage (as long as not too small). if really are no ideas, say so. give up on CISE? [me: no!] don't call infovis impoverished given nvac. nvac fast - 3 years. maybe nsf longer term. not just medical - want a dozen issues. ---------- schroeder. was gonna make outrageous statement, pat already said it. vis is thinking. you sketch to think, to present. all disciplines. resort to pictures and abstractions to think and communicate. make this point to funders again and again. vis is thikning. if you do something and have no vis, project is not complete. tail on dog. extension of X. sometimes tail chases dog. gain understanding, learn how to obtain more info or how to reason about things around us. put in bottom arrow. funding cycles (small business perspective): research wants to be driven by problems. needs marketing. ripe opportunity. embed vis in all aspects. identify problems - build interdisp teams, communicate. vis results in these things. change thinking process about where vis sits. NCBC funded by NIH. project that accomodates basic research, engineering, transitional. driving problems: schizophrenia, etc. algorithms experts. engineers - if you want to do this right, here's how. architecture, data/source open. basic/engineer/domain. vis is thinking. get this point across. open access to data - critical. hamstrung by inability to share data. open source is second. ---------- maceachren. token geographer at all of these events. background in cartography. visual thinking going back to thematic mapping. 1700s. not a new area. lots of history. lots of mistakes that cartographers made as they tried to turn themselves into a science, especailly looking for users. the first time around was in the 50s 60s. communication model of how maps work built on information theory. it was just wrong. didn't make sense. it was too tied to low-level perception, too driven by user studies that weren't grounded in what we would now call cognitive science. do like idea of broadening perspective out in both directions. both basic fundamental theoretical principles how humans think. as much a cognitive scientist as geographer. one of the things that has been missing, hit on well in hvac report, visual analytical thinking getting as not just how to they see displays, but how do they think with them. a lot of the basic psychophysical answers are there. but we don't know hardly anything about how people think wit hhighly interactive visual tools. broader vision that what was hearing with user studies. think carefully about which domains to connect with. no disagreement with those in report, but need to add more. he's on fringe as a geographer. no strong opinion one way or another as how tamara position: purist. cscw - visually enabled group work. hard problems solved by groups, but our tools are targeted at individuals. semantics/ontologies - how meaning is created. some funding programs try to span all three areas, it's hard. mixed success. nsf digital government has pushed researchers to ground in realworld practice. tension in that program is how do you do basic research and satisfy nsf that you're doing basic research and also satisfy collaborators that are solving realworld problem. have been some successes in terms of basic research challenges progress and impact too. also arda gi2vis, even harder to be successful given two year grant. some parallels exist between research cycle diagram and human cenered systems and user centered design. how to borrow from those domains to learn what problems users are really trying to solve. building a perceptual cognitive model of how people think with visual tools. contract work from govt agencies. build small exploratory vis toolkit, with epidemiologists. detailed case studies with epidem, figure out how they think and build in. halle: - visual encoding as vis specific basic research. agree that vis is taking a lot of stuff from other communities. filtering and taking little bits of extra info - what will said, perception and understanding of data on either side. fairly unique from pieces coming together. wrapped up into app specific info. vis peole have expertise but not unique. is vis irrelevant, if pieces taken out, what would fail, what would be worse, what would be better. funding case. if apps did what vis did without a discipline. more redundancy less taxpayers advantage for dollar spent, worse results. remove and see what happens is interesting approach. chris - what do physicists do to get advantage of small focus field: dont yuck someone else's yum. subcontract, provide to community. no less or more a vis person. crj: will/pat - in recent years, when went to doe/dod meetings, and prog managers said why do we need vis. answers like will/pat, more than half brains used for vis processing, this is how we think. more than half brains -> more than half funding :). because so ubiquitous, people don't think about it as process that needs research and tools. we need to make compelling case. argument of 'give us half the money' doesn't work. response to tmm second slide - fundamental research vs. work with other applications. both not either or. some carve out as our own (abstractions), algorithmic level. there are fundamental vis algorithms that need research. concentrate on them as vis researchers, abstraction, interaction. but also very important to motivating problems of high priority for nation and we show ward: agree to keep pure and applied. read report and quite concerned that no domain expert, no funding. grand challenge problems that go across lots of diffderent discipline areas, eg uncertainty vis, could be applied to dozens of potential fields. asked program director at nsf whether it would be beneficial to move from general purpose to tying tightly to particular domain. as long as have motivating text of here is where impact, not so important to have tight bonding wiht domain persson. alan - you mentioned uncertainty vis, he's not so convinced that it's as general as you characterized. different domains treat uncertanty very differently. maybe that's why we don't have a good general answer! maybe that's the challenge. pat then mike. mw: attribute of data like other attributes. alan - if you say it's general attribute, then how is uncertainty different than anything else? larry - don't agree. uncertainty not built into data. hard part isn't vis although that's interesting challenge. getting physics of uncertainty before you can begin to visualize. pat - facilitating thinking is better than just vis. reasoning/thinking with uncertanty is very different than just vis whichis how to draw wierd uncertain regions/glyphs. but thinking about uncertainty, lots of stuff in stats and science. we're fundamentally good at it. vis is not really working with us to understand at all. maybe sometimes can show error bars. domain knowledge. what biases we have. mike: some case to be made - innovations of vis that translate. different than saying can approach without *any* domain motivating it, working in either to do general solution without something driving you. can transfer really good research into other fields, more likely to be fruitful recourse. need to find a domain partner that subcontracts to you since you're helping liz: narrow deep perception of one single field. many of you broader and [BL: no depth]. there are things that are so badly needed. 3D visualization of uncertainty so that i can operate and if segmentation is wrong i don't kill nobody will trust image-guided surgery until they have that. PH: is that a problem in vis or in segmentation. liz: yes, understand where segmentation is going to fail. give me ability to say how wrong is it. likely to cross a number of differnet fields. have grad student working on, has outlined in thesis of other areas that it's applicable for. she would be that one of these ways that this whole recommendation can be stronger is to give four general things that she has need for, one is uncertainty. that can be translated into specific questions but then can cross fields. terry turner penny terry: only uncertainty accessible to public - storm tracks hitting florida, cones showing where might hit. best he's seen. oh, look, might be in danger zone. temporal aspect too. killer apps - congress/people/insurance. still have no good assesment. real estate values. map that to insurance, stock could go up and down based on storm tracks. or emergency evacuation planners. pat: no doubt that it's an important topic. but he thinks it's not just vis, mostly other topics. terry: invite to think how we look at daily weather. we need some grasp of that. pat: but vis don't help us. easy to show, doesn't mean you believe. actively confusing. book on risk communication. if you present uncertainty in ways that are incorrect (health), known vs. unknown, can get sued. really really hard problem turner: uncertainty is not single phenomenon. both direct and indirect. turner - charting uncertanty in measurement. very well understood in measurement. homeland security and through the wall imaging uncertainty of interpretation, timeliness matters. all very different, not single topic, very domain specific. penny: within a domain have multiple forms of uncertainty. circle back - in order to do intelligently, do need a base of domain. hard, large investment of time and effort in both sides. need to get somebody to talk to loony vis person and will use halfbaked vis tools. and for us, to learn enough about field. over the years, there's two kinds of vis people: marry to app area and become experts. and those wedded to ideas and concepts and apply to multiple app areas. what can we suggest to facilitate that kinds of partnerships. rather than say she needs to spend 100 hours each on each of these app areas. hard to justify. how can we make recommendations hanspeter, me hanspeter: foley said tech transfer is a contact sport. need recipient, personal relationship with domain scientist. customers not atending our conference. try to make conference more appealing to domain scientists. can make connections. [me: no one right answer. not just Vis] larry - tried very hard to get scientists into Vis confernce. brought molmodel people the first year. couldn't do it. they have their own. effort was made.] felice frankel - image and meaning. mostly scientists and communicators. penrose, el wilson, etc talking about how they use vis. vis fanatics. penrose would go on and on about visula thinking. same with comunicators. surprising that no cross-fertilization. wanted us to solve problems. lots of interesting knowledge. and ideas. separate fields helwig/terry/mike/lunch helwig: interesting question of why app people don't show up. are we vertical or horizontal field? we don't have specific market. many markets, enabling technology. on the same floor and building another center on telecommunications. one market, lots of problems - vertical. we're horizontal. if app people have need for vis and vis experts, they would come and ask. why don't they? terry: question for the panel to think about over lunch. larry esp. how do we bridge it? need interdisciplinary folks. don't throw baby out with bathwater. vis has not affected science, how do we go about this. one of the things they don't have is resources to approach us. crj/larry - big overstatement. volrender and marching cubes every day. terry: if i started funding biologists - you can have another 50K if yoyu work with vis people and give the data. will that solve the problem. larry: but might solve with joint proposals. pat: don't want to start with position that they're a bunch of idiots. they think we're a bunch of idiots. bad working relationship. classic way to create divisions. must go out of your way to pay attention to what they're saying. career out of figuring out how hacks work. there's some great insight, you can put onto scientific foundations. mike: we're going to be doing thta kind of stuff - sporadic high risk. lowhanging fruit to harvest. might be able to find. systematic, rule that comes to mind - people learn from people like themsleves. if vis layer is one person deep, don't have a biology person inclined to go to vis conference - crossover person who will talk to us, then it won't happen in systematic way. less hack motivating way for government - data starvation? then govt requirements for open data and accessibility can help. ------------------------------------------------------------------------ viz literacy, rheingans [me: belief -> assertion] [halle: sciam Mind, buy at airport] ---------- hanrahan: surprised and delighted that it's in report. two directions. overall visual literacy not that high. need to help. opens doors to thinking about vis more broadly. powerful/dangerous word - people have associations. reading - how to read a graph. explicit training. ross perot widely criticized for showing graphs, americans couldn't read. misleading - cat/dog/horse/fish seeing/observation - hard to learn to draw, hard to learn to become observational scientist. graphic design - style is important, not arbitrary. perception and cognition - cognitive most important. test for spatial ability, if low ability only use much simpler. herb simon - why is a diagram worth 1K words blueprints, etc: most advanced visualizations that we have created. visual literacy - people don't trust images anymore. symbolism - dutch masters "everything means something". representational realism. dennis woods - van sant - most realistic picture ever. literary criticism deconstructing this. impact - hubble space telescope. symbols of science. exciting unexplored. seeing unseen worlds. ---------- joy right: integrated lots of thing that he didn't get out of the initial meeting. wrong: audience. current document is too focused towards us. would you give to senator or chancellor? unification of fields? is that explicit objective of report? shouldn't be. missing: grand challenges. levers to communicate passion jim thomas did a visionary thing with naming it 'visual analytics'. not just visualization, something else attached. analytics meant different things to each person. driving app: driving. grid changes dynamically. started s/vis/data exploration/ in grants a while ago. agrees too much transitional research. no medical doctor will play with transfer functions. new visual tools. nonscalar datasets. [me: sure, very infovis-y.] decouple from siggraph-y origins. not just pretty picture. may be 2D plot. hard to interpret isosurface w/o context. vis is microscope. education is driven from below. students never the impediments. much reach lower tiers in educ structure. maybe not k-12 yet. definitely ugrad/grad. ---------- hanspeter technology - we're missing the boat. world is changing. cell phones becoming tricorders. wall displays, table displays. surrounded by lots and lots of pixels. too focused on rendering, not enough on interaction. talk at mit - vis people do not have objective measurements so the work is junk (machine learning guys). insights? no, what about actions. [me: chris north paper from infovis04]. terry: cell phone as tricorder. tricoder collects, not just displays. sensors are elsewhere. and small displays are hard to use. pat: all media have limitations, and craft is learning how to exploit those limitations. many of those things are low res, low fidelity. learn how to take advantage of media. biggest thing to learn with new devices. mistake is let's just think of device independent rendering like rgb->bw. have to change design and way info is portrayed. not about high fidelity. mike: back to question of what do we do. either we have the data or we don't. real multiband, they don't just happen. if vis partners with that bigger entity, could work. i don't see how "there will be sensors on taxicabs for no reason, and now we can use this great data source" will work. has to be driven like reason. don't see how vis can bring it into existence. applications that drive it. seems contrived. privacy issues. hp: mit talk: assistive tech for people with brain injuries. they'd forget bus stop, etc. using cell phone they could track person. machine learning technique signalled divergence from usual pattern. ken: deployment of sensors is happening. solved problem. data is there to be analyzed. we can have role in that. terry: you said us generates the most data. sensor data, or statistics or what? (ken: nsf report from a year ago). will: vis literacy makes me think of game theory. are we going to leverage the gaming community? kids more literate than we. maybe we need training, we're the impediment. pat: i don't believe that. talked to don norman. he's also against this way of thnking. for scientist, certain kinds of visual representation good for some stuff not for others. experiential learning of games not good for e.g. general relativity. analytical more subtle thinking won't benfit much from traditional gaming interfaces. supermario running around the web - people keep trying, it never works. you have to understand what visual representations work in what situations. good for certain things - maybe surgical training. but not knowledge worker activities. will: exploring the data. pat: you can, but... on being in the right space. in the right space problem is trivial. flying around the data is hard. if isosuface is realistic thing, flying around it won't help. in phase space, can find parameters that change from turbulent/laminar flow, that's what you need. just flying around isosurfaces isn't so useful. will: flying and interacting. simulating physics experiments. get sense of changing physical constants, gravity. pat: ok. helwig: mom and pop vis. who do we focus on. the more general our user community is (cell phone users = everybody) the less vis. no doubt that census collects lots of data, difficult to make sense of it. must be really simple for mom and pop, is it still visualization. advanced users. scientists/analystis. experts not general public. [me: specialists] steve: one word for many things. escape black hole, twitch game winner different from relatively specialist. speedreed not good for general relativity book. same thing for vis: domain expert vs. wait for traffic jam. both vis, just like both reading. me: specialist, crisp task. hundreds of hours on photo collections. steve - deep needs if not special abilities. nonscientist specialist can do. harris diagrams over 25 years. look like bus diagrams. sophisticated graphical language. specialist vs nonspecialist better than scientist vs. nonscientist. pat: textbooks to get sense of visual sophistication. biologists have great range of diagrammatic representations. invented to solve problems, both educational and reasoning/research. a lot of these fields are visual thinking fields. he's talked to people who produced those diagrams, they're using illustrator. their tools don't leverage visual analysis. vis has underappreciated visual languages. infovis people know how hard it is to do this. mike: mom and pop taxpayer. if govt takes in money and provides for common good, and you had to make case for benefitting enduser, is there really no ansewr? yeah, we could do something. might be outlandish or transtional, and later commercialized. visualize every recipe from epicuroius for apple pie, to see differences? experience using vis to teach holography. relatively esoteric. same effect that learning artists can beneft from painterly photoshop. exploration without fear, because can undo. can find new optics configurations that couldn't intuit from equations. had a lot more luck than would have expected. thought of as a crutch to help people learn, but made them into better lab people. report said simulations as something when real is too dangerous. when have grounding of real physical attirbute, all of a sudden realize you're seeing different perspectives of same thing. lots of visual thinkers that are latent. we didn't have these tools. what kind of cast have we been put in that won't hinder others. terry: speaking more broadly about visual literacy of mom and pops. when undergrad, computer literacy = write computer program. statistical literacy = must read scatterplot, stddev, etc. ken; thinking literacy - lots of discussion in academic circles of students solving analytical problems. pat: summaries of critical thinking book. huge problem, people don't learn to think analytically. but - we're not saying that all thinking is visual. need to know when visual will aid and what's the right format. and when should we just use text. terry: i can write complex equation. change params, pat: but knowing quartic is more fundamental then knowing params. intuition isn't whole thinking process. need conclusions that are correct. ken: and other mechanisms to give intuition too. pat: don't say are all equivalently useful. ken: vis as amplifier. that's how we should advertise. terry: graphing calculators as better than graph paper. chris: pat sez vis scientists haven't appreciated visual languages. short article about tufte at caltech, feynman diagrams on van. numerical analyst at cambridge: pde coffeetable book. he's been so frustrated teaching numanalysis with no figures, putting together book with math page, pix page facing. to give intuition. hans: publishers want experts to help them design book illustrations. scientist is too busy. disaster when editors doing this. alan: bunch of groups have tried to get visual literacy into curriculum (and failed) for a long time. if you keep this in report, need to go back and check what happened. 1972 cartographers tried, math national standards tried, statistics tried. would find in various other national standards. steve: good little ways of visualizing things, from ivan sutherland 10 unsolved paper in datamation. to understand circuits, little springs etc mechanical arrangements. that kind of transfer from one way of energy propogation to another can have nice appeal. turner: i get springs in terms of circuts. chris: question of audience. how many have taught? how many have written books? 3. crj: once came out with graphics book, it just boomed. thought it had positive impact in field. solidified what is graphics, people didn't know. codified. put into practice, moving forward. ken - research suffers from writing books. terry: posit that if couldn't do it without making book this thick. pat: then you haven't gone to trouble to distill it down. infovis and scivis on different planets. cartography, statistical graphics. perceptual stuff. one of the big problems is these things are not unified. need one umbrella. everybody could write a book about their course. but challenge is to unify. the people outside are doing better than people inside. our books are mostly on algorithm issues. mike: we talked a little bit last time, idea of field guide to vis algorithms which specifically is not coherent whole. this is technique, kinds of datasets, adv and disadv. have that as community resource. lorenseon: was book (keller and keller). pat: challenges as teacher is to distill into fundamental concepts. so field guide is cheating somewhat. people should be thinking about it. it's hard. mike: engineering deliverable in the meantime while waiting for science deliverable. kwanliu: come up with agreement of what to include in textbook. three at davis never use same content. lorenseon: how many copies of FVD? 82 first ed, ken: ieee tutorial on graphics sold 28K on graphics. largest selling tutorial for a long long time. market is there. lorenseon: 200K copies of object oriented book. vtk book 30K. tufte in millions. (shows limit). hanrahan: IT department at stanford sponsored tufte visit because they thought everybody should be visually literate. mike: felice's book appeal is beyond the people who take photographs of scientific objects. pat: steve: tufte books are beautiful. coffeetable books, never having read. can certainly imagine books full of scientific pictures. one could make coffeetable books of vis. but big fat comprehensive book on vis - if he was the knuth, he would be doing well. but split among four people plus don't make much money. ken: does it have to be a book? collection of web pages? steve: part of justifying it is the money. robert - alan, has gis tried web-based courses? alan: i haven't been affiliated with that. at his own university, making money off web-based courses in gis for nondegree professionals. don't know how much they've made. more than enough for staff of 12 continuing to go on, professional masters in GIS. penny: need book when talking to new person about collaboration. harder to do with collection of web pages. mike: no, collection of web pages is great, can be browsed. steve: describing a world where couldn't whip out small highres display and just show. ken: [me: rewarding repository building with 'academic credit', books, etc.] alan - university invested 250K in first pair of course. eventually got investment back. upfront costs. just a bunch of web pages won't work. mike: what hans said about resources appearing around us, missing the boat. graphics watched gaming become graphics, didn't notice until too late. could be that some group uses resources in "vis" and we missed boat. make money, touch people, etc. hanspeter: google maps are great. terry: smaller display. lots of people say not vis. windows systems not called os research. if you ask ma and pa what's os research, they know what mouse is. crj: panelists on spot, we need rec/findings. what is your rec? terry: k-6, k-12 have some pushback ken: want a literacy project at university including visual thinking. pat: 1) outreach part (educ, explaining how to use visuals). 2) broadening agenda - programmatically quite different. better undersatnding of cognition/perception. those two directions are different. image and meaning - what are you allowed to do to an image and still call it science? me: is the word dangerous? pat: thought fonda/kerry not science/perception. toolusing, like using excel computer literacy. so if do use, address in writing. mike: visual learning and understanding would be less loaded. his sense is that isolation/purity issue does come in. tread very carefully when step outside science and intersect with educationally. we believe have provided tools allowing opportunity for education in wide range. but be careful and partner with people making educational statements. will get ripped apart. evert: visualization literacy or visual literacy. steve: literacy implies being able to read different kind sof document: newspaper, rent bill, stock page. collection of stylized kinds of writing. ability to write as well as read is beyond literacy. so authoring is different. basic research big new ideas blank ---------- fastforward bullit: multiD plus 3D spatial mapping. fourth: beyond flat screen display, need true 3D. ebert: too scolding on lack of funding, be more positive. hauser ? who was this?: if they don't like it they throw it away. we should stay in CS research. halle: learning - don't talk about NCLB before impact in schools pat: vis is useful but won't cure cancer. chuck: good goals (impact, big picture). not there yet. collab and infovis both important, but... not only thing. concerned about interagency jealousy with nsf/nih. hauser: get rid of scivis/infovis discussion. co-financed i.e. industrial funding, not just govt funding. joy: tell the funding agencies how much money we want? 100M/10 years, or what? potential impact? keim: need to be clearer on infovis/scivis. omit distinction ok. or if leave in make more clear. infovis is not covered well enough. not right set of things. challenges: customer relationship management, marketing, network security, etc. includes gis. [jack and dan should talk :). jack said no interaction, dan said over 50%] bill: decide whether we're engineering or research or both. i am an engineer. and engineers do research. new interaction/navigation [me: zooming issues, f+c, etc]. [me: mine the first workshop a lot more.] must get good students or it will die. kwanliu: overstress societal impact - impact of vis will continue more on science/engineering/research/medicine. still room for innovation of volvis. rosenblum: lots of shorter vignettes instead of big writeups on three or four. "kamal said don't ask for funding"? vanwijk: spotlights need to be written by domain experts not vis people. vis alone will not save the world. vis dying, looking at feet, wondering about difference between infovis and scivis. bright young girl, 17 years old yet, not mature but has many directions in which to grow. mw: [me: don't like his use of the word 'hacking'. change to something else?] turner: infovis as catalyst. stats probably a discipline. vis, maybe not. pull not push. this point missing from report. nclb test scores will be misinterpreted. ---------- cafe definition of vis halle: vis speed dating. pat, keim, ward. analogy to stats is really good - provide high value. keim defn: mapping from data to pictures. taking data as input, picture is output. ward: for what purpose? presentation, hypth generation/validation/exploration? keim: any of that. ward: data is too narrow. could be vis of concept, or of relationships. keim: then what is input? have to get something. algorithmic point of view: what do you feed into alg? must be data. relationships can be data, hierarchical relationships can be data. but vis in general might not be data. computer-generated. me: computer-assisted not just imagination. pat: to first order like definition. process of mapping data to pictures. mostly about process. ward: communication. keim: interactive aspect. pat: then getting into use when talk about interaction. then things quickly get really complicated. dk: much more important for explore than for present. ward: so usually user-controlled mapping. keim: like in bar chart, click to get more. ward: how that control is granted to user is the interaction techniques. pat: but you're not just wanting to control vis itself, you're doing reasoning. analysis not just fiddling with mapping. three principles: congruence (mental model to visual rep), apprehension (easy to understand), affordance (as you manipulate, manipulate actual rep). natural mappings are the ones where doing stuff, dont' think about doing vis per se. mayaviz principle - manipulate info not vis. ward: porcessof immersing user into their data or information. more yo ucan have vis disappear and have user feel they're in their data, better the vis. keim: when people hear viz don't think interaction as well. me: interactive vis. keim, no not just that. then vis still main term. pat; discourse vs. interaction (dynamic queries, brushing, menu). but discourse is problem-solving, check data to see if it's ok. eventually lead to conclusion. what do we call this? ward: perceptualization. keim: somehting to tell people. me: multimodal, what's your take? ward: all serving the same task - building up mental model of information whether vis auditory, haptic. ---------- halle: art store analogy. make suppplies, sell them at art store, are purely artists, help artists figure out which suplies best for then, which teach art. people who make art supplies like algirthm designer. lorensen: no iea on defn. what he doesn't do. maceachren; not excited about analogy. viz as verb not noun. not label visual display wich is the end result, but likes process of visual thinking. art store ends up with final display. alan: don't have alternative analogy, but like to see focus on process of interactively using different kinds of displays in order to develop new knowledge. biased by cartographic background. they've been doing wha tmight be called vis for centuries, so they need to think about hard line between communicating and exploring. vanwijk: is it an insult to call a cartographer a vis guy? alan: no. some object to vis for geographic content since cartography is a good enough word. most of those folks not focusing on maps to help think, instead communication as end product. hauser: other terms come to mind - helps think about vis. sensemaking for example, or visual dialog with the data, or knowledge crystallinzation should be tightly connected. a pool of terms that relate to each other. bill: in one sentence at a high level: process? set of tools? jack: vis is technology. what goes on whe npeople watch picutre. bill: in this report vis is a X. discipline? science? process? mike: sensory conveyance of data. process by which data is conveyed by senses. alan: like general idea but conveyance makes it sounds like just delivery mechanism. but most of the times the person using tools is making the knowledge. when exploratory. bill: vis is not a source of something, it takes something and does something and produces. not like ct scanner that produces from something. jack: but ct starts with body. mike: one could argue that vis is enlarging the smallest possible defn of visualization. we can accept that it does involve. but some vis is unbiased. me: impossible! mike: you rpurpose is not to make judgement. me: one sentence, one paragraph. alan: nvac had once sentence, followed by two or three more. helwig: try to explain to students, come back to few important things - user, data, context of task. jack: vis is interactive computer graphics applied for data analysis. mike: biggest problem with visual analytics is that it takes the potential creative side that vis could include i.e. visual explression. but analytics is Not That Part. helwig: vis. bill: what does wikipedia say? alan: mostly targeted at funding agency people. how does it get defined asscience or engineering discipline, remember. then completely different. jack: vis is idea that making pictures is the only way to communicate with superior visual system etc. mike: feiner: like to see more sensory-modality neutral, not just visual. and other senses. disclaimer. helwig: vis is way t oenable efficient access to data. present data efficiently and understandabvly. from pov of presenter might be efficient. bill: transformation of data into form that promotes underswntading. feiner: otherwise you're showing off, user is dazzled/confused. coffeetable book might not come away with uany understanding. me: art not vis. steve: exactly. alan: we aren't creators, users are. ---------- schroeder, kwanliu. liz. will: vis is process of mapping information into sensory input. visual haptic sound touch taste. kwanliu: don't think you can define it. same as will: transform data into picture. concern about report is that we're including everything just to get funding. me: good, you;re a purist! where to draw lines. klm, will: how to draw lines. will: inherently a thinking process. liz: understanding my data, and that to me is a huge thing. will: includes interating and havin git pushed at you. klm: not claiming designing interface for cell phone is vis. focus on data. liz: if it's understanding my data, not just viz. me: need human perception of data in there. kwanliu: depends on purpose - expository or exploratory. wanna go there or not? (infovis vs. scivis) will: yes. kwanliu: not for funders, but yes for us. will: but makes sense to avoid confusion. for me, scivis is spatiotemporal where have space/time context. inherent reference frame. immediately relate, no need to explain. liz: what is infovis. at each point in body i can give you 150 attributes. will: crossover - if reference is body. kwanliu: in scivis often doing infovis. me: need to talk about the past in order to kwanliu: ok, i see. 2d or 3d. liz: space with multivectors. will: yes multimodal. transformation of information into sensory input. transformation may also include human input and interaction. kwanliu: personally focus on visual. liz: we're testing out haptics again. not simulators, but surgical guidance as tone or forcefeedback to show uncertainty or something like that. integrated not separate. ---------- crj, ebert, chuck, thomas, hp crj: hanspeter: short and concise. one sentence. chuck/david started out long, then shrunk. turner: put up a call for definitions. assignment for tomorrow is to come back with one-sentence definitions. don't need education panel. crj was too busy to do computational science. ertl: vis is thinking. no, visualization is graphics that's useful. ebert; imagination is not part of vis. it's analytical thinking. cognitive process. crj: multipieced. algorithmic, effective mappings we make from some space to visual spcae we understand. software algs - speed, usability, hw. 3-pronged process communicating. sw/alg, map/cognitive, hw. theory, sw, hw. ertl: bad category - theory in algorithms not just perception. crj: user interaction, software, hardware. ertl: transitional/basic/applied maps to this threeprong. [me: no!] crj - surprised nothing we're hearing about graphics hardware. turner: there is research in it but vis is not driving it, just along for the ride. chuck: i agree. we sholdn't drive it. but there are useful things we can do along for the ride. crj: we should figure out how to do multiuser game that gets the geatures we want to we can help manipulate graphics hardware. hanspeter: that' smy panel at siggraph, what's wrong with apis we have. crj: bandwidth off bus is so slow. turner: don't make up with this. multiuser games are nothing omre than instant messaging. when they realize it's the im medium they'll start supporting it better. i want to navigate through quantum dots. ebert: hard to do. design etc. turner: take hardware designed for game and do cfd on it. ebert: some things map well. luckily we have some hw manufacturers that are willing to listen. turner; hw is running along at a rapid pace. in 87 hw had so far outpaced application it was an attempt to get scientists caught up to graphics communicty. crj seems like display going faster than hw right now. but most of the things are low resolution. is it a driver problem or what? turner: how do you convinced somebody they need a wallsized dispay? you build them one. starkweather built a 3x6 display, put t up seams and all. but for gis it was great. step back see whole picture, walk up and see every tiny detail. the only other person thta gets it is bill gates because he has so many in his house. lots of interest. crj: on friday in san diego because of microscopy advisory. new algs for electronc microscopy. 8kx8k resolution they have tiled wall. neuroscientists coming over all the time to use. zooming in and zooming in, have never seemed that detail before. big revolution. turner: off the shelf parts,it's cheap. ------------------------------------------------------------------------ terry wrapup: pat said need call to action. haven't heard yet: what action? jump - where? how much money and when? missing case studies and success stories. what are they? [me: cafe topic for tomorrow: success stories]. where are europeans going? [jack: showing you help economy is important] what motivates americans? being afraid. ken: impact! short/long versions. yep. 2-page exec summary, findings/rec. first chapter: 15 pages. whole document: 100-ish pages. apps, success stories. 1/10/100. success stories. impact. do we want to make america more afraid? gick. pat: reason we're arguing for this is that somebody might be a customer and pay for it. have to line up customer and fund it. give and take. if we get a good message together we have an opportunity. funding is opportunistic. that's how it's been in high-perf computing etc. not enough to just say impact. hpcc grand challenges: Free Money. hpcc inititive came from peace dividend. new evil empire post-wall falling was japanese supercomputers. hpcc communication tacked on as afterthought, pat: question is, who would fund this? terry: spiralling cost of health care is big scary issue. didn't cure polio with iron lungs, did it with viral agents and vaccines. doesn't cost that much and saves lots of lives. pat: then focusing on how vis is a tool to help medicine and biology would be a tangible report. other things could also be int there, even if not spotlight. time to put these techniques to use. budget doubled because cost of health care scares people. it's taken two workshops to get us from national focus instead of collision detection and mesh refinement. stretching whole group up to high level as taken a while. me: kwanliu. 3 pat, liz, me. terry: hp, mike, ebert: helwig, daniel, thomas, jack. ---------- cafe summary process of mapping from data to pictures interactive computer graphics applied for data analysis process of interactively using different kinds of displays in order to develop new knowledge transformation of data into form that promotes understanding process of mapping information into sensory input thinking computer graphics that's useful sensory conveyance of data cognitive/mapping, software/algorithms, hardware perceptualizaton vs. visualization other modalities: aural/haptic. sensory-modality neutral defn? purposes: expository/presentation exploratory hypothesis generation/validation/exploration computer-assisted, not imaginationn interactive, user-controlled 3 principles from NVAC: congruence, apprehension, affordance art store analogy: those who - make supplies - sell them at art store - are artists - help artists figure out what supplies best for them - teach artists viz as verb not noun process vs static end-result picture defn for us vs. for funding agencies / policymakers sensemaking, visual dialog with data, knowledge crystallization process? set of tools? technology? discipline? science? transformation vs. data source improve vs. omit scivis/infovis historical distinction ======================================================================== sensors/financial jack, daniel, [stephen, leland] pat - jack's first two negatives he considers huge positives (simple data). PIM is grand challenge for AI, after all. he meant not timevarying multidataset. pat: but consider all info on desktop, and you're working with a group. integrate with larger community, then the world. no clear boundary. so it's not curing cancer, but can have huge impact on the world. something we can do in the near term. plausible in the next five years to build new desktop. halle: google has cured somebody's cancer - helps them find information. every day :). ---------- dan: report missing enthusiasm! scivis feeling glum, no mission to accomplish. we want to change people's lives. they can buy into. like email. he needs better interface. hundreds of emails in inbox. maybe #152 is important. many folders with misclassified data. full search is not what works most of time - server doesn't support right, etc. not an easy problem, many of you have it. either just organize by time, or if better difficult to access. gmail doesn't solve full problem. this is inspiring him to do new research. if we have this type of vision, main point we're missing. logging/sensors much too tecnical, peopledon't relate. they're dealing with business. automated data generation is sensors. but for money, need selling point! likes term visual analytics a lot. nvac did a great job of promoting and selling as something new. cannot use the word "visual data mining" in public in this context. tight integration of data mining and vis don't want to do spam filtering visually - wnat machine to do it, then occasionally intervene. human judgement at critical points. business analytics: crm, marketing, finance networks analytics: monitoring, security importance analysis automatically, then show to user. network admin has 1K computers, display of port scans etc, can drilldown current companies for doing BA have poor visual interfaces report generation is state of the art. no interaction, maybe drilldown. excel for business analytics. tableau going in that directions. GIS typically involved. important where customers live. email/spam - interesting to see where it comes from. despite relays. differences to nvac? leaving politics out of it, strong relationships. different applications, join forces. very few people in vis community know the dm algorithms, so it's hard for them to help steer it. 95-98% of credit decisions based on computer analysis - classification. classification trees get very large, why did decision get made? visually construct decision tree. you can guide the algorithm. automatic + visual. homeland security, millions of people each year visiting us. georeference is important. oldest - roman time. maps of paved roads. complete street map of roman empire. interesting cartographic technique :). abstraction, stretch out streets as backbone. cartograms. rectangular cartograms - ooh, cool. email/spam - geolocation, paths, time correllations. told student to fix his own need. try to solve your own problems. ---------- halle, hp, pat, helwig, terry, kwanliu. me. turner. moorhead. bill. halle. halle: interested in apps, trouble classifying vis - what piece of it. based on how much vis fit each area. pure: vis theory primary: vis is major element - disk drive app enabling: makes work, but not primary piece enhancing: makes better, but not strictly necessary eg sensor network: start with enhancing, maybe. report should have pure/primary/enabling hp; why is matlab not enough. machine learning people keep using it. keim: interaction missing completely. don't want 2D plot to just see, then switch tools to solve th eproblems not enough for interactive analysis. hp is interaction helping them do their job. is there any empirical data that he can point to? with interaction could do so much bette. keim: must be. jack: want two things at once: big picture, then find some detail of interest. then have problem. keim: scatterplot not answer for everything, if complex relationships. hp: need terminology: MD segementation, feature extracton, clusering decision trees. umbrella of all of that is data mining. keim: don't are about umbrella name , analytics not bad. data mining tech term for clustering/classifcation umbrella. knowledge discovery i s process. feature detection pre/post processing steps. in kdd commnity data mining is technical part, knowledge discovery is process. kdd defines everything, including feature extraction. analytics is similar. pat: election map was phenomenal success. nyt had pullout section. communication info about govt and society to people. and communicating science to people, i sunderappreciated. govertnment is relatively commited to this. noble cause. lots of other agencies - if you do anything - flows of data. environmental organizations all enthused. lots of folks that want to make the world visible. they love visualizatio. that's not done so much by vis folks. alan has. bens has also. census data. 9-track tapes. hard to deal with data. they spent a lot of money on census, health stats, spectrum allocation. all there to be used - wold go over well in government if proposed as agenda. open government. visible government. halle: beyond government - visual society. california has countingcalifornia web site advocacy group to make info available, but web site pathetic. jack: in a lot of media everything condensed into sound bytes and three numbers. can confront with lots of data - if enable zoom in grpahicsally topically then really cold serve purpose enable people to make more sensible statements about their neighborhood. global->neighborhood. keim: must zoom. helwig: on emore coment to hp comment - interaction, size both re matlab. r and s cool - but can't act on large data. hp: our people only do toy problems so we're fine. terry; report from uncertainty NAS meeting. statisticians are stuck in 2Dland. dan said scatterplot isn't good enough. but re pat, visual languages are hard to invent. reason the statisticians use scatterplots is they're elemental. won't replace circuit diagrams. hp question is still germane. [me: visual metaphor design we're working on] solve problem that they can't' do otherwise. government wants to give data away. helwig: significatn piece of infovis work done by statisticians. ty: true, but statisticians pat: must ask why is that. reaction: you're not using my favorite metaphor. larry: seeing in lots of diferent areas. sonar displays - horrible. but those guys have been trained for 10 years. [me: training vs. novice/education]. you won't change old guys, less need, but new guys. same reaction from medical. better vs accepted. but accepted by new. bill: or info from one field to other - radiology to surgeons. crj: i've followed up with statisticians asked specific question. re bens demo, they thought was great wold like to use it. they can't get their algorithms into his system easily software barrier. his algs aren't good enough. [me: open software problem]. not outdated another place not made it to user community. demos only. pat: r is tremendous opportunity. alan - ? keim: starting with last - R is great. his students use it a lot. lots of techniques implemented in R as tools. works on large datasets. interaction is difficult though - not easy to do linked views within R. main discusison of training on new - i don't see any statisticnas saying scatterplots/histograms good for everything. they know don't want to do that on 1M data points. if noise in data, see nothing. they know they need diff technology. getting difficult to throw standard R stuff at it - arbitrary hierarchies etc. but they're careful about vis people because they always come with fancy 3D interfaces. pat: they've proved they don't work. jack: everybody should read - design attitudes colin ware 2.5 D. kwanliu; wnat to start with vis throughout - over/zoom, or data minig integrated. where we do vis. like using color, want to be careful. like text. 3D not necessary. turner: discussion of taking new visual metaphor and ramming them down the throats of people happy with old ones. for their own good. don't show sonogram to anybody above rank of commander. translation of metaphors. making them available to nonstatisticans. [me: we do too] bill: if statisticians are happy, great. but they have to descrive to others. pentagon weather guys. if they go to a general, should we fly today. they pull up charts and different layers. he wants to see sun/cloud/lightening. translational metaphor. dumbing down. pat: appropriate visual representation. ***** bill: radiologists, complex CT scans. 3D, high frame rate. they look at slices. but when they show us, they show 3D volume rendering. translate their findigs into someone else understand. ken: do they show eurosurgeon these things. liz: show some things in 3D - bones on CT scanes. jack: use for visualization - simnply for presentationcommiunication neglected far too much by us. lots of competition from stats, people are expert, so they don't need us. presentation - vis really is better. bill: urricaines. weather folks have all kinds of ocmpmlex stuff to explore. summary for public. band of uncertainty. terry: when we take data from one area and want to transfer from fields - those representations have to be intuitive. because they're not elevated. literacy, so thta people can look at treemaps. but not so intutive when first looks at it. threshold. metaphors we need to build, we need to be really aware of less expert. turner: not only that, sometimes don't have time to sit and look - must react to it, very different type of vis. consequwnces of misinterpretation. terry: ATC, etc. miek: why do people use matlab? because they can take their data and look at it. if they hear about another vis technique, how do they just try it - do an R plugin that provides a base vtk interface. then people will try. don't need to shove down throats. adapt to their world, not make them adapt to ours. [me: institutional issues in releasing software]. vtk doesn't have fron tend of R. crj: do have that in scirun - matlab interface, stick matlab code in. halle: do that some with slicer. crj: good response so far. halle: good government - good ma and pa app, how does data get to people. the googleization of information will hurt us. unstructured search, their algs get unstructured search just fine. grand challenge - if we're trying to improve health and decision making ability. sensor,population - for tsunami relieef. famine/drought/pollution or correllate disease with polluation. hard to visualize.big problems lots of data/sensors. public must make decisions on it. tsunami hit country, not a good time ot ask somebody to build vis program. health of world/humanity. common platform to visualize pieces exceiting. me: sw release - pot of worms with gold in it terry: reward structures, what we're up against. we don't accept papers on tech transfer. not being published. also getting shot down in review. at nih other institutes have continuing software grants for support. those programs do miserably. reviewers say not new, not hypoth driven. our own community killing it. pper reviewers blowing it away. liz: actually, those ppaers are being accepted if evaluation is included. look at micai papers, lots of those that say we are using software and here is how we evaluated it. with eval it becomes important thing. i'll pan too if you say isn't it pretty. [me: yeah, eval papers published] jack: also support. would love to read papers on what have we learned, what was good what was bad. example: libraries for UI, love to have. if your problem is X use Y. how to organize things etc. liz: cg&a. can publish something that's description as opposed to research/eval. terry: haven't seen so many survey papers. concrete example when it failed. author thought it wasn't worth saying. leonid zhukov said diffusion. next year in app paper showed same technique -marching least squares muscle fibers in cardiac dog image. showed spiral structure of canine heart .terry said - so, how much of your sw did you change to make thta happen. i didn't change anythign - you should shout that out loud! yay, go team! we should reward thta, not be embarassed. thought he would be shot down. we have messed up. crj: followup with jack - we all want the cookbook paper on cmparing UIs, and people aren't rewarded for it. academics, survey papers aren't rewarded. not intellectually significant research. larry: reached point if they'r eshot down if not compare in AR. jack: but can get lots of citations. so there is some brownie points there. ------------------------------------------------------------------------ healthcare panel [me: have slides from rosenblum, maceachren, vanwijk, schroeder. need slides from hanrahan, keim] oncologists don't care about imaging - only care if cancer gone. clinical trials can last 15 years. ross: distinguish biological imaging from clinical imaging. for many years, focus on helping doctors do jobs better. diagnosis/prognosis by experts on individuals. not easy to have an impact. resistant to 3D stuff. surgery/planning did have impact. biological imaging: organisms or populations, not diagnosing individuals. capecchi lab - knockout mice. transgenic mutants - bat-mice image-based phenotyping wiring diagrams of retina. currently "not constrained by data". tens of thousands of pixels in each slice 1TB of data. can't volume render it. tissues not represented by homogeneous intensity level. ken: liked structure: here's a great problem, and here's the role vis is going to play. ---------- bill five years ago realized couldn't exist as vis grou panymore. now combined with vision group. more emphasis on analysis and detection. open research - won't talk about it, but critical. navigatoin - needs to be in report "leave no image behind" right image to right person at right time digital human, again, dammit. ---------- liz bullitt casilab.med.unc.edu - multidimensional data with preservation of 3D+T anatomy - intraoperative deformation, changes over time - - we're interested in blood vessels. ca ncreate detailed blood vessel trees down to limits of sensing modality. can get statistics on variety of measures of shape - how wiggly is single or group of blood vessels. database of healthy subjects. 8 atlases. think they can diagnose cancer early because of blood vessel shape - they get wiggly. how to display multivalued vectors wih preservation of 3d anatomical information? huge database. where in space is distribution of how things are. [me: TALK TO LIZ] turned out stddevs too high in specific area of head. tissues stretched or lost. how to map to preop image. map/vis deformed segmented thing into operative field. odd surfaces. error. displaying two surfaces. interaction. endovascular surgery - thread catheter through vascular network to target. can't see. standard to project little puffs of contrast. flat image. all you can see is downstream of catheter. if you had 3d object perfectly registered. but when move, changes things so out of ground truth plane. don't like that. now in qualatitive mikey likes it stage. only one study by fuchs on AR vs. standard techniques, AR won. time is right. kwanliu: what do you think about NPR? liz: want useful. sometimes photorealism most useful, sometimes outline most useful, sometimes radiating lines. me: multidim data? two patients had failed to diagnose cancer in, both in back of head. turns out that stddev of tortuosity measures was wrong. realizing that anatomical distrib wrong makes her wonder. through pure stats would have been very hard, didn't kinow was teh question she wanted to ask. so maybe there are all sorts of interactiosn at various scales that she's not picking up. sometime do know what to ask stats guys, but sometimes don't. helwig: was supposed to give talk. suggestion to make use of cheap data - ultrasound, xray. lots of expensive data ct/mri. xray easy/cheap. volvis mostly focuses on ct. meaning from ultrasound much harder. liz: highquality like multimodal fuse with intraoperative ultrasound. current funded project. helwig: where is spot for vis in the problem. great to do diagnosis. and bring in infovis. esp for prevention. incl environmental, stats, census data. the future will be time-dependent, since living bodies are. integration with vision is very important. tried so hard to do good transfer functions for segmentation. forget it, will never work, not reliable enough for critical questions. good algs in vision domain for segmentation. terry: when started itk, said wouldn't fund vis, just image analysis. hailstorm of criticism. guido gherig said can't do segmentation without vis, gotta see what's happening. the reason they didn't was distraction from virtual endoscopy. they didn't fund systems, connected to existing ones. but sidesteps the vital role for vis. helwig: reminds of what daniel said in morning- tight integration. steer algorithms. volvis community should be less fond of beautiful images. slide of bernhard breim from dagstuhl03. golden rules - be friends with 2D. 3D important, but don't discard 2D. terry: ross has TBs in 100MB chunks. 1TB won't fit into main memory. large data challenge at nsf - can't come to us with anything smaller than 100GB. bigger than both visual humans put together. bill: bio is starving for help. crj: microscopy and imaging - 8K by 8K resolution microscopes. thousands of images. exploration and discovery, multiscale. turner: do you relaly have 1TB or just happen to sample at 1TB resolution. information vs. raw data. not just storage and bandwidth, also representation issue. terry: are faced with TB - can be compressed but not clear where and how. cellular anatomist at cell and cell distribution. visible pancreas at cellular level (10 microns). want the islets of lagerhans cells, they're scattered, want to know distrib within organ. but others care about vasculature or bile ducts or ... ross: issue is where is redundancy in data. given how early we are in field not clear where to go with that. terry: all the islets are different, whereas kidney is more uniform. mike: any compression problem - swapping/caching - quickly to get answer, hard when don't know what questions are yet. compression never works when you don't know what you can give up. astrophysics with huge photographic sensor arrays - all sky, lots of channels, in 24 hours. same kind of thing, don't know dataset. from high tech to low tech - even lower tech than ultrasound, digital photography hasn't penetrated clinical practice. transfer function question - parameter space visualization question. change detection/viz - where is it going. bill: important, but how to do it? can detect, but how do we present? halle: impedance matching to our own ability to notice change? bill: volumetric change in size. alzheimers very small brain volume change or size of structures. halle: would algs for exact volume in the middle of noise. liz: part of detecting change (my bias - segmenting and looking at objects) blood vessels are easier than ohter stuff. response to treatment of malignant tumors. they normalize. because they can put number, malignancy probability dropped from 95 to 11%, worry at 40% if have recurrence. but what if Q is how shape changes. will: they see lots of data from biological imaging - interesting as vis puritst - room for new data structures. nested volumes. amr grid misaligned. combinations of volumetric data plus imagery. all requre new data structures and algorithms. adaptive multires representations might be adopted by scanner companies some day. alan: for report - we're not saying much so far about potential of infovis in this domain. in terms of making sure people don't get cancer - screening, understanding why not working, why people not getting screened. make sure it's in report. population epidemiology stuff. dan: has anybody thought about including molecular biology. lots of dna, microarrays. no molbio guy still has ball an dkit in his desk, they now use computers for this. dan carr could be useful for this as well, stats guy who's doing really good visualization and has targeted general public, but also now applications in genetic structure. bill: molecular modelling and vis, art olson. was driving problems in 87 report. different level now. if you look at this report, what should be grand challenges. medical no-brainer. who will fund. biological imaging. something for everyone. terry: close with comment. if nih was in charge of curing polio we'd have the best damn iron lung in the world. who said that? former head of nih. (west wing). lewis thomas talks about difference between high tech of health care halfway medicine, no medicine. handholding comforting. internet and web better tools for informing patients. halfway medicine is "hightech". we can suppress immune system. real medicine is immunology, comes out a bottle, 50 cents. march of dimes was grassroots. hope for infovis/molbio/medicine. not elaborate technologies for surgery. liz: partially agree, partially strongly disagree. days of open surgery are numbered - split them apart. but for foreseeable future, need to deliver, even if perfect agent, to targeted are. have to get there somehow, whether percutaneously with needle or by threading catheter! helwig: yesterday in cafe session had list of grand challenges more than medvis. in fastforward app spotlights need to be extended. sim data, engineering. how will this get there? no panels on this! danger that this report only addresses subset of field. vis makes engines work better and better every day. picture, caption, paragraphs. grand challenges apps, grand challenges vis. useful split. a few grand challenges in apps might be sufficient but grand challenges in vis need examples. kwanliu: large data - size, dimensionality, [me: sources]. terry: for larry onr volvis funding - i know who got money. same for itk. but what about large data initiative. larry: you may not know we were able to interpret broadly and turn into AR initiative. ------------------------------------------------------------------------ comp science and educ crj: pitac. definition. slogan. third pillar: theory, experiment, computational science. [me: toread from pat: marti hearst paper on which search vis has failed] www.nitrd.gov/pitac david: review criteria change already happened - cg&a is a magazine for reporting on this stuff. tvcg publishes the right stuff too. need first year course required in science, like stats or lab technique. five week course at purdue. [me: need reverse image search, put image into google and get source.] unattributed pix on slides. computational steering - he's talked to lots of scientists who thought it was a bad idea ot change model based on picture. decrease cycle time. scientists get really upset if you take their carefully tuned grid and make it regular. ---------- ertl: gap between 1st grad course and publishable research. [me: huh? can't they just read papers? ] vis phds, too many? [me: job market great for infovis phds, jim thomas alone wants to hire 50.] disability stuff - braille. educ: IBR relativity, weiskopf. kwanliu: some projects have little interaction - a few emails. others require visits to lab, several hours. slac: student more than a year, working closely every day. fourth: had to read their code. [me: wow. first two are really short. ] pat: i bet those pictures are cool and hard to make. but how do those help the scientists. what is their question, what was their experiment, how does picture help them answer question? what is value? stress is how we do, not how we help. supernova, looking for shock structures. potential: see how tube changed shape. particle data - banana-shaped structure. one picture does not help, but interaction on desktop. in past just showed pictures. now giving them software. [me: wow, never gave software in past?!] ---------- matt: thinks education is critical. as important as health care. [me: draft matt to help with this section.] we'll have to do all our own programming if we don't solve this. imagine understanding structure if color-coded nouns/verbs etc, then showed clustering based on sentence structure, etc. education, nsf, army research labs, three or four other agencies. state money. has never had proposal rejected. intelligent tutoring. just say "education" and they pony up. matt doing computational steering for simulations that last weeks. [me: interesting, talk to him?] integrate the multivariate techniques well-developed techniques linking them to spatial visualizations. uncertainty: certainty, confidence, quality all names for this google for features. larry: surprised by statements he saw that could have been in 1988. matt: didn't have infovis in 88. david: sure, but people have been trying to link stats with graphical images for a long time. in terms of computional science, big change is interactivity. crj: so is your point similar set of challenges: scale etc? larry: what points does panel bring out of it? scale argument won't sell, except to doe. won't sell to nsf. chuck: big issues with scale to nih, particularly for large-scale imaging in biological sciences, frankly not sure anything will sell to nsf. terry: scale isn't just little-> big problem. means xray crystallography to membranes to organ systems. not same language or lexicon. ebert: important across all science disciplines. nano - quantum to nonquantum. weather: microphysical particles. hurricaines don't show up on global climate models. boundary problem is hard. that boundary problem didn't exist in 87 we weren't trying to int [me: pnnl wants 50-70 phds. straw poll. never, sometimes, mostly, always deliver software to users. no nevers - a few sometimes, lots mostly, several always] thomas: what do you learn by collaborating software? halle: history as something not just in past - but is this vis? lots of education is funded for public and schools. token person. crj: all over the map. k-12, museum, university. all of those things. common in nsf? yes. larry: yes, have to have hqp. halle: does that have impact on what goes into report. ebert: separate program too - educational directorate. cunningham's term is ending. halle: overlap is way to drag in others. crj: what from your presentation isyour key recommendation for report? ebert: tight integration of visual representation into the computational science instead of post-processing mentality. that will make CS a more powerful tool. thomas: engineering analytics - success story of vis in all types of engineering fields. larry's impression of 18 years of no progress is not right. comparative vis, eval of many computational results. kwanliu: importance of integrating vis solution into overall problem solving and scientific discovery process. not only vis process and sw/hw but also people - vis researcher, should be in loop. mw: two cents for keeping education as high priority domain. also tight integration not only of vis and computation but seamless connection between spatial and nonspatial. ------------------------------------------------------------------------ final session highest level finding: enabling technology that is necessary, but not sufficient. funding: vis, like stats, is a research field of its own. its home is computer science. joint funding between cs and other domains, or other domains including funding for vis as a supporting or necessary enabling area, is highly recommended. app grand challenges digital human - physiome situational awareness mapping the universe intelligent sensor fusion health of humanity and the universe change detection [me: don't think first five area really necessary the only ones] health clinical genomics epidemiological space exploration vs. man on the moon digital human - nice hard problem transcends spatial scales spatial scales, temporal scales, one application of digital human. turner: axis problem. things that are linearly enumerated situational awareness and management / change detection short-term / long-term themes that span. not applications pick climate, tumors - good driving apps with change detection. health of humanity and planet health individual public climate/weather visible government education security engineering science business sensors need killer app for this homeland security to tsunami. intelligent manufacturing can use to study climate/weather/public health/visible government etc crosscutting change detection situational awareness killer mechanisms, killer apps sensor fusion isn't a vis technology. sensor fusion - if appeal to dhs soldier in the field, bring in other agencies than nsf/nih larry: need to say vis not fusion for sensors, that's outside us. crj: way to better sell it is to think about - here is an important application that everybody cares about. here's the vis research challenges, if we could solve them it would enable X. here's something important, how is vis going to solve app problem that we care about? or other way: if we have Y, what kinds of X can we do? hp: keep message simple, tell story. jack: matrix for fighting cancer. just atmospheric - or earth, oceanographic, climatologists. halle: for the grand challenges - not more and more stuff, want compelling application. climate - hurricaines, storms, droughts, (temperature change). agriculture, food supply. halle: why do you wnat photoshop for scientists? so they can use vis. digital human cross-cutting. me: success stories vs grand challenges vs app spotlights bill: broad grand challenge - without vis, digital human will fail. way to pull all info together in a meaningful way. rest of work doesn't have value if can't get info out of it. if you can just take pill, don't need vis. man on the moon - if people hadn't done computational fluid dynamics, could they have gotten there? terry: 10 years from now people will forget tsunamis. people who do the job cannot do without maps etc. satellite photos of before/after. cannot be studied at global scale without vis. bill: compare to hurricaines that are studies and visualized. tsunami example of lack of sensors and lack of vis. everybody interested in climate/weather. ty: opening anecdote. halle: relief efforts start, logistics of planning is maps etc. how much water can i get to people in short amount of time. alan: comment about weather/climate example. out of the 87 report, weather and climate is area where lots of resources put in. stopped short of being able to help people answer questions. creating visual display, nice animations, understand how process worked. but didn't get out of analytical reasoning. making big leap from making pretty movies t ohelpint people understand problems. matt: description of success like that or applicability is good. but in grantwriting, learned that people are suckers for this - scenario of how vis is ubiquitous part of our lives and critical component ten years from now. if given enough funding. success story vs. app. list to attract people. then a few in detail. examples have to be visionary. crj: specific pieces out of one of those bigger people. need very specific piece. here are vis research challenges. could design airplanes etc. except for first one, not grand challenges. liz: if you're looking for grand successes - medicine has to be at top. mr and ct have transformed how medicine is practiced. it's a vis of the inside of body. go on from there. vis challenges that will, to an equal extent, transform how medicine is practiced. as new markers come in, as new ?/ percutaneous and endovascular techniques. advances will be just as great as mr/ct. *** bill: organization of old report broke up into research opportunities in science and opps in engineering. for each one they gave science and engineering. maybe molecular modelling. medical imaging brain struture function, mathematics. cartography. pretty good playout. security is not mentioned - new area. FE. then was stovepipes. now is croscutting piece. some cross horizontal pieces. keim: but cooperative / collaborative plea was in there - bring in the cognitive scientists. terry: challenge vs. area. turner said axes. space for describing this. jack picked two axes. steve: wondering about medical imaging - what were funding sources for older revoltionary tecnologies. nsf or nih money or what? or big companies? but things companies can't do on their own. bill: initial funding came from nih. later they dropped the ball, companies picked it up. steve: maybe a little bit too far out for company to do now, but thta makes sense in terms of research. bill: companies do badly at interdisciplinary. weather modelling has far greater impact than company could have derived. RM: the guy who funded vis5d is now at nasa. should govt be putting radar systems out there. commercial entities now make money off weather stuff. rewritten 5d and use for project. bill: companies don't do things for the public good. govt does. ebert: still ncar and noaa. ---------- recommendations chuck: what is diff between second and third. 2: interdisc research ken: first one (restructure) does not belong in report. chuck: agree - but fundamental research in algs as science needs to continue. and promote interdisc discovery. ken - don't want us to be compared to mathematical sciences. chilled by the statistics comparison. supplementary field not enabling field. comparing us to them is wrong. crj - did you have a bad relationship with statistician. halle: enabling technology that amplifies effectiveness of many others. crj: thing i like about stats analogy is that it's both enabling and its own field. fundamental research goes on, and it's used everywhere. at nih, if don't have stats guy, it's rejected. like the fundamental research problems and still enabling/essential. ken: ok with that above. if the like statistics goes further down. alan - stats guys feel underappreciated, don't want to put yourself in their position. pat: sure maybe underappreciated, but hey - probabblistic reasoning, machine learning - hottest area in cs. google hiring in that area. nih requires statisticians on grant. can't get clinical study through without them. we'll be really lucky to become like them. hp: i'm a good guy, i do good work, give me a raise. turner: value proposition. what is value of understanding data or phenomenon. sell it based on that not essential goodness. will turn it over to him when done typing. terry: tension between multidiscp and fundamental field. probability is still statistics. robert: tamara's diagram - far left and far right. finding vs. rec vs. supporting evidence finding: without me, company would fail. rec: i need a raise. evidence: here's why. chuck: what does it mean to build a national infra? crj: sw infra, funding infra = longterm. ken: funding structure long term. halle: must move their funding structure. larry: before my time - program officers who supported it at the program officer level. will they be happy getting musts from program director level people? kamal said peter freeman said no more workshops but i will make an exception. every workshop that year had generated a new funding initiative. don't do that. but last report lasted longer than we ever anticipated. can program directors who initiated this, are they in a position to come in and say they supported something that tells nsf what to do with money. maria said yes. frederica said yes, go ahead ask for the moon. chances are you won't get it. if parts are well justified and reasonable, we'll see what we can do. crj: we are trying to prioritize nsf/nih future investments for vis or multidisc that include vis. we won't say 10M/100M/whatever. we give you ammunition. must is powerful. must <-> should consider. turner: i wouldn't hesitate to march into somebody's office. larry: ok, then my name won't be on it. halle: if that's biggest question we're facing we're doing pretty well. chuck: need third one - fundamental research continues to take place and must be funded. enabling new interdiscp fundamental research long term democratization keep identity as CS vis research alan: looking at nvac exec summary. dhs vs. nsf/nih. recs are much more specific. old report: 1% of industry spending on vis should be spent on vis. turner: first statement says more of the same. what's new? where is the specific new thing. bill: continuing is bad word. halle: reason - we've been talking about datasets emerging that none of us have any idea how to attack. new applications. hp: what is new - analytics not in 87 steve: can we think of specific situation that because of magnitude/complexity, dataset that didn't exist in 87, and high payoff, and sitting there not being used and people are dying. ken: how to put new into there. new comes in from new problems that we must address through vis. the fact that our field has outgrown problems that it was designed to work on. resync to new problems of today and new technologies of today. in hands of everyone. end scientists end users. [me: democratization] crj: first couple sentences - info big bang. 87 pre-web. steve: ok, lots more computers and data. sounds like all you need to do. not continuing. terry: sound like dissing nsf. matt: sources of data don't have needs. turner: do not use words challenges and opportunities. steve: address problems or reap benefits. opportunity to do something really good. we can reference report. [me: 87 vis getting invented. now vis critical enabling] ebert: science has visible successes. dhs successes are invisible. nsf has been doing turner: give nsf credit, and want to say redirect? [me: huh, no] crj: one thing that would be good use of email time is concise findings and recommendations, work on those until we all agree. primary and secondary. then writing report becomes much easier. ---------- larry: what will and won't appeal to nsf. what can happen, other than gather dust. 1. area study just been initiated. freeman looking at everything and reprioritizing. so something seen as 15 years old and not doing much is at risk. so make case that it's dynamic with new challenges. if it gets out in time to influence the area study, first round in this week. probably 3-4 months for action there. 2. other divisions - can help force interdisc at program officer level, maybe to make link up. 3. at division director level, mike forsman taken over for kamal abdali as director of ccf, has ability to move funds around. 4. if peter freeman decides is great, can get something big to happen. what report has to have, talking mre or less generically, to succeed. 1) must have new areas. not just scivis and throwing out some infovis. my own interest is sensor vis, given ties from tsunamis to homeland defense. can span location-aware computing, building up information about where i am. sensor case is very multidisc. not in 87 report. second high impact is nvac card. nvac has short-term crisis things to do, will focus on next three years. manhattan project of security. in meantime, somebody has to lay out science for next and next-next generation too. halle: is that a good explanation for long term vs short term third point? no, fundamental science support. infovis, evaluation as new things not in 87. vis integrated with other fields. argument for scale in sense that terry presented was good - scale across problems, not just bytes of information. a little bit of gig to tera ok, but don't play it up. only doe cares. take a hard look for other areas - old science, 87-era. what's new, what has to be continued. but they don't make strong arguments. educational component adds value. would be good to get a real educator so we're not saying ridiculous things. i.e. blinn work since late 80s. has anybody proved that there's pedantic benefit to that? [me. heh. pedagogical. c'est toi qui est pedantic.] finally, lots of missing topics. take the nsf directory and go through science divisions. put in word networking in report. put in security. etc. need a couples of sentences from each area. enfranchise everybody. ---------- larry's handwritten notes: sensor visualization -AR - interfaces w. mobile sensors - reconstr urban environments nvac -if nvac agrees argue that nsf needs to support the long term, basic researc for next generation and generation after next infovis/eval/others as new vis as integrated with other fields is important scale in sense of terry (fusion of areas) is good. gigabites to terabytes an be mentioned, but is weak argument. education component adds value - keep expand may need education expert to write where is network vis security vis etc - must be ... similar for wide range of nsf interest: math phys, bio, engineering what impact at nsf keep graphics alive with improved funding at nsf interdiscplinary activities +- ups at division level cise-wide nsf-wide impact ---------- bill: you should present something at vis. bof if not workshop or panel. everybody in room has sent comments. wow!