Oren’s Weblog

Sandra Braman (University of Wisconsin-Mulwaukee)

Why does it matter?

- Research success is mission critical for many institutions, which means you need researchers. The Ability to do work is central to researcher identity. Serious researchers are willing to trade salary or switch institutions to improve support. It’s clear to researchers that IT are collaborators, not just service providers.

How do we know what we know? Direct and indirect info from national reports, conversations with faculty, anecdotal & systematic reports from CIOs, scholarly publications on research trends and methods, and trade press reports.

Factors affecting research use of IT – Disciplinary cultures; professional development on computational techniques (more widespread at elite institutions); institutional and general incentive systems – there are many ways in which research trends (like collaboration or development of new algorithms) are being undermined by incentive systems; Efficiency – how do we evaluate the efficiency of research? Trade-offs will be made in how people use technology based on efficiency – if I have to add metadata to get data into an archive, it won’t likely happen; Relationship between research & teaching – the use of research computing among grad and undergrad students is growing; Diversity of research approaches – people use high performance computing in conjunction with other research methods.

Institutional issues: competition for resources; history of privileging certain faculty and units; emphasis on homogeneity – can generate problems for researchers; assumptions about activity in units – IT may not be engaged with units new to research computing, but the assumption that support is happening in the unit may not be correct; inertial regarding institutional motivations.

Research culture issues: the rapid spread of computationally-intense research across disciplnes, e.g. music and dance or databases of videos in humanities; speed and continuous nature of innovations in research methods (needs for training); generational issues.

Decision making about IT for research: Hiring commitments by deans etc without discussion with IT units; Resource allocation – faculty may want to participate; collaborative infrastructure development (Princeton’s new supercomputer purchase from funds shared across units); policy implementation – if faculty are in on development of policy they are more likely to abide by it.

Collaborative Decision making – multiple options not mutually exclusive. They are quite rare at present. MIT has multiple working groups that orient differently around research problems. Deans and chairs don’t often really have any idea what’s going on in the research areas in their units.

Computing needs – capacity; stability (and help with transitions); architectural flexibility – people who know about OS and research app architectures are usuall not talking to each other.

Training – there’s a gap between how you were trained in grad school and current practices. Inadequate reliance on current students – lots of issues with use of students in terms of knowledge transfer, security, etc. speed of innovation. Range of diffusion techniques – working groups for getting peers in communication. Institutional and inter-institutional synergies. Linkage with methods courses.

More about software – needs range from simple scripting to custom software creation; software specialization as national institutional niche; finding researcher-authored software (NSF is talking about funding software archives)

Data needs: Collection; storage; preparation; presentation

Storage & preparation – multiplicity of types of storage (project-specific repositories to long-lived data collections of global importance) – UCSD is offering up 100 years of data storage to researchers from any institution. Multiplicity of storage venues. Rising & complex preparation needs. Policy issues (access, control, raw vs. cooked, – feds are pressing for release of raw data to federally funded research; etc).

IT & Ontologies – NSF is funding lots of work in metadata and ontologies. Driven by effort to bring info architects and disciplinary scientists into system design. One approach to user-centered design.

The new big two issues – more support for learning, adatpting, and writing software specific to their research problems; Researchers need help managing their data as it enters worlds of public presentation & long-lived data archives. UMich has a lib school training work for disciplinary specialists to do ontological and metadata work. Peter Murray from UMBC quarrels with these two issues – the hot topic is sophisticated pre and post-award information systems. Faculty and PIs want to reduce the administrative burdens. Everybody’s dealing with compliance.

I didn’t see any mention in this presentation of what is probably the biggest issue for us, which is demand for housing research computing in a centrally run data center environment. That need is being driven by security and recoverability issues.

Greg Jackson makes the great point that this analysis doesn’t recognize that many of the shortcomings are the results of tradeoffs made as a result of incomplete analyses of cost, risk, and opportunities. We need to do the analysis of which services are actually worth spending scarce money on.

Technorati Tags: ecar2005, research

While I had to stop blogging due to dead Powerbook batteries, the conference wrapped up with a cocktail reception hosted by Stanford’s Media X group.

I got to use the time to chat with O’Reilly’s Rael Dornfest about having some calendaring events at the Emerging Technology conference in March, to talk with Rael and IBM’s Stephen Farrell about managing activities as opposed to projects or calendars, to introduce myself to Joyce Park and Adam Rikin from Renkoo, to carouse a bit with Lisa Dusseault and Tantek Celik (how do you make that C with the thing hanging down?), and to finally meet Dick Hardt from Sxip.

There’s some video from the event up on CNET, as well as link to more coverage.

I thought that all in all it was a terrific day – kudos to Esther for organizing and leading a really thought provoking event that brought together a lot of disparate threads – I’m not sure it all melded together, but I think that there are definitely some themes – the merger of organizational and personal calendaring is one (which obviously has implications for identity management too); the closeness of discussions of time and place; and the whole world of event management and notification. This feels more and more like a set of topics that is rapidly reaching critical mass – at last!

Technorati Tags: Calendaring, when2

While I had to stop blogging due to dead Powerbook batteries, the conference wrapped up with a cocktail reception hosted by Stanford’s Media X group.

I got to use the time to chat with O’Reilly’s Rael Dornfest about having some calendaring events at the Emerging Technology conference in March, to talk with Rael and IBM’s Stephen Farrell about managing activities as opposed to projects or calendars, to introduce myself to Joyce Park and Adam Rikin from Renkoo, to carouse a bit with Lisa Dusseault and Tantek Celik (how do you make that C with the thing hanging down?), and to finally meet Dick Hardt from Sxip.

There’s some video from the event up on CNET, as well as link to more coverage.

I thought that all in all it was a terrific day – kudos to Esther for organizing and leading a really thought provoking event that brought together a lot of disparate threads – I’m not sure it all melded together, but I think that there are definitely some themes – the merger of organizational and personal calendaring is one (which obviously has implications for identity management too); the closeness of discussions of time and place; and the whole world of event management and notification. This feels more and more like a set of topics that is rapidly reaching critical mass – at last!

Technorati Tags: Calendaring, when2

Will Wright (Maxis/Electronic Arts, and the creator of Sim City)

Models are a good way of abstracting data to make it understandable – building physical models is a good way of starting.

The idea that there could be a whole world living in your computer is fascinating. He showed the first version of Flight Simulator and the latest as a comparison.

One of the interesting things about games is that you can restart games from the begining – an iterative way of experiencing time. Games are a good way of understanding chaotic systems because you can see how small changes can create large differences in outcomes.

Possibility space – the shape of the landscape of possibilities is how you model the difficulty of achievement. Tried to model the Sims so that people needed to balance material and social success – but people tend to go for one side or the other. Now they’re doing dynamic tuning of the game based on data from real players and how they negotiate the space.

A story is a way of how to displace someone’s experience in time and space to apply it to another person. While you’re seeing one linear path of events, the drama is created by imagining all the other things that didn’t happen (“what would’ve happened if he had tripped here?”).

Kids will mash all the buttons, look at what’s happening on the screen, and build a model of the cause and effect relationships – kids are great at building a mental model of arbitrarily complex systems. Hey – just like science!

Games are doing the same thing – looking for simple compact rules that can create large spaces of possibilities.

Dynamics – the change of structures through time. We tend to think in terms of topologies, which are the parts of the systems, then there are the dynamics that occur to these things, then there are paradigms (network theory, adaptive systems, chaor theory, etc). Game player intuitively understand topologies within two or three minutes of playing.

A unique property of time is nested interaction loop with success and failure at each level. First you have to learn basic control, then you can deal with their needs, then you can get to the next level. The most intereseting side for most players is the failure side – that’s where people spend the most time. As long as people understand why they failed they are willing to go back and try it again.

Interesting book – the User Illusion. Most of our intelligence is pre-conscious.

When players look at games a similary thing happens – use visuals in the game to tell the player what the nouns and verbs are. In first-person shooters you discover things by going around and bashing everything with the noun you have.

Games run on two processors – the computer and the player’s imagination. A lot of what gets built in games is scaffolding for the player’s imagination. A lot of the trend lately in games is to go to more open-ended games where the player has a role of authorship.

Technorati Tags: Calendaring, when2

Using time and time patterns to detect things.

Bob Pinner (CDC) – An internal medicine and infectious disease doc, now working on epidemiology – three major dimensions are time, place, and person. The structure of things they think about are similar to events. Infectious disease surveillance is what he works on – monitoring trends, or more recently, early detection. Can you go earlier than a specific diagnosis, to a syndrome, or even earlier, to a set of conditions. The earlier you get the less specific the signals are. Public health functions are organized locally by state and county – that’s good for local response but not so good for national distribution of medicine, for instance.

Steve Hofmayr (Sana Security) – Trying to detect malicious software on a single computer. Gathering information can leave you vulnerable. If you’re looking at time on your computer, looking at dynamic behavior of a system. When it behaves a little strangely, it’s not enough to define it as bad, but when it’s a lot strange it’s too late. With machines you can roll back what happened once you know. Analagous to the immune system, which doesn’t mount a massive response right when it sees something new, but waits to gather more information before responding. A classic example – You’d think if something on its machine that tries to hide itself is bad, but it’s not necessarily. Or if a process survives a reboot. But if enough of those kinds of events are correlated in time, then it becomes more likely that there is some malware.

Dan Doman – At doubleclick they had a vision of highly targeted advertising. Keeping track of demographic data is difficult – he got interested in inferring demographics – e.g. people who go to sports sites are likely to be guys, etc. Contextual advertising is delivering advertising within the context that the consumer is in now. You look for the numbers of times people are looking at things (“velocity”) over a period of time which indicate an intensity of interest.

Omar Tawakol (Revenue Science) – Behavioral targeting – advertisers and marketers have always wanted to reach people based on what they care about. Behavioral targeting talks about the person reading the page, not the text on a page. It also brings the notion of time into the equation – if you go to a car site, and then to an entertainment site, the entertainment site can show you car ads. In advertising there are two uses for time – one is branding, which is all about your interests; the other is direct response, where the goal is immediate response. Branding is more time independent – it you’re a golfer you probably will be in five years, but if you’re looking to buy a mortgage, you probably won’t buy again for five years.

Technorati Tags: Calendaring, when2

Using time and time patterns to detect things.

Bob Pinner (CDC) – An internal medicine and infectious disease doc, now working on epidemiology – three major dimensions are time, place, and person. The structure of things they think about are similar to events. Infectious disease surveillance is what he works on – monitoring trends, or more recently, early detection. Can you go earlier than a specific diagnosis, to a syndrome, or even earlier, to a set of conditions. The earlier you get the less specific the signals are. Public health functions are organized locally by state and county – that’s good for local response but not so good for national distribution of medicine, for instance.

Steve Hofmayr (Sana Security) – Trying to detect malicious software on a single computer. Gathering information can leave you vulnerable. If you’re looking at time on your computer, looking at dynamic behavior of a system. When it behaves a little strangely, it’s not enough to define it as bad, but when it’s a lot strange it’s too late. With machines you can roll back what happened once you know. Analagous to the immune system, which doesn’t mount a massive response right when it sees something new, but waits to gather more information before responding. A classic example – You’d think if something on its machine that tries to hide itself is bad, but it’s not necessarily. Or if a process survives a reboot. But if enough of those kinds of events are correlated in time, then it becomes more likely that there is some malware.

Dan Doman – At doubleclick they had a vision of highly targeted advertising. Keeping track of demographic data is difficult – he got interested in inferring demographics – e.g. people who go to sports sites are likely to be guys, etc. Contextual advertising is delivering advertising within the context that the consumer is in now. You look for the numbers of times people are looking at things (“velocity”) over a period of time which indicate an intensity of interest.

Omar Tawakol (Revenue Science) – Behavioral targeting – advertisers and marketers have always wanted to reach people based on what they care about. Behavioral targeting talks about the person reading the page, not the text on a page. It also brings the notion of time into the equation – if you go to a car site, and then to an entertainment site, the entertainment site can show you car ads. In advertising there are two uses for time – one is branding, which is all about your interests; the other is direct response, where the goal is immediate response. Branding is more time independent – it you’re a golfer you probably will be in five years, but if you’re looking to buy a mortgage, you probably won’t buy again for five years.

Technorati Tags: Calendaring, when2

Had a good time at lunch chatting with Lisa Dusseault from OSAF and Tantek Celik from Technorati about the hcalendar microformat, calendar urls, and lost of other interesting things.

The afternoon panel, on time and functionality is just starting.

Munjal Shah (RIYA) – Tagging lots of digital photo data – uses face recognition and text recognition to infer what and who are in a photo, and turn it into a searchable database. They do it for consumers on the desktop, and have a permissioning model for letting people search their friends. One insight is that consumer photos come with timestamps, which they used to enhance face recognition through time-based clustering. For instance, if there are ten photos of you at a party wearing a particular shirt and only one of them is full-face, they can infer that the other photos of the same shirt around the same time are also you.

Tantek Celik (Technorati) – Time searching on the web is terrible – try looking for just this year’s version of the conference. Technorati relies on pings from blog software for indexing. Before an event people are talking about it, during the event people are blogging it, after the event people write about it. What happens in short time windows? We find that humans are the best at knowing what’s going on right now – the most popular ten searches on Technorati. News – let’s look at what bloggers are linking to in the last forty-eight hours – turns out to not being the same topics a newspaper editor would choose. They see the names of bloggers in far-flung places around the globe that we may not have heard of – it dramatically flattens our view of the world.

Esther asks when most documents will have timestamps on the web – Tantek asks whether you can trust time information in documents? The ping is a more reliable time stamp, because Technorati knows when the ping was received and when they went out and retrieved the information. As we get used to copying info from the web we’ll make much more use of time-based information. Technorati is working on microformats, small extensions to html that enable information within web pages to identify the same kind of information as ical and vcards.

John Arenas (Worktopia) – Worktopia is about the premise that the ubiquitous network and collaboration tools are freeing the workforce from physical limitations. Matches demands such as temporary space with supply. Enables companies to have a distributed workforce. This kind of relationship allows hotels and other meeting spaces to tap short-term markets between big meetings, for example. John notes that Sun now has 1.5 workers per desk, so this is an accelerating trend.

Ben Cruze (Demand ID Systems) – enabling users to request live music events, and over time other kinds of events. Provide a market intelligence to show level of demand for a performer in any part of the country – that doesn’t exist today. On the back end, when an event is scheduled, they alert consumers who requested the event, so they can purchase tickets and merchandise. They can also alert sponsors to how many people might be likely to attend an event, so they can better plan and target their sponsorship dollars. The consumer service is under the brand name of Tourvote. Enabling people to have a voice in creating an event is important.

What’s the business model? Munjal – Search is the model – you’ll search for places, products, and things – not professional pictures but user generated pictures, which reflect reality better. The premise is that travel advertising will support the business. Tantek – Marketers are using Technorati to do research on their brands.

Mark Johnson from Intuit on how does time influence decision making? The (somewhat silly) example he gives is the decision of whether to spend $4 a day on a venti cappucinno vs. $4 a day on Starbuck’s stock. Esther comments on making latent demand visible – knowing that you’re part of a larger group that all want Bonnie Raitt to show up. How is our discount factor changing with respect to time?

Ben asks Munjal the question about whether people have the right to put up pictures of buildings and other businesses that they don’t own, and about the tension between business owners who want to link to photos of their businesses and the reality of what user photos might actually show (e.g. the cockroach in a hotel).

Technorati Tags: Calendaring, when2