Computational Science and Engineering "Advance" Meeting

January 22, 2007
1147 Math Sciences
University of California, Davis Campus

Summary Notes

(Compiled by Robert Pattison, Dawn Sumner, Jim Crutchfield, and …)

Welcome and charge (Deans Ko and Lavernia)

Dean Lavernia: CSE is important. Both Deans recognize the importance of the computational science area. Emphasized engaging all of campus.

Dean Ko: There is good turnout and great interest in CSE. There is a lot of expertise in CSE and it is evident from the SciDAC awards and outside partnerships like that of Google with LSST and the Hazards Institute. Initially, the CSE objective was very simple, create an environment for world class education and research at UC Davis. CSE is a very good program and we are here to listen to everyone’s vision and hope for a consensus in campus vision and structure to enable world class education and research in the future.

Introduction: CSE initiative Present and Future (McCurdy)

View this presentation

• Reviews by MPS and Engineering, two reports produced, each sent to each dean.
• Goals - Need recommendations for:
  • Education - almost universal agreement that we have a problem and can do something about it; focus on commonalities
  • Science and Synergies
    • New grants - what role should the center have
      • Requires individuals to step forward to lead
    • Need to know what each other are doing across campus
• Organization and Leadership for CCSE
  • Including recommendations of who should lead it
• Service role of the CCSE
  • Tar baby - "who’s going to run my computers"
• Report
• Circulate to participants
• One report will be delivered to both deans

CSE Education and Discussion (Joy)

View this Presentation

• What is (are) the educational goal(s)?
• Programs around the country
  • Those trying to define a “new field” of computational science: Rarely successful; tend not to have the core courses.
  • Single Department
    • Computational Mathematics; Computer and Computational Science
    • Mixed success
  • Designated Emphasis Programs
    • Set of core computational programs for domain science and engineering applications
  • Programs that integrate Science & Engineering Applications
    • Usually associated w/ a high profile center
    • Tend to be successful
• Working Programs
  • U of Texas; graduates now working in other top programs
    • Consists of lots of “Centers”; include many topics in their program
    • Courses: offered through faculty in this institute
    • Hundreds of associated faculty; started as a one person show
    • Lots of facilities: well funded by university and through grants
      • funding came after a long period of grass roots efforts
  • Texas A&M - smaller program; able to get graduates in pipeline
    • 4 faculty; 5 courses integrated with the science applications
  • Columbia, Stony Brook, Brookhaven
    • Model of collaboration with a national lab
    • Brand new
    • Built around core faculty including courses provided by lab personnel
  • Princeton
    • Institute for Comp Sci & Engineering
    • Brand new, 2002
    • Similar statement as CCSE in terms of importance of work
    • Also in conjunction with a national lab?
  • Harvard
    • Trying to get all people interested in large-scale computation co-located
    • Earth science projects not listed: e.g., none of the climate science people; rather it’s focused on medical/bioscience with some other areas. Not on campus, which will intrinsically limit the participation of some faculty.
  • UIUC
    • Large number of courses for grad students
    • Mature grad program
  • UCSB
    • IGERT in computational science
    • Small program and struggling
    • In only a few departments: Math, Computer Science mostly
    • Joy said NSF is looking for something different for an additional IGERT
  • UCSD Supercomputing Center
    • Not old enough to have many people coming out
    • Looking at putting more funding into the center to expand resources
  • Purdue
    • Older, struggling, but integrated with the Computer Research Institute and now taking off
    • Started fall ‘95
• National Reports: Agree on CSE needing to be integrated into grad curriculum
  • Methods will complement methods of theory and experiment
  • Computational methods will replace and enhance many current research approaches
• At UC Davis
  • Want to develop designated emphasis
  • Get graduates out into the workforce/academia
  • Have
    • Graduate groups for interdisciplinary groups
    • Many existing courses in domain sci, math & stats, CS
    • CCSE
• Previous effort here was a designated emphasis between applied science and chemistry; but died because of lack of enrollment
• To date not coordinated on campus
• Biggest barrier
• Coordination of classes
• Need guarantees that departments would teach the necessary classes
• SciDAC Model
  • Triangle of Domain, Applied Math, Algorithms & Computation
  • Designated emphasis programs concentrated on Domain point of triangle
  • Ph.D. programs cover the rest, mostly between Applied Math and Algorithms & Computation
  • 1/3, 2/3 Ph.D. model
    • Domain science students essentially take a minor in computational methods, ~1/3 coursework or less
    • Computation students take 1/3 in a domain science, rest in computational

Discussion

• Orel: There was a designated emphasis about 5 years ago, when CSE was started. A set of computation courses was required but Chemistry/Applied science did not want to use it and it was ended because of low enrollment. Could be re-started?
• Anon: Tremendous collaboration is required for it to work. And would need to include other schools.
• Anon: Should include medical computation
• Anon: Need to include Ag school as well to make it work.
• Lavernia: We agree that there should be more schools involved and have discussed informally with Biological Sciences Dean Ken Burtis.
• Anon: There is no funding on campus for designated emphases. Institutional commitment is required. This is what a graduate group provides.
• Joy: And if we teach HPC we would need HPC infrastructure. A graduate group could help with this also.
• Anon: The Structure of CSE is a problem. Perhaps it should be setup similar to the Genome Center where they report to a committee with one lead dean. Perhaps CSE should become an ORU. Genome center has been organized to report to a committee of deans – good to integrate
  • Reorganizing sort of like an ORU, coordinated by the office of research
  • Ko says future of CCSE might be an ORU
• Question to Joy: Do any educational programs exist w/o a strong research program?
  • Probably not, Joy didn’t look for those; he looked for those with the strongest research programs
• Joy: Start by looking at all offered courses and define a list of core courses for the grad group
• McCurdy: Most grad students arrive here ignorant of numerical techniques and parallel computing. They think that there aren’t classes to take; but there actually are in the context of each department. We need more than cross listing, also need to modify the content of course to fit broader needs.
• Pickett: hurdle to getting interdisciplinary courses taught regularly because taught by departments; first priority is departmental needs
• Hoppman: Ag has a rich history in cross department and college grad group; model in place;
• Freeman : Major research efforts are the foundation for education. How are the other CSE programs in regards to this? Does University of Texas have strong education with strong research?
• McCurdy : Cross listing courses is important, but it is not enough.
• Anon : There are hurdles to getting interdisciplinary courses taught. There are already high requirements in departments for teaching that must be met first.
• McCurdy : Theoreticians are generally a minority, so these classes are not a high priority to departments.
• Lavernia: If you bring the department chairs on board the courses will get taught.
• Ko: Davis campus culture is that courses come from faculty, coordination is important because if departments offer similar courses then enrollment will be split. Need a critical mass of students; deans need to put resources where the students are; depts that are willing to give those courses can be rewarded.

• Anon: For courses to be truly interdisciplinary it requires they be team-taught. Bio Sci has a good example of this. In their graduate groups many of the classes utilize team teaching. Added value for team teaching is faculty learning from each other when coordinating lectures and class materials. This would be very good for developing interdisciplinary research.
• Ko: If you only teach 1/3 of a class you would have to teach 3 to get credit for teaching a full course.
• Kellogg: Courses in a sequence could also be interdisciplinary.
• Anon: Do not see why if a graduate group was in place that courses would not get taught? Departments would still get credit for teaching the students.
• Anon: Simply putting the list of courses on the web would help
• Hoppman: Student numbers is where the rubber hits the road; service courses need coordination and motivation is increased student numbers
• McCurdy: Don’t always have a person in the same department teaching it; need a commitment?
• Temple Lang: Talking about core courses; what about inter/multi-disciplinary courses; need co-teaching; is there a way to do that here?
  • Joy: This is done on campus; How do we actually do it?
  • Ko: Yes, you are allowed to do this.
  • Anon: Depends on the flavor of the course; don’t need to co-teach
  • D’Souza : wants to co-teach to provide broader expertise
  • Kellogg: Could have a sequence of classes rather than team teaching
  • Sumner: Faculty members get a lot intellectually from team-teaching courses
• Anon: Designated emphases do not have any financial support
  • Joy: That’s why we need something like a grad group
• Kleeman: Need a major research program to support the educational program; they need to be a thriving research group; success of students is more a result of the people, not the structure/classes students take; success is more correlated to the people they work with
• Rustad: Students also need to feel they are part of something big and new
• Kleeman: Have individual stars; need a center for the research
• Seigel: Funding model at Illinois – they were looking at a proposal to link educational funding to growth of research funding by involved faculty; don’t know if that’s a good model or not;
  • Joy: Almost exactly the same as Harvard
• Hamann: Can’t separate research from education; have to ask what is in it for ‘me’, my merit, what do I want to get out of it, how will it help me do bigger and better things? Articulate what you want this unit to do?
• Hamann: Google just invested $6M in data center in North Carolina; adjust to changes in market place; where will the jobs be tomorrow? Need that to make campus attractive and distinguish it from others? How can we be bigger and better than other fields?
• Joy: Grad program gets the students together; IDAV has 40 grad students, benefit for students dealing with other students on campus.
• Siegel : It is important to look at funding models. There is a link between growth and co-investment by campus. Look at the model at Illinois.
• Hamann : To interest and motivate faculty, there are basic questions that need to be addressed by each: What’s in it for me? Merits? Fame? What do I get out of this? Bigger and better things, better graduate students. What can this unit enable me to do?
• Anon : Google is investing $600 million in a data center in Carolina. We as a campus need to be willing to invest to keep graduate students.
• Joy : Graduate students are my lifeblood.

Research Group Presentations

Computational Biology (Toby Allen)

View this Presentation

• Characterized by very diverse questions and techniques.
• Electronic level; molecular (Allen’s level), cellular, networks, bioinformatics, system - organisms, systems - populations, evolution, ecosystems
• Why? – need to unveil hidden links; experimental probes imperfect resolution in space and time; computers can provide quantitative explanations
• Large range of space and time scales
• Many examples
  • Mogilner – cell motility and division
• Center to nucleate a broader group of researchers
  • Funding
  • Common educational needs and goals

Computational Materials (Mark Asta)

View this Presentation

• Moderately uniform techniques compared to biology, diverse applications for materials.
• Faculty already interacting from many departments
  • Mostly molecular
  • Missing continuum materials - center could facilitate
• Education - opportunities for collaboration, some duplicated courses taught
• Need high performance computing - bread-and-butter of computational materials
• Methods similar across group, lots of applications
• Big emphasis in funding on collaborative experiment and simulation/modeling
• Need
• Support for large grant development
• RA support to seed interdisciplinary projects
• Increased external visibility
• Expand web site, budget for workshops, visitors
• Education
• Coordination, designated emphasis, interest in graduate group focused on Applied Math and Computational Science
• HP Computing Infrastructure

Complex Systems (Jim Crutchfield)

View this Presentation

• Very diverse questions and techniques
• Benefits of a cross-cutting theme
  • Asking why is the world organized the way it is
    • Large-scale systems; emergent properties; how are we going to address these questions?
  • Innovations: Need to rethink foundations of how to address problems, how to questions our assumptions about them; use computing to approach large-scale systems with emergent patterns.
  • Talk to local industry (Silicon Valley) people about what they want, e.g. Google:
    • Google system emulates biology, redundancy rather than perfect reliability
    • Data mining & semantic web
    • Emergent web social organization
• Drivers for success
  • Build an engine for innovation
    • Intellectually adaptive, open environment
    • Nurture research to point of independent funding
    • Graduate fellowships
    • Faculty UCD-internal sabbaticals, ...
  • Education
    • Grad group – tools, applications
    • Theoretical side needs to also be developed

Computational Transport (Jean-Pierre Delplanque)

View this Presentation

• Similar techniques applied to fluid behaviors of diverse materials
• Group that solves the same set of equations
• Need
  • Research center - organization
    • Facilitate communication, some in the group didn’t know each other
    • External visibility
    • Grant development
  • Education - designated emphasis, graduate group?
    • Educational & research structure parallel, but distinct
      • trying to do everything in one structure my not be good
  • General
    • HPC infrastructure
    • Involve more colleges and schools
    • Interest in a superstructure w/ focused modules in terms of the shape of this center

Networks (Raissa D’Souza)

View this Presentation

• Developing novel techniques to understand a new field, applied to diverse problems
• Cross cutting theme again
• Center has been very successful at facilitating cross-campus collaboration
• Goal: come up with a general, unifying theory for understanding networks
• NAS/NRC Network Science report
• Classes starting on Network theory classes, in various departments where taught

Data Management/Analysis and Visualization (Ken Joy)

View this Presentation

• Diverse problems, but presentation focused on IDAV rather than integrating efforts across campus
• Emphasis on national labs
• Need
  • A place to focus activities, a place to collaborate; get people together in the same place
  • Graduate group, supports qualified students, centers educational program

Computational Environmental/Geological Science (Louise Kellogg)

View this Presentation

[Notes not available.]

Computational Applied Math (Naoki Saito)

View this Presentation

[Notes not available.]

Discussion

• Lavernia : There are many examples of Graduate Groups on campus. I’m even more impressed now, computational science is clearly an extremely active area.
• Anon : Designated Emphasis is quick, we should do that first.
• Klein : Focus on problems as a group. Find a program to emulate. The reward structure is yours perhaps you would get accelerated merit, but the true reward is solving problems, being successful, publications, and the legacy you leave. Also think about how to raise money, for example, using the problem of global health you could maybe solicit the Gates foundation.
• McCurdy : We can agree on a graduate group if we don’t get down into the details. How do we define a graduate group? Breadth?
• Ko : Graduate groups has to be research oriented. It needs a focus on research to find a graduate group focus.
• Anon : Why get a Ph.D. in CSE vs. Physics graduate group?
• Bai : Why not setup CSE as an arm of Applied math, it is a successful program already. Then a graduate group might come of this.

Research Computing (Pete Siegel)

View this Presentation

• There is a clear need for physical space and machine space.
• In the near term we are reviewing options for space in Sacramento, Chiles road facility basement, changes in existing space. For the future long term, perhaps allowing a contractor to construct a building on campus and allow the UC to have space in it in exchange for reduced electricity rates.
• Klein: The campus loses money every year on indirect costs. This can’t be used for computing facilities and the way the debt works UC can’t mortgage buildings to keep up with demand. Having other developers is one way and another is off-campus space. When using off-campus space the indirect is only 26% and the difference between that and the on campus rate of 52% can be used to pay things like mortgages.
• Anon : Why not outsource the whole thing?
• Siegel : Could be problem with knowledge about infrastructure, if vendors slip.

Final Discussion

View this Presentation

• McCurdy: Strawman model
  • Director, steering committee, advisory committee (outside)
  • Current groupings based on presentations, leaders would be the steering committee
  • No new money on the table
    • Need to be able to make the case for 5 new faculty
    • Need to make the case for $
  • Who does the director report to?
    • Currently to the two deans
    • Hoppman suggested including more people
    • ORUs report to Klein, but new ones aren’t going forward
    • Kellogg : University has defined structures, so it depends on the direction of CCSE
  • What is the best structure to make the argument that this is a useful?
    • Kellogg: It is the energy from this group that will make any of this move forward; so the structure should be chosen as something that we can get behind
      • focus on the next $10M that won’t be here without CCSE

• Hamann: Structure and leadership should be set up to make everyone feel equal in value to the efforts, not as a data provider to someone else, algorithm provider, etc.
• Anon: Draw organizational chart differently - applications and tools on two axes
• Galli: One wants predictive computation for scientific discovery
  • Overarching problems to which we want to contribute
    • energy, climate, etc.
    • we propose solving some of these problems with predictive computation
  • Think about where should we be 5-10 year from now
• Turcott: Interdisciplinary is a buzz word; look forward, be proactive
  • Hamann : Teams breaking down boundaries, doing things that we can’t do without each other, addressing bigger problems, what happens as we merge things into the next level;
• Temple Lang: Software is more expensive than hardware; can we develop things together; a possible way of building something entirely new and more productive; shared development draws a thread through various topics
• Disagreement from several on a small number of critical problems: If the chosen problems aren’t directly relevant, many won’t want to work on them;
• Kleeman: Focus on problems, but not have them chosen by this group; let successful topics nucleate based on success in funding, individual efforts; have to have the commitment from individuals to get those big proposals submitted and funded
• Galli: Software-sharing highlights the question about “how we do computation?”
• Kellogg : Software method development - KeckCAVES; methods - built out of existing software to new problems; research directions have shifted; it’s a lot of work to make these changes; people have done it based on their excitement – different things have come out than expected due to the people involved. They found the problems, collaboration and opportunities worth the time.
• Anon: Bottom up – and after things have grown, the administration supports them
• Sumner: We have a large grass roots effort from a number of groups; what structure will harness that energy and short-cut the time to a program with a large impact?
• Hamann: JMIE = several independent centers under one umbrella; a meta structure that sits on top of existing centers on campus to share overhead, etc.
• McCurdy: Need to represent the diversity of a large group of people; maybe the overarching meta structure can do this

Leadership?

• McCurdy : Advisory committee of people from other entities review CSE every couple of years. Steering committee consists of heads of focus groups.
• Kellogg : There are specific reporting structures. Graduate groups report to Graduate studies, ORUs report to Barry. The important thing to ask is for the next 10 million how will we organize?
• Anon : We should assume that there are no more FTE. It’s a sensitive issue and might distract the deans.
• Anon : If we go for an ORU then grants would go through the ORU and Barry would be able to kick in money for this, however, departments won’t like that grants are being taken away from them.
• Galli : Which model do we use? How do we plan for the future? I would like to see this be more predictive. Where will we be at in 5 years? What will the computational world be like in the future?
• McCurdy : This brings us back to what Barry said about “problems”.
• Temple Lang:Nothing new is coming out of this. How do we do new science? What about software, Can we share development software?
• Kleeman : People are disenfranchised because problems are too specific. It would be good to orient problems based on the market. With maybe 5-7 large-scale problems.
• Anon: The center’s mission should still be facilitating faculty getting together to solve computational problems, educations, representing community. ORUs are designed to facilitate interdisciplinary collaboration.
• Anon : Organization is required before we can go out and get money.
• Anon : The proper approach is for people to get together and do research then once you have successful research you get center status.
• Anon: The center already has a grassroots effort: Current CSE efforts, Geology, material sciences. Bottom up is already here.
• Bernd : Perhaps a structure like JMIE is appropriate? CSE would be just one of many groups under a larger umbrella.

Graduate Education:

• Goals for the Center: Getting faculty together, a conduit to the administration and the outside; interdisciplinary nature suggests an ORU;
• Ko: Faculty decides what is taught; maybe if we get together and do these courses and get agreement from department chairs;
• Turcott: No computation in stats, thermo, etc.; need MatLab or something that could be adopted universally to provide a fundamental basis
• But this is probably fundamentally for undergraduate education
• Bai: Graduate Group in Applied Math is huge; use this as a model for the breadth and depth – probably half is already computational science; Computer Science graduate group – interaction w/ genome, etc.; start with these two.
• Benham: Grad Groups that are successful usually have a single department (maybe one more) that really cares about it; or need something like 10 faculty who really want to make a new one happen
• Anon: Designated Emphasis – has to involve 2 or more graduate groups
  • Enough consensus that department faculty could come together to support this

• Anon: Faculty decides what is taught on campus. If a couple of faculty in each department get chair to buy in and allow new courses then it will work.
• Bai: Graduate groups in applied math and computer science exist. It might take years to start a new group but a strong director might be able to coordinate with existing groups.
• Anon : The BioTech graduate group is successful because it has funding and administrative support.
• Anon : If 10 faculty were willing to work extremely hard on this project, it might happen.

Computational Infrastructure:

• Anon: Serious commitment/plan to build a building south of campus
• Rustad: Concerned that there isn’t someone focused on HPC in the center
• Joy: Siegel really wants research computing; didn’t have at Illinois;
• Kellogg: Do the people who provide Siegel’s funding have this as a priority? They need to understand that it saves money and brings in money.
• Anon: We have more active clusters than Berkeley
• Anon: Can’t under estimate the costs - electricity, etc. - we already pay these costs, in individual departments, but changes will shift to admin for these costs

• Anon: There are convincing plans to build more space for computing needs by possibly having outside developers come in.
• Anon: Research computing support should be separate from administrative and CSE could the advocate for this.
• Anon: Can’t be a 21^st research institution without HPC.
• Anon: Grant support is necessary.

What is Needed?

• Anon: Hiring more FTE is not the biggest or best and might influence the decision in a bad way.
• On whiteboard: First Year: CCSE begins as collection of groups which have self-nucleated.
• Anon: Then can with colleagues improve their research.
• Anon: Some may choose not to join.
• Anon: This does not assume you will stay in the same research. Conversations and intellectual relationships have changed research.
• Anon: Shared space is necessary for collaboration.
• Anon: An administrative staff dedicated to helping with interdisciplinary research.
• Anon: Need intellectual leadership to articulate vision and also need grassroots.
• Anon: What goals do we set? Goal of new grants, goal of ORU, goal of campus computing infrastructure? goal of Graduate group? IGERT?
• Anon: Interim director to carry momentum of advance and refine vision.
• Anon: A reporting structure with a single dean.
• Anon: A director is like a department chair. It’s part of service. Resources are necessary for a director to achieve anything though. How to make the position sufficiently attractive?
• Anon: Craft the job description for the director to make it a powerful position - (Hamann’s point); negotiate before accepting the job.
• Sumner: For research there are foci of people in the room:
  • Those dedicated to making High Performance Computing more efficient for Computational Science and Engineering
  • Those that use HPC for discipline science who need the infrastructure and support to do their research better
  • Those that are interested in using Computational Science to develop new theories and approaches to computation and science.
  • These three directions need to be simultaneously supported and developed in a way that encourages the strength of each to contribute to the others.
  • Space and communication are essential. Communication is fundamentally up to the participants, but the CCSE should facilitate opportunities. The characteristics and needs of the community require an ORU-type structure.
• Sumner: For education a new graduate group is essential for a significant segment of the participants. Others are likely to benefit from it. The 1/3, 2/3 model seems a very appropriate starting place.