Session Notes

Session 1 Overview - All Groups

CI GROUP OVERVIEW

Campuses are responsible for persistent infrastructure. Track 3 level does this at campus level. NSF funds at meta-campus level.

GAPS AND WEAKNESSES

Windows dominates desktops but not academic CI

• RFI must be addressed

• CI must be lightweight, robust, reliable

• Decision supported provided by integrated CI

• Validation and verification of models

Opportunities

• Great leverage of Windows tools in NSF projects

• Tired service versus net neutrality

• CI should provide leveled playing field for rural US, smaller university; big future challenge for us

• NSF opportunity; conflict between municipal vs. commercial studies

• What would the chemical industry do with wireless access on every lamppost

• Communication is an issue in international companies; import for data and method

• Global continuing education with change with pervasiveness of CI; must support innovation approaches and distance learning.

OPTIMIZATION and SUPPLY CHAIN GROUP

Grand Challenge – the supply chain

1. Getting the desired products to effectively and efficiently to consumer can have a HUGE societal impact (jobs, economic curvature)
   
   
2. Can we have a well-coordinate supply chain
• With CI – and old idea whose time may come (gives computing and data)
  
  
3. Chemical Supply Chains
   • very large scale
   • highly uncertain
   • difficult scheduling aspects (variable process times; changeover times)
   • nonlinear

Challenge #1  Optimize the entire supply chain

            Needs to be Multi-scale (strategic, operational, tactical, real-time)
            There is Uncertainty – multiple objectives
             
CI Can help handle these issues e.g.   couple simulation with optimization

Challenge #2 - Modeling the national chains

            Collaboration/competition issues in supply chain management
            Impact for National Security

Knowledge Gaps

• People don’t use optimization – Why?  How can CI Help?

• I don’t have the data – CI can get it for me; systems will be dynamic

• Answers are non-intuitive – visualization, simulation for insight

• Answers aren’t robust – consider systematic treatment of data uncertainty

• Requires an expert – problem solving environment; e.g. collaborative OR + Chem/Engineering Communities

Conclusions

• HW is only part of the solution
• Might want “high throughput” not necessarily high performance
• Algorithm development is AT LEAST as important as hardware

MODELING AND SIMULATION GROUP

Problems

Consider these areas:

• drugs/health
• energy
• environmental
• homeland security
• sustainability

With this overlay:

• Multi-scale modeling
• Large datasets
• Education and training
• Economic prosperity
• Algorithm development
• Physical model development
• IP issues
• Visualization/data display
• Standards
• Community models/codes – impacts all other areas
• Lack of them right now – atmospheric modeling community – notably lacking in chem./engr communities

Areas that could use a model

CFD for biological systems
Cardiovascular
Industrial fluid property simulation

Community Based Models – and what they optimally need

• open-source
• many funding sources (what tech communities; could be many within CBET)
• Collaboration among different communities
• knowledge sharing/data sharing
• dissemination vehicle
• validation/standards development
• Curation/management/responsibility for peer-review (who would do take responsibility to ensure quality) (By whom?  Academics?)
• ITR – community based models?
• multi-scale
• multi-physics
• programming language issues?

Community-based models can engage DOE, etc., better than individual PIs

Discussion:

Question:  Who pays for open-source development?  Where’s the reward system for community based models?

How does atmospheric community get it funded?  Who underwrites? 

Will industry underwrite if it’s relevant?

Would be nice to leverage work that’s been done and build on it?

What’s the central hub?

Initial steps in CFD community, but embryonic.

Point:  Don’t we have this? (Vince)

Developing models within CFD – not easy;  no code-sharing; commercial vendors are not providing code;

Want to move into non-traditional areas.

Take exception to idea that partnerships with vendors haven’t worked out well (Vince).

Does each vendor represent a separate community?  Agreement.

CBM – doesn’t just mean code.

Social/organizational  issues - we need some cultural change; proposals always get “that’s no good; we already did it”

SMART PLANTS - HARRIS

Define smart plant & vision

Ability to adapt to a wide variety of change ( failure of equipment, changes in process; business changes, etc.).  Smart plants would adapt fast and take optimal corrective action fast – including shutting down the process – if that’s the optimal thing to do.

Industry Problems

1.  Economic problems – owners of industrial plants would like 100 percent availability; ability to run production all the time and never having a problem that affects:

• Profitability
• Safety
• Environment

2.  Lack of data to prove benefit of improvements

• Sometimes it’s straight-forward
• Technical measurements
• Sensitive data (e.g. near misses, problems in plant); Method of monitoring and sharing

3.  Uncertainty of future business environment – e.g. cost of oil; demand for heating

4.  Employment issues- number of jobs vs. people who can do them; loss of experienced people

Gaps in knowledge

• Security vs. openness (want excellent communication and interoperability;) more you have the more security vulnerability; continual patches.

• Ability to turn data into information and turn that into valuable knowledge

• Need models to use the knowledge

• Requirement  for dramatically more data; sensors, networks; - cost is high (but coming down); need networks to feed correctly; s/w to interpret

• Ability of plant operators to handle emergencies – some plants are using simulations, but that’s not the norm (airline pilots do it – not the norm in our industry)

OPPORTUNITIES FOR CI

• Have a much smarter, simpler process flow – will solve others

• Improvements in energy usage
• Integration of all communications networks p- sensors, etc. 

Where should NSF throw money?

1.  Data and communications – affordable sensors (installation, actual cost of sensors, maintenance costs; interoperability)

2.  Operator training (plus use a more sophisticated approaches to reacting to abnormal situation); but group is describing abnormal situation as one the system can’t handle

3.  Cyber security (without the side effects – i.e. forgot the password with an urgent requirement waiting)

SMART PLANTS (2) – PROCESS SYSTEMS ENGINEERING

Four Challenges

1.  Unacceptably long lag times from tech innovation to commercialization

Gaps

• Lack of understanding business process to adequately automate translation to commercialization
• Lack of case studies demonstrating characteristics and other aspects of business process
• Appropriate use of dynamic modeling (currently as a point solution)

2.  Lack of a systematic process for innovation

Gaps

• Appropriate data acquisition (VOC, VOM) (including pulling info from wherever it’s available)
• Analysis of the resulting heterogeneous data
• Isolation of “vested” communities

3.  Control and operations based on hierarchy of multi-scale models for flexible manufacturing

Gaps

• Sufficient value proposition (so what?)
• Applications demonstrating value
• Interoperability – I might have the models but don’t know how to actually use them in my work

4.  Sensor network design; full utilization of sensors and sensor data

Gaps

• Nature of novel sensors (esp. image, acoustic
• Integration into PSE practice
• Handling massive quantities of distributed data

Opportunities

Background question – what’s the role of academia vs. industry vs. vendors, vs. standards development?

• Tools for facilitating dynamic modeling (make a transparent part of the practice)
• Tools for data acquisition, heterogeneous data analysis for innovation target identification
• Tools for facilitating communications and integrating communities locally and globally (interoperability of human assets)
• Tools for finding relevant information (e.g. a “specialized Google”)
• Tools for integrating models across multiple scales
• Tools for novel sensor data analysis for ops, maintenance and fault detection
• Tools for knowledge extraction from divers sources and heterogeneous data

BIOLOGICAL SYSTEMS

1a)  Funnel problem – 5,000 drugs in; 1 not so good drug out.  Pharma continually gets new tech thrown at them; most doesn’t; work – turned them into pessimists.

• Modeling for pharma - hard to predict outcomes

• Can we ID who should/shouldn’t have a specific drug?

• How to conduct clinical trials?  Need CI to do the trials.

• Can we compile data (Who smoked?  At what age?  What’s the prior medical history?) to populate trials?

• Outside the U.S. there’s good clinical data now being used to lure good researchers to those countries.  In US – data is hidden due to IP, privacy, bad press, etc.

• Biological variability – chemical or engineering vs. a huge variability in biological systems

1b)  What are the IT enabled economic drivers?

• Answering the questions asked in 1A would have big impact

• Problem:  research community wants open source; inhibits company’s ability to function – what do you sell?  If companies can’t survive it’s bad for research community.

• Personalized drugs are already happening – proof of principle is out there.  If pharma becomes more efficient, it can target smaller pools of patients; can address rarer diseases

• Drug resistant crops – other non-pharma applications. 

2.  Where are the gaps?

Group focused on how to address the gaps (already identified as: communications, multi-scale multi-physics modeling, education, etc.)

Answer – Need resources

• Won’t happen by volunteer
• Requires infrastructure; haven’t seen convergence; need an enforcement body.

Common language discussion

Current standard SBML

ODE mafia –

DARPA funded effort to produce next generation language – based on productivity rather than just code – debate over whether should be one language or several

Should NSF fund language-development effort????

ALEX:  SBML is barely funded; nothing else is funded at all.  People trying to put everything into that, rather than develop anything else.

Frank:  It’s mandated to publish in certain arenas – can’t publish if you don’t use it. 

Linda:  Don’t think publications should dictate language.

Kim:  Mafia redefines the problem when they can’t solve it.  Leads to community going on tangent that isn’t relevant.

Useful to figure out how to share algorithms, data; 

Shouldn’t force everything into a data model that can’t handle it.

Won’t have a modeling language to handle everything so should have subsets of the language.

Can’t have one framework for all models

Structure to keep track of model & model assumptions;

How to transfer data – information - knowledge?  Is/can CI provide the structure to do this?  How?

How to get people to talk to each other.  How to create an infrastructure that allows people to find common problems and to collaborate?  Need biology equivalent of math centers?  Gateways/other programs at NSF?

CI OPPORTUNITIES

Scale X detail – roughly constant (may increase constant a bit by faster computation)

GRAND CHALLENGE PROBLEMS

1.  Multi-scale multi-physics modeling
2.  Minimize collateral outcomes in drug targeting – 99.9 percent works .1 percent fails; fix that
3.  Data integration – biological and chemical and clinical and from old literature (can we get a database that provides structure in old papers – the results, their validity , error bounds, which other results they pre-or post cite
4.  Synthetic biology – re-engineering natural systems to produce things we need
5.  Run with the models – use them for control  and system identification
6.  Understand biological systems; for now models for us are models for basic science

REX ON MAIN POINTS

Each community surfaced their own favorite problems.  Needs to see the connection between CBET and CI. 

Connecting themes:

1.  Little explicit demand for high-end computing (except within CI);  want reliability; want to move data, want a utility  etc. no real demand for the high end stuff

2. Demand for algorithm/software engineering.  That isn’t featured as much in what we talk about in CI generally

3.  Lots about “give us the tools to interpret data; extract knowledge; assuming it means more than conventional data mining”

4.  Need for managing dynamic data and information flows

5.  Security – everyone needs it and wants it, but don’t want to be bothered with it.

6.  Didn’t hear much about visualization of data; but lots about interpreting data.

7.  There was a flurry of enthusiasm about community based models – but didn’t hear it echoed by the whole group – would see this as an area where we could identify groups – for sharing and developing algorithms.

Discussion:

Stan:  Integrating theme – coupling an economic force to make progress; key force for him is the set of mark-up languages (i.e. to write web); languages are in the public domain, but the algorithms behind them are behind the languages.

Really heard that we need languages to express ourselves; to give clarity of what we’re trying to model or achieve.

Alex:

Need:
1)  to exchange models in a real way 2) need tool interoperability 3) a way for people to describe or build without having a huge programming expertise.

Harris:  lot of good info on visualization – need to get people to use it.  Too “touchy, feely” for engineers.

Marlon:  Inflicted geographic mark-up language on some grade students.  Probably don’t want to get a consortium of folks to create a language.

Linda:  Put WWW at the top of tools.  Call it performance for high computing – rather than high performance computing.

Miguel B:  Need to determine where you need advances – in computing?  In concepts?  Don’t want to spend time proving a computational method to solve my problem – want to work on my problems.  Need to strike a balance between HPC and algorithms.  Can’t establish a general rule.

Kim:  Optimistic data:  Industry knows:  very gross variations from company to company on success rates on clinical trials.  At high-end – only 50 percent chance a drug will make it to market.  At low end – 10 percent. 

People in 50 percent category – has a better clinical trial process. 

Great CI will put more trials in the 50 percent success rate.

Floudas:  Advancing in HW and SW have to go hand in hand.

Tunde:  Follow up on clinical trials.  Where’s the customer feedback?  If you take Tylenol and it doesn’t cure your headache, the info doesn’t get back to the company.  Not a good standard in manufacturing of meds – e.g. active ingredient amounts vary widely.  Lack of customer feedback is a big problem.