How to get that first grant:
A young scientist’s guide to (AI)
funding in America
Jim Hendler
University of Maryland
All opinions are the author’s and do not reflect official policy of the AAAI,
the University, the US government or any other fund-granting organization
This is not a grant writing seminar!
 Many places offer grant writing seminars
– If you think you need help WRITING a grant,
take one.
 Offered by Universities, US Govt, others
– You should be able to take one at low cost or
free, high cost ones also exist
 This talk focuses at a different level than
those seminars
– grant strategies as opposed to writing tips
Outline
 Entre Acte: grant writing is a skill
 Background: the main funders of AI/CS
– Corporate
– DoD services
– DARPA
 Grant Advice
– some hints from researchers and funding agts
Entre Acte
 The SCIENCE must be there
– Doing good science is a necessary condition to
receiving any grant from anyone
– It is not, however, sufficient
 Getting a grant is not a “game”
– it is a skill (like writing a good paper, giving a
good talk, etc.)
– Like these other skills it cannot be easily
taught, but there are pointers and tips that can
help
Terminology: Gifts, Grants, Contracts
LOW
s
t
r
i
n
g
s
a
t
t
a
c
h
e
d
 Gifts have few strings attached
– Corporate Education Funds
– Industrial Affiliates Programs
$
$  Grants are “monitored”
– Deliverables generally underspecified
$
– PM primarily interested in the Science
 Contracts, legal documents
– Deliverables specified
– PM has specific programmatic goals
HIGH
– PM may often direct/re-direct efforts
AI/CS funders
 Where do grants and contracts come from?
– Companies
– The National Science Foundation
– The Department of Defense
• Darpa
• Service Funding programs
• Labs
– Other Govt Agencies
• NIH, NLM, DoE, NASA, etc.
 The DoD funds more than 80% of CS research in
the US! (not incl. High performance computing)
Corporate Funding
Corporate Contracts
 Corporate funds generally answer to “The Bottom
Line”
– Your research must be “value added”
• one exception is education/training function
• Gift funds, unusual but not unheard of
– usually small compared to a contract
– You need to show where it will have a real, and usually
fairly direct impact
• Corporate contracts generally have strings attached in the form
of deliverables
– prototypes
– improvements to their systems
Corporate Contract Caution
 (Local) Corporations are often interested in more
than your technology
– Your name on their grants
– Your school’s reputation
– YOUR STUDENTS
 Take this as both opportunity and caution
– Your name is an asset you should keep control over
– Companies may be more interested in the university’s
CS undergrads, rather than your grad students
• You can end up a local corporate recruiter
-- for better or worse
Corporate Contract tips
 Personal contacts are everything
– Work closely with groups at local corporations
– Take advantage of university contacts
 The written proposal is often a “formality”
– money is committed based on statement of work
– but may be needed for higher level
• Often require iterations and patience
 Pitfall: Legal issues
– make sure licensing/patenting etc. are worked out in
advance (if applicable)
Government Funding
US Government Funding
 Important to understand the mechanics
– How is funding announced?
• RFPs and BAAs
– How is govt funding categorized?
• 6.1 - 6.3
– Who makes the funding decisions?
• Funding Decision making
• Program Oversight
Calls for proposals
BAA: Broad Agency Announcement
RFP: Request for Proposal
–
–
–
–
–
–
–
Goals of Program
Requirements for Proposers
Format of proposal
Program Budget
Deadlines and dates
Where to look for further information
Who is in charge
Finding out about funding
 All govt external funding (BAA/RFP) must be
announced in the “Commerce Business Daily”
– but that goes way beyond AI/CS
 Fedix Alert is a push service (keyword based) for
routing CBD announcements
– but free service doesn’t include all agencies (incl.
DARPA)
 Agencies announce RFPS & BAAs on their Web
sites
– But this means you need to monitor
Tips
 Many universities/companies have someone who
monitors the CBD
– You need to search them out if your school/company
doesn’t have an active grants group (ORAA)
 The PMs send out email to lists of people they
think might be interested
– some of the folks in your community are likely to know
• Form Alliances (it is NOT a zero-sum game!)
– If your PM knows your work you are more likely to be
informed
• more on this later
Government Funding Types
 6.1 Basic research
 6.2 Exploratory Development
 6.3 Advanced Technology Development
 (6.4 Development, Test and Evaluation)
 (6.5 Management and support)
Funding Decisions
 Virtually all federal grants receive some sort of peer review
– Scientific Review Panels (NSF, NIH)
– External Reviewers (NSF, AFOSR, ONR, ARL)
– Source Selection Teams (DARPA, DoE, etc,)
Panel/Reviewer Recommendations
DoD
NASA
NLM/NIH
NSF give PMs differing constraints
DoE
– Different agencies
re. these
reviews
• Regardless, someone must be impressed by your proposal
Program Managers
 The PM is a scientist or a scientific manager
who has proven him/herself to be very
knowledgable
– solid scientific background in their field
 The PM has different scientific objectives
than yours
– yours: Give talks, Publish papers, get grants
– PMs: Give talks, Publish papers, get grants
• but very different than yours
Program Manager’s Needs
 The PM must defend his/her program to higher
management
– Program Objectives
• How it will have scientific &/or military impact
– What great research/researchers have been supported
• Awards won
• Papers written
• Transitions to industrial or military development
 Your work must help him/her do this!
National Science Foundation
NSF Requests for Proposal (RFP)
 NSF accepts two kinds of proposals
– Unsolicited (Sent in on “open” RFP)
• goes to a PM in an area for action
• Reviewed by panel or by external reviewers
– Special Programs
• Yearly
– Career
– Instrumentation
• Big “one of”s
– KDI, Govt I, RI
NSF Open RFP
 Objective: Support best basic research
– Typically have one or two Pis
• academic researcher
– generally small
• $50-100k/yr
– Stress personnel support
– PI support
– Grad Students
 Review criteria: Scientific basis, work plan
– Lack of work plan is #1 rejection reason
NSF Special RFPs
 Science with specific characteristics
– Support some specific govt objective
– encourage interdisciplinary research
– provide research infrastructure for universities
 Features
– Multi-PI (often multi-dept, university)
– Some go to $1M/yr or more
– Personnel and infrastructure
NSF RFPs cont.
 NSF Special Programs are a very good way for a
younger researcher to get funding
– but generally NOT as PI
• Especially good for interdisciplinary or team-oriented
researchers
 These larger grants are team oriented
– Most common error:
Team is formed for sake of grant
Grant makes that obvious
– Common Example: N PIs, N sections, each section has
one author
NSF Career Awards
 NSF has a special program for young researchers
– Science Based (whatever you may have heard)
• grant is more important than track record
– Panel reviewed
• Your work must stand out among similar projects
• Not everyone will be a specialist in your subarea
 Most common errors
– This is a grant, not a thesis chapter!
– Lack of enthusiasm (overly passive writing)
– “Trust me” doesn’t work in this program
• Assume everyone else is as good as you are!
NSF tip
 NSF is one of the few govt CSD funders
that returns reviews
– USE the reviews to revise and resubmit your
grant
• Never resubmit without serious rework/rethinking
– Assume the reviewers will be similar
• Assume the mistaken review was YOUR fault
– Pretend you must have written it wrong if you could have
given such a misleading impression
» (Example: If the reviewer says you didn’t discuss X,
but you did, assume you didn’t make it prominent
enough)
Department of Defense Funding
DoD Service Funders
 The Services each have their own 6.1 funding
organizations
– Air Force: AFOSR
– Navy: ONR
– Army: ARO
 Each organization has an “Open BAA”
– don’t get overly excited when the new BAA comes out
 These offices administer special programs for the
OSR(OFC DOD DDRE)
– These are like NSF special program RFPs
• generally for large funding, big groups (MURI)
• Sometimes specialized for govt priorities (Depscor, HBCU)
DoD Labs
 DoD labs can administer/award 6.2 and 6.3
funding contracts
– AF: AFRL (Wright, Phillips, Rome, others)
• Rome Laboratories (AFRL/I), Rome NY for IT and
especially AI
– Navy: NRL
• Naval Center for Applied Research in AI
– Army: ARL
• Adelphi Maryland: Agents, language, robots
 But, ...
DoD Lab BAAs
 Getting funding from the labs (without
Darpa in the loop) is pretty rare
– BAAs usually call for White Papers
– Geographic Proximity often a must
– Personal contacts and joint work often
prerequisite
 The labs put out calls for White Papers
– Watch for these!
– These are a good opportunity to get lab people
to know your work - can lead to funding in the
longer term
Know your Program Manager
 The people making the funding decisions
are trying to make sure they support high
quality work
– The quality of the scientists is important
• Track Record
– Knowledge of your past work by PM is often
important
• The PMs are typically scientists
• They enjoy meeting and talking to you
– (with certain caveats)
Know Your Program Manager
 You should try to get to know your PM
– Useful for NSF, very important for services
and DARPA
Note:PMs often come to conferences
– attend talks
– ask people about their work -- knowing who
is who can be useful!
Visiting (or calling) a PM
 Use the analogy of visitors to your labs
– You treat an interested student different from a know-itall kid
 Do your homework before your meetings
– Think about what you want to ask
– These are busy people, but they’ll give you time if
you’re not wasting it
• like you would a perspective student
 Be prepared to listen
– learn what the PM needs and is looking for
The White Paper
 A good outcome is if the PM asks for a
white paper
– A SHORT description of your research
 This is a great opportunity if you do it right
– Think of it as an “advertisement” for your work
• Why is what you are doing IMPORTANT
• What are you doing differently than others
– Use 3-4 pages
• Technical arguments must be focused
• Technical arguments must be precise
Service/Lab Open BAAs
 Rarely, if ever, is an unsolicted grant submitted
under an open BAA funded
– Don’t just send grant
• All grants received MUST be treated correctly
– lots of paperwork, headaches for PM and reviewers, especially
when there is no money in the budget
• Better:visit or call, white paper, invite for grant
– You can iterate the grant ideas with the PM
• don’t submit a final grant until it is ready
• Be responsive (especially about budget)
 Money is usually very limited
– being told there is no funding available is the expected
value!
DARPA
(Defense) Advanced Research Projects Agency
http://www.darpa.mil
The Key to Understanding DARPA
 DARPA is a results-oriented funding
agency
– They do like science and want/need to support
it, but that is not the primary mission
– The primary mission is to bring science to a
military customer
– That science must be “revolutionary”
• by MILITARY standards
– innovative science coupled with innovative doctrine or
new technology
Common Misunderstanding
 “Some people seem to get a lot of Darpa funding,
all those grants must be wired”
– DARPA is torn between two needs
– Support the best scientists
– GET RESULTS
• Going with a proven performer lowers the risk
– Already a well-known scientist
– track record of getting results
– The DARPA review process is closed, so there is no
feedback if you are not funded
• helps perpetuate this misunderstanding
DARPA Reviewing
 A panel of reviewers, usually all from DoD
entities and/or national laboratories, rates all
grants and ranks them with respect to fixed criteria
– All proposals are ranked into two categories
• Selectable
• Not selectable
 The BAA specifies the criteria of review
– The PM chooses weighting factors within criteria
 The PM funds some, but not all, of the selectables
as his/her funding allows and based on program
needs
Example Review Criteria/ranking
1. Overall scientific and technical merit of the proposed
program (30%)
2. Applicability of proposed technologies to <program area>
(30 %)
3. Expertise and experience in <program area> and ability to
complete the proposed work successfully (25%)
4. The degree to which technical data and/or computer
software developed under the proposed contract are to be
delivered to DARPA with unrestricted rights (15%)
5. Proposed cost, cost realism, and availability of funds.
Applicable only in case of significant cost estimate error
(0%)
– Legally budget not included in selectable/not selectable ranking!
DARPA desires new ideas!
 Darpa wants to bring in young scientists and
new ideas
– e.g.: Darpa sponsored a workshop at AAAI ‘97
for young AI scientists to present their ideas to
a number of PMS
(many of these folks were funded under 1998 BAAs)
 BUT… DARPA is a results-centered
organization
– most young scientists haven’t learned to portray
their ideas appropriately
A Funded NSF Proposal
 NSF grant on planning (Nau/Hendler):
– Developing control strategies which use knowledge about goal
interactions to guide planning by making better selections among
alternatives
– Exploring how limitations on goal interactions lead to
characteristics of task networks which constrain the size and
structure of the search space, thus enabling hierarchical planning to
be done more efficiently
– doing a theoretical analysis of several competing planning
strategies in terms of how knowledge about limited interactions
can improve their search efficiency
– implementing the results of the above research in a planning
system for the efficient solution of problems of practical interest
A Funded DARPA Proposal
 DARPA grant on planning (Nau/Hendler):
– Make realistic manufacturing example problems and knowledge
bases available to members of the <program> in a manner usable
by the technologies being developed under current <program>
support
– Develop an ontology of machining information in the standard
ontology package being developed by <program> researchers
– Evaluate <program> work on realistic manufacturing
problems…and provide comparisons with current tools used in
manufacturing and manufacturing research
– Provide case-bases of manufacturing plans accessible to
<program> participants …
– Make <program>-based manufacturing results and tools available
to the manufacturing community via the NIST tesbed
See the difference?
 For NSF we stressed the scientific goals
– we later backed them up with details from the
application of the technology to show that the
theory was viable
 For DARPA we stressed the application of
the technology
– we later backed it up with scientific details to
prove we had scientific credibility and
knowledge
Specific DARPA hints
 Work with people who already know the
DARPA world
– DARPA generally funds multi-PI projects
and/or “consortia” of contract/subcontract
groups
• Form strategic alliances with better known
researchers
• Team with corporate research groups that have
DARPA experience
Specific DARPA Hints
 BEFORE a BAA is written, the DARPA PMs
often sponsor a workshop on the topic to help
them put together a credible scientific program
– Invite only, usually only to senior researchers
• great if you can get an invite
– Notes from these workshops are made available on the
Web and elsewhere
• PMs and participants will steer you towards them
– After a BAA comes out, search for those notes!
• Can provide valuable hints as to what aspects of the
technology are seen as most important
DARPA Note
 Once a BAA is announced, the PM is legally
constrained NOT to provide information to
potential bidders UNLESS the discussion is shared
with all potential bidders
– A web site usually is created with an address for
sending questions and a FAQ of those that are legally
answerable
– The site may also include pointers to lists of potential
bidders who have expressed interest in partners
• Don’t hesitate to contact them
Grant Tips from the “Pros”
TIPS From the “Pros”
I asked a number of program
managers, senior scientists, wellfunded researchers, and
participants in funding review
panels to join me in providing tips
based on our experiences
The Number 1 Tip
Respond to the BAA/RFP
– Provide EVERYTHING that is requested
– Make sure your description matches the
program needs!
 This may sound simple, but virtually all of
us have encountered proposals that were
about good research, but not responsive to
the BAA or which omitted key points!
– They were not funded
Good Grants
 A good grant “tells a story”
– What is the problem to be solved?
– Why is wrong with current approaches?
– Why is your new idea?
– What is your technical approach to this idea?
– Why are you (your team) the one(s) to do it?
– How are you going to show that it works?
More general grant tips
 Back up what you propose to do with what
you’ve already done
– A funded proposal must describe work that
doesn’t yet exist, but at the same time, the
reviewer must be convinced you can do it.
 Show enthusiasm for your work
– if you don’t love it, neither will the reviewers
Know your audience
 It is critical to consider who will be reviewing
your grant
– NSF: (Academic) scientists in AI &/or CS
– ONR,OSR,ARO: military lab scientists
– DARPA: military scientists, operational military
 Rule of Thumb: Someone on the panel must think
yours is THE BEST
Readability is important
 A typical reviewer (on a panel) is reading a
lot of similar grants in a short amount of
time
– Make his/her life easier!
• Highlight key points
• Repeat things you want them to be sure of
– tell em what you’re going to say, say it, tell em what you
said
• Use figures/graphs where they can help make an
obscure point understandable
– space is limited, but this is worth it!
Budgets
 A smaller grant is ALWAYS easier to get
than larger
– example: There is a myth that DARPA budgets
should be large -- this is false (and a bad
strategy)
• remember selectable  funded!
 Don’t ask for more than you can
realistically spend
– A key new issue: expenditure rates!
• Years don’t have to be equal
– example: if you can’t hire all students immediately, ask for
less year one
One last tip…BE VISIBLE
 Too many young scientists avoid “time
wasting” things like program committees,
ed boards, workshop/symposium
organization, outside talks, etc.
Save some of your time for these!
– name recognition is important
– a reputation as someone who “gets things done”
looks great on a review form
SUMMARY
 There is no magic to writing a good grant, it
is a skill that can learned
– Learn from mentors
– Learn from your mistakes
– Learn from good examples
 The PMs are crucial
– Get to know them
– Get them to know you
Http://www.cs.umd.edu/~hendler