Received: 11 May 2021 Accepted: 13 January 2022

DOI: 10.1002/aaai.12032

ARTICLE

The NSF Convergence Accelerator program

Chaitanya Baru Lara Campbell Aurali Dade Pradeep Fulay Alex Loewi
Douglas Maughan Ibrahim Mohedas Linda Molnar Michael Pozmantier
Michael Reksulak Shelby Smith Nicole Tehrani

National Science Foundation

Abstract
Correspondence The National Science Foundation’s Convergence Accelerator is a unique pro-
Chaitanya Baru, Convergence Accelerator,
Office of the Director, National Science
gram offering researchers and innovators the opportunity to translate research
Foundation, 2415 Eisenhower Avenue, results into tangible solutions that make a difference for society. Through an
Alexandria, Virginia 22314. intense innovation curriculum and a mentorship program, researchers gain skills
Email: [email protected]
and experiences that are of use not only during this program but throughout
Chaitanya Baru is now at the University of their careers. This article describes the NSF Convergence Accelerator program
California San Diego. He was previously
and its initial funded convergence research topics—or “tracks”—funded in 2019
on assignment at NSF.
and 2020. In almost every track and NSF-funded project, artificial intelligence
Aurali Dade is on assignment at NSF from
George Mason University, Fairfax, VA. and machine learning (AI/ML) approaches and methods are playing an essen-
Pradeep Fulay is on assignment at NSF tial role.
from West Virginia University,
Morgantown, WV.
Michael Reksulak is on assignment at
NSF from George Mason University,
Fairfax, VA.

INTRODUCTION include hardware prototypes; new functions and capabil-

ities integrated into existing systems; open-source tools,
As a funder of research and education across all science data, and knowledge products; expansion of tools and ser-
and engineering fields, and with well-established relation- vices into new markets, for new audiences—keeping in
ships with global universities and funding agencies, the mind, in all cases, the sustained impact beyond the dura-
National Science Foundation (NSF) is uniquely positioned tion of the particular project itself.
to accelerate discovery and innovation for societal impact. The program is founded on the fundamentals of con-
In early 2019, the NSF launched a unique program, the vergence research—requiring a deep integration across
Convergence Accelerator to build upon basic research and disciplines and expertise to address specific challenges or
discovery and accelerate solutions toward societal impact. opportunities arising from pressing societal and national
The goal of the Convergence Accelerator is to serve as needs (NSF 2016)—and acceleration. Acceleration is
a platform for transformative cross-cutting partnerships key to ensuring that teams deliver impactful results to
that bring together stakeholders from different disciplines, an identified community of users within a 3-year time
expertise, and organizations to solve national-scale chal- frame. To achieve these objectives, the program requires
lenges. While the main focus is to enable long-lasting soci- partnerships from industry, academia, nonprofits, govern-
etal impact, results from each solution will vary and may ment, and other communities of practice to support the

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited and is not used for commercial purposes.
Published 2022. This article is a U.S. Government work and is in the public domain in the USA. AI Magazine published by Wiley Periodicals LLC on behalf of the Association
for the Advancement of Artificial Intelligence

6 wileyonlinelibrary.com/journal/aaai AI Magazine. 2022;43:6–16.

 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AI MAGAZINE 7

NSF Convergence Accelerator Model

Convergence Research Focus

• Selected track topics are released in a program solicitation
• Awarded projects are selected based on research track focus, acceleration of
basic research, multidisciplinary approach and national-scale societal impact
• Funded teams are composed of diverse disciplines, expertise, sectors Societal Impact
Ideation Process and communities
• Teams merge ideas and share approaches and techniques to speed • National-scale
use-inspired solutions toward real-world application
• Topics begin by gathering • Sustainability beyond
ideas from the community • Teams have no size limits and are expected to evolve as projects advance
NSF support may include:
through a Dear Colleague • Partnerships strengthen each funded effort by providing end-user insights,
Letter, Request resources, services, infrastructure and transition-to-practice pathways • Follow-on funding
for Information and investment

• Promising ideas that meet • Integration into

the program’s criteria existing systems
Phase 1 Phase 2
are further explored in • Tools/knowledge
funded workshops accessible through
• Funding up to $750K • Funding up to $5M
• The workshop findings open source 
• Nine months planning phase • 24 months solution
assist NSF in selecting development phase • Solutions expanded
• Teams participate in a innovation
future convergence curriculum to strengthen • Focuses on developing sustainable into new markets
research track topics collaboration, foster partnerships and impactful project deliverables
and transform basic research into
high-impact solutions
• Teams participate in an
idea-to-market curriculum to
• Teams participate in a formal phase unleash an entrepreneurial
2 proposal and evaluation to be mindset and skill set and to
selected for phase 2 ensure each project reaches
its full potential

FIGURE 1 The National Science Foundation (NSF) Convergence Accelerator program

teams and strengthen the solutions. Partners may provide plinary group that spans a wide range of disciplines and
end-user insights, resources, services, infrastructure, and sectors; focuses on outcomes and deliverables feasible in
transition to practice pathways. a 3-year time frame; and considers how the set of projects
within the track might work together to deliver success-
ful results. The findings of these workshops assist NSF in
PROGRAM MODEL developing the final track topic to be funded in the next
year.
The Convergence Accelerator program model has three
phases: topic identification; convergence research phase
1; and convergence research phase 2. At the completion Convergence research phases 1 and 2
of the convergence research phases teams are expected
to deliver high-impact solutions that meet societal needs A solicitation is issued each year by NSF with the identi-
and continue to have an impact after NSF support ends. fied convergence research track topics to solicit proposals
Figure 1 provides an overview of the program model, from multidisciplinary, multisector teams to develop solu-
from ideation to convergence research phases 1 and 2, and tions aligned to these topics, as exemplified by the FY2021
anticipated societal impact. solicitation (NSF 2021a). To encourage submissions from
industry and nonprofit submitters, the solicitation is
released through a Broad Agency Announcement (BAA),
Ideation process: Topic identification in addition to the standard NSF research solicitation.
In addition to the key NSF criteria of intellectual merit
Convergent research topics emerge during the ideation
and broader impact, phase 1 proposals are also assessed
process. Each year, the program issues a Dear Colleague
based on the extent of their multidisciplinary approach to
Letter, Request for Information (DCL/RFI) to gather ideas
the problem; identification, or potential to identify, multi-
for potential track topics, such as the ones for FY 2021
sector partners for the team; clearly articulated deliverables
and FY 2022, respectively (NSF 2020a; NSF 2020b). Ideas
and the likelihood of achieving those in the 3-year time
must be built upon foundational research; require con-
frame; and how well the project goals are aligned with the
vergence and a multidisciplinary approach; and be suffi-
theme of the track.
ciently broad in scope to address large-scale societal chal-
lenges. Ideas that meet the program’s criteria are funded
by NSF to organize community workshops. An estimated Convergence research phase 1: Planning
10–15 workshops are funded each year to explore poten-
tial track topics that incorporate convergence research. A In the Convergence Research phase 1, teams may receive
successful workshop would bring together a multidisci- up to $750,000 for a 12-month duration. During this
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 AI MAGAZINE

phase, teams develop their initial idea into a proof sition, and the challenge the solution is addressing. This
of concept while refining team composition, including includes developing a Public Executive Summary.
adding new team members and partners, if and as
needed. Teams participate in an intense hands-on 9-month The experience thus far has demonstrated that teams
Innovation Curriculum designed to help transform their benefit greatly from this innovation curriculum. Teams
ideas into proofs of concept, including refining the com- have applied the curriculum to modify their original
position of end-user communities, understanding their approach and plan, form new partnerships, interview
requirements, and identifying key partners vital to phase larger numbers of diverse stakeholders—all of which have
2 success. greatly assisted them in refining project goals and deliver-
ables.

Innovation curriculum
Preparing phase 1 teams for phase 2
The Convergence Accelerator’s Innovation Curriculum is
a critical aspect of the program and phase 1 activities. All phase 1 projects are eligible to submit a phase 2 proposal
The curriculum is delivered through a hybrid model that for up to $5 million over 2 years. The review process at the
includes content delivered asynchronously by online video end of phase 1 includes a formal proposal evaluation and
modules and supplementary activities, as well as syn- an oral pitch.
chronous live sessions. Members from all teams partici- The phase 2 proposal is a traditional NSF proposal that is
pate in the curriculum. “Core members” from each team evaluated by an NSF peer review panel. The review criteria
are required to participate in the live sessions. for phase 2 are similar to phase 1, but include an empha-
In addition to regular engagement with the NSF pro- sis and focus on convergence research, partnerships, deliver-
gram, each team in phase 1 is also assigned a coach to ables, and track alignment. It is expected that phase 1 activ-
help navigate through the Acclerator program, develop ities and experiences would influence the project’s direc-
pitch presentations to diverse audiences, prepare phase tion for phase 2.
two proposals, and develop strategies for creating a Proposals are required to describe the impact of their
sustainable solution beyond phase 2 NSF funding and phase 1 activities on their strategy and tactics going forward
support. Coaches participating in the Convergence including modifications, if any, to their original project
Accelerator program have seasoned entrepreneurial plan. The proposals are expected to provide a detailed time-
backrounds and are experienced in mentoring early line of phase 2 milestones and deliverables, along with a
stage startups and transitioning solutions to the project coordination and management plan to help ensure
marketplace. that projects are realistically able to achieve their goals and
The innovation curriculum consists of team science, objectives over the 24-month duration.
human-centered design, user discovery, early-stage proto- Projects are also evaluated on an oral pitch made to a
typing, communications, storytelling, and pitching and is separate pitch panel, which includes panelists with techni-
designed to position teams for success in accelerating their cal expertise as well as expertise in entrepreneurship and
solution forward and preparing them for phase 2. industry technology transfer. The pitch panel receives only
a 2-page Public Executive Summary in advance. Each team
∙ Team science – helps teams develop an internal dynamic is given 10 min to communicate information about the
that supports effective engagement focus on a unified project, such as phase 1 key activities and outcomes and
project vision and common goals. the team’s phase 2 plan. This is followed by a 20-min Q&A
∙ Human centered design – teaches how teams can lever- session.
age a collective process to produce tangible outputs Proposals are evaluated based on the outcomes from
rooted in the needs of people who will use the solution. both the NSF peer review and the pitch review and selected
Through this process, teams are able to identify end- teams advance to phase 2.
users and key stakeholders, conduct user interviews to
glean new insights, and build low-fidelity prototypes to
elicit feedback. Convergence research phase 2:
∙ Communications, storytelling, and pitching – commu- Implementation
nication is critical to getting the team’s ideas in front
of various stakeholders including potential partners, Throughout the 2-year phase 2 effort, teams continue to
investors, and end-users. Teams learn how to effec- work on their solutions while also participating in an
tively communicate about their project; its value propo- entrepreneurial curriculum. By the end of phase 2, each
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AI MAGAZINE 9

solution is expected to have found a path to sustainability phase 1 booths for track E: Networked Blue Economy and
beyond NSF support. track F: Trust & Authenticity in Communication Systems
(see Table 1).

Entrepreneurial curriculum
PROGRAM STATUS
The entrepreneurial curriculum in phase 2 is
based on the Idea-to-Market curriculum (I2M The Convergence Accelerator program is currently com-
2021) supported by Global Innovation Catalyst posed of three cohorts, started in 2019, 2020, and 2021,
(https://globalinnovationcatalyst.com/) consisting of respectively. Each cohort focuses on two research track
modules in product development, intellectual property, topics. The 2019 cohort is in year 2 of phase 2; the 2020
financial resources, sustainability planning, and communi- cohort is in year 1 of phase 2 and the 2021 cohort is in its
cations and outreach. It is a structured approach designed phase 1. Table 1 shows the status of the program’s portfolio.
to assist teams in defining and developing an execution Themes for the two inaugural tracks in the Convergence
roadmap and sustainability plan and to help push the Accelerator—track A: Open Knowledge Network (OKN)
projects in reaching their full potential. and track B: AI and the Future of Work—were derived
from the NSF Big Ideas on Harnessing the Data Revolution
and the Future of Work at the Human–Technology Fron-
Track integration tier, respectively (NSF 2019a).
As shown in Table 1, in September 2019, 21 projects were
To ensure that each track outcome is greater than the “sum selected for track A phase 1 and 22 for tracks B1 and B2
of its parts,” phase 2 teams also work together on track inte- phase 1. Of these, five from track A and three from track
gration to try to create solutions that are more than what B (two B1, one B2) were selected for phase 2, in Septem-
may be produced by each individual project. Up to 5% of ber 2020. All projects are typically characterized by mul-
each team’s phase 2 budget/effort is set aside for this inte- tisector partnerships involving academia, industry, gov-
gration activity. ernment, nonrofit, and other sectors; a multidisciplinary
project team consisting of a breadth of expertise includ-
ing subject matter expert as well as technology experts,
Convergence accelerator expo and, clearly articulated deliverables to be created by the
end of phase 2. The following sections provide an overview
All teams participate in an annual Convergence Accelerator of track A and tracks B1/B2, as well as a brief overview
Expo that is typically held in July featuring the program’s of the eight projects that entered phase 2. Seven of these
full portfolio. The Expo provides teams the opportunity to projects are described in further detail in separate articles
showcase their solutions to a wide range of stakeholders. in this special issue. The NSF Program Director for track
Teams may find new users, customers, and partners A is Dr. Lara Campbell and for track B is Dr. Linda Mol-
to assist in accelerating solutions forward, and/or even nar. This special issue also includes an article that provides
investor prospects. This outreach event is an opportunity a brief overview of all phase 1 projects in track D whose
for teams to highlight their project deliverables and theme—data and model sharing to enable AI innovation—
achievements and also help teams prepare communica- is directly related to AI. The NSF Program Director for
tion and marketing products that can be used well beyond track D is Michael Pozmantier.
the event per se.
Due to the pandemic, the 2020 and 2021 expos were
virtual. In 2020, the Expo featured 43 phase 1 booths Track A: Open knowledge network
(21 in track A and 22 in track B). In 2021, Expo 2021
(https://nsf-ca.vfairs.com/) featured 40 booths, which Track A focuses on use-inspired research to create an
included 10 phase 2 booths (five in track A, three in track Open Knowledge Network (OKN) representing informa-
B, and two track integration booths); 29 phase 1 booths tion about real-world objects and relationships among
(11 in track C and 18 in track D), and one Convergence objects, as a critical part of a national data and knowl-
Accelerator program booth. Expo 2022 will also be vir- edge infrastructure, as originally mentioned in the NSF’s
tual and is scheduled for July 27–28, 2022. At Expo 2022, HDR Big Idea (HDR 2017). Vast amounts of data are pro-
participants will have the opportunity to see novel solu- duced daily, yet many organizations lack the tools and
tions across six convergence research track topics, includ- resources to draw insights from these data and to make
ing 18 phase 2 booths from tracks A, B, C, and D and 28 data-driven decisions. Proprietary versions of knowledge
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 AI MAGAZINE

TA B L E 1 National Science Foundation (NSF) Convergence Accelerator Portfolio

Projects funded during each convergence research phase
Cohort year Track: topic Phase 1 Phase 2
2019 Track A: Open Knowledge Networks 21 5
Track B: AI and the Future of Work 22 3
2020 Track C: Quantum Technology 11 4
Track D: AI-Driven Innovation via Data and 18 6
Model Sharing
2021 Track E: Networked Blue Economy 16 September 2022
Track F: Trust & Authenticity in 12 September 2022
Communication Systems

networks/graphs already power some of today’s most pow- information via link traversal and also enabling deducing
erful applications in web search (for example, Google of linkages among entities, and (d) “horizontal” and
Search, Bing), dialog-based systems (for example, Siri, “vertical” elements. Horizontal elements and activities
Alexa), e-commerce (for example, Amazon), and social would support development of a common technological
media (for example, Facebook, Twitter), while Wikidata infrastructure, regardless of the knowledge domains, such
provides the most significant example in the open world. as data curation/refinement services, data provenance,
Access to an OKN that contains open scientific and govern- data integration services, and user-friendly tools, appli-
ment data and information, while also linking to access- cations, and interfaces. Vertical elements and activities
controlled information, would open the gates to a new would encompass ingestion and curation of content for a
generation of powerful applications serving societal needs. wide variety of information domains with activities such
Furthermore, researchers would be able to develop expres- as extraction of information from structured/unstructured
sive frameworks to capture knowledge and design user- sources, defining ontologies, and mapping across ontolo-
centered interfaces to access knowledge at scale for real gies, supporting applications in many different domains.
applications. Organizations—regardless of their size or A report released in March 2021 by the US National
sector—would be able to take advantage of this networked Security Commission on AI noted that, “OKNs (or repos-
infrastructure to provide better services and products. itories) with massive amounts of world knowledge could
In a workshop organized in October 2017 by the Big Data fuel the next wave of AI exploration, driving innovations
Working Group of the Federal Networking and Information from scientific research to the commercial sector” (NSCAI
Technology Research and Development Program, stake- 2021). It went on to encourage “A vision to create an open
holders from Federal agencies, industry, and academia knowledge graph of all known entities and their relation-
concluded that an open, community-driven OKN effort ships, ranging from the macro (for example, have there
would enable stakeholders from all sectors—government, been unusual clusters of earthquakes in the US in the past 6
industry, academia, nonprofits, and others—to share the months?) to the micro (for example, what is the best combi-
effort and cost of developing this framework and infras- nation of chemotherapeutic drugs for a 56-year-old female
tructure (NITRD 2017). This OKN would be an essential with stage 3 brain cancer?). The OKN is meant to be an
component of AI-based solutions by adding context and inclusive, open, community activity resulting in a nonpro-
world knowledge to machine learning systems, enrich- prietary knowledge infrastructure that could facilitate and
ing data with extensive information about the underly- empower a host of applications and open new research
ing objects, and enabling natural language systems to link avenues, including how to create trustworthy knowledge
words and sentences to descriptions meaningful to human networks and graphs.”
users. Workshop breakout sessions discussed cross-cutting
“horizontal” topics in technology as well as domain-based Track B1 and track B2: AI and future jobs
“vertical” issues. The need for this impactful new mode of and national talent ecosystem
data-driven discovery was further endorsed by the NSF Big
Idea on Harnessing the Data Revolution. Track B focuses on the convergence of AI technologies
Key characteristics of this OKN would include (a) across a wide range of future of work issues including
dynamic structure, reflecting real-world information workforce training and education, curricula and skills
updates and changes as they occur, (b) open system, to training development, and worker-work matching. The
accommodate input from a variety of sources, (c) link- landscape of jobs and work is changing at unprecedented
ages among entities, to enable linking across disparate speed, enabled by advances in many areas including
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AI MAGAZINE 11

computing technologies (for example, AI and robotics); cles for recruitment and engagement, with the potential
deeper understanding of societal and environmental for wider implementation by industry, educational insti-
change; advances in the cognitive and learning sciences; tutions, and other stakeholders engaged in the cocreation
the development of pervasive, intelligent, and autonomous of a national talent ecosystem. AI convergent research
systems; and new concepts of work and the workplace. efforts may include leveraging AI to adaptively personal-
The ongoing COVID-19 pandemic has brought about even ize the learning content and learning experiences, includ-
more change and more urgent need to address work—how ing AR-based learning applied to fields such as emergency
and where it is done and, importantly, the workers who response (ER).
are doing it. Many American organizations currently
suffer from a talent gap. Not only do current workers lack
the requisite skills to perform 21st century work but many Track A phase 2 projects
new graduates moving into the marketplace also lack
those skills. In addressing these Future of Work issues, A brief outline of the five track A phase 2 projects is pro-
the goals and objectives of track B are in line with NSF’s vided. Specific issues related to each project are further dis-
R&D priorities in ensuring accessibility and inclusivity; cussed in separate articles in this issue.
securing global leadership; and advancing the frontiers of The Open Knowledge Network Infrastructure project
research into the future (FWHTF 2020). (Cafarella et al. in this issue) is developing software to
Track B1 on Artificial Intelligence and Future Jobs support all phases of the lifecycle of knowledge net-
focuses on R&D leading to technological tools to connect work creation and use, addressing the lack of an effi-
individual workers with jobs, and on providing periodic cient and easy data-oriented programming environment
retraining and reskilling outside of traditional educational for this purpose. Tools and capabilities are being devel-
settings, which will be increasingly necessary and integral oped for data ingestion, data curation and knowledge
to successful lifetime careers. Ensuring fair, ethical, and graph refinement, provenance tracking, and develop-
equitable treatment of workers is a key guiding principle in ment of applications that use knowledge networks. The
the creation of tangible deliverables. Projects may consider CORD-19 dataset on COVID-related publications (CORD
a broad range of factors impacting employment and train- 2019) provides one application domain. The project is
ing, such as intersectional identities (for example, gender, also working with the SCALES project on court records
race, ethnicity), disabilities, socio-economic status, and data.
contextual factors (for example, family responsibilities, as The KnowWhereGraph (KWG) project (Janowicz et al.
well as access to education, and technology). Projects may in this issue) is addressing the need for efficient repre-
focus on particular industries or regions, specific popu- sentation and use of spatial information in knowledge
lations such as veterans and autistic/neurodiverse indi- graphs while also developing tools to provide geospa-
viduals, or particular workplace types such as small busi- tial data (“enrichment”) that can be tailored to a broad
nesses, decentralized manufacturing, medical facilities, or range of applications. Spatial information is “semantically
K-12 schools. AI technologies utilized may include NLP lifted” and representing in the knowledge graph to enable
and understanding; machine-learned models for matching answering questions about spatial properties of entities,
skills with training; and an intelligent coach “virtual teach- and spatiotemporal relationships among them, such as
ing assistant” to gain trust and encourage engagement. “What entities are located here?,” “What events occurred
Track B2 supports convergence R&D leading to inno- here in the past?,” and “How does region X compare with
vative approaches for employers to support their work- region Y?” for any given region on earth. The project is
ers seeking the skills required for 21st century work and focusing on applications in disaster relief and supply chain
technologies of the future. Track B2 efforts may devise management.
R&D plans addressing multiple approaches toward re- The Systematic Content Analysis of Litigation EventS
envisioning the concepts, structures, and technologies Open Knowledge Network (SCALES) project (Amaral et al.
needed for employers to support continuous learning for in this issue) is a “vertical” project in the judicial records
dynamic, digitally intensive work, and provide access to domain. It seeks to transform the transparency and acces-
skilled talent pathways, mentors, and authentic workplace sibility of court records by making federal court records
experiences. Projects are informed by research on STEM information easy to access and query. Current court
learning, engagement, and its social context, as well as recordkeeping practices produce rich but largely unstruc-
research on organizations and collaboration. Projects may tured text data, the study of which is further complicated
focus on prototyping innovative approaches such as learn- by being behind a paywall (Sanga and Schwartz 2020).
ing environments, platforms, interfaces, or simulations, The project has developed AI techniques for computa-
tools for analysis, assessment, or prediction, and vehi- tional understanding of the free text of docket entries,
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 AI MAGAZINE

that allows records to be understood through the different of the underlying knowledge, skills, and abilities (KSAs)
practices employed by each jurisdiction, judge, clerk, and required, or offered, to match training opportunities with
involved party. The team is already answering questions skills needs in an efficient, effective, and equitable manner.
for the courts themselves, and as they build their corpus of A “virtual coach” uses human-centered AI to answer ques-
data is also focusing on users including legal scholars and tions about SkillSync’s own operation, and reasoning func-
researchers, journalists, and policymakers. The project is tions, to improve usability. The project is creating “Intelli-
developing their Satyrn platform to integrate natural lan- gent Links,” which are AI-supported connections between
guage querying and visualization in a notebook-style inter- Employers and Colleges for training and skills develop-
face and make it easy for nontechnical users to effectively ment. The project is run by Eduwork with their partners
explore, query, and analyze the court records knowledge at Georgia Tech.
network. The Learning Environments with Augmentation and
The Biomedical Open Knowledge Network (SPOKE) Robotics for Next-gen Emergency Responders (LEARNER)
project (Baranzini et al. in this issue) is developing an OKN project (Mehta et al. in this issue) is developing an acces-
in the precision human health domain. The foundation sible, modular, personalized, and scalable mixed-reality,
for this is a large property graph that integrates approx- cloud-based learning platform for ER use to acceler-
imately 40 data sources to date covering a broad spec- ate skilling and reskilling of ER workers, particularly
trum of biomedical data. As part of the data ingestio and on nascent augmentation technologies, such as pow-
curation process, SPOKE uses established ontologies to ered exoskeletons and head-worn augmented-reality inter-
markup data to enable consistent linking across the knowl- faces. These new approaches have the potential to change
edge graph. The project is exploring development of visual the very nature of work in this domain, and improve
interfaces for easy navigation of the complex graph and a efficiency, health, and well-being of workers. A unique
“neighborhood” explorer to enable analyses that approach human-centered adaptive training framework is being
the data from different users’ perspectives (for example, developed which incorporates physiological, neural, and
diseases vs. genes). Specific tools are being explored for behavioral markers of learning into real-time AR and
biomedical researchers, the pharmaceutical industry (for virtual reality (VR) and force feedback-based exercise
example, drug repurposing), and clinicians to apply the delivery, along with user preference and past training
insights of this knowledge network. SPOKE is also striv- history.
ing to align with Biolink, a biomedical semantic standard
currently being established by the NIH/NCATS Biomedi-
cal Translator Consortium (BDTC 2019). Track integration
The Urban Flooding Open Knowledge Network (UFOKN)
project (Yeghizarian et al. in this issue) is developing As described earlier, projects within each track are also
flood forecasting and response capabilities by integrating required to work together on “track integration,” to
data across the urban multiplex. The general approach endeavor to create an overarching outcome that is greater
is broadly applicable to the natural disaster “vertical” than simply the sum of the individual efforts. The five track
domain. The approach is to integrate the Urban Mul- A phase 2 projects are developing a Data2Knowledge Con-
tiplex Inventory (UrMI) of multiple linked knowledge sortium, described below, to provide resources and insights
graphs (for example, building structures, transportation drawn from the individual projects’ tools and efforts, as
networks, power resources, utilities) with flood model data well as from cross-project connections, to support other
across the continental US, incorporating local flood mod- efforts to embrace the significant promise of knowledge
els and tools where available, and socio-economic data (for graph and OKN approaches.
instance from the US Census’ American Community Sur- The track B phase 2 projects are developing a Skills-
vey). The project is working closely with local and regional based Talent Ecosystem Platform for Upskilling—STEP UP
governments in pilot municipalities in Florida, North Car- to help develop evidence-based guidelines for selecting the
olina, Ohio, Virginia, and Minnesota as they work toward best type of training for acquiring a given set of skills in
a national rollout of this OKN. a multireality environment. In parallel, STEP UP is devel-
oping skills frameworks and associated skills-based creden-
tials for technical areas covered by the LEARNER project
Track B phase 2 projects and for first-line management leadership training, which
is a steppingstone to jobs that provide a living wage and
The SkillSync project (Robson et al. in this issue) is creating an upward career path. These will be used in a proof-of-
AI-based tools to develop an understanding of job postings, concept STEP UP portal scheduled for release by the end
work requirements, and training opportunities in terms of 2022.
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AI MAGAZINE 13

Track A: The Data2 knowledge (D2K) STEP UP will provide a platform with a set of capabili-
consortium ties that will (a) allow employers to create job profiles that
identify the skills needed for in-demand jobs; (b) enable
Projects within the track are working together to iden- individuals to create (and validate) skills profiles that iden-
tify specific areas of mutual opportunity. For example, tify the skills they have; (c) use AI to match skills profiles
the OKN Infrastructure and SCALES projects are work- to job profiles and to identify skill gaps; (d) enable individ-
ing on common tools and technical capabilities for infor- uals to find and enroll in training to fill skills gaps; (e) issue
mation extraction from and understanding of documents. skills-based credentials that validate newly acquired skills;
The KWG and UFOKN projects are working together on and (f) enable individuals to provide existing or potential
spatial representation of information and the Biomed- employers with those credentials. All of these functions
ical OKN and OKN Infrastructure projects are work- will be supported by AI-based techniques that extract skills
ing on use cases connected to COVID-related informa- information from unstructured data for use in job match-
tion, which can also draw in the KWG and UFOKN ing and recommendation algorithms using well-validated
projects (for example, compounded effects of flooding AI de-biasing techniques. A special focus is on ensuring
during a pandemic) and the SCALES project (for exam- that research in this area is well-connected to all segments
ple, impact of pandemic on court cases broadly, but also of the population, especially those that have been tradition-
characteristics of cases related to vaccination and mask ally underrepresented and historically excluded as part of
mandates). research focus and case studies. It is envisioned that new
These projects have identified a set of broad cross- workforce technologies and associated skills can be added
cutting themes across all OKN efforts such as: the impor- to the STEP UP platform providing new skills-based cre-
tance of tools and well-defined processes for ingestion dentials, training profiles, worker profiles, and job profiles
of new data into knowledge graphs and networks. This thus creating more opportunities to support different types
includes tracking provenance of data; understanding the of learning that are no longer served by a standard resumes
role of the human-in-the-loop—how far can an OKN go and cover letters (Cardenas-Navia and Jyotishi 2021).
in providing automated responses using inferencing and
other AI techniques and at what point does the human-
in-the-loop need to come into ensure accurate data inges- New tracks
tion side as well as effective and fair interpretation, and,
provisioning of core infrastructure, that is, what types of Tracks in the Convergence Accelerator are selected
computing and data platforms need to be readily available through an NSF RFI workshop process (other than first
to ensure effective end-to-end applications and maximize two inaugural tracks in the 2019 cohort, which were
benefits to society broadly. selected by NSF). The 2019 cohort included tracks A and
B, which are in year 2 of phase 2; the 2020 cohort included
track C on Quantum Technology (program director Dr.
Track B: STEP UP—skills-based talent Pradeep Fulay) and track D on Enabling AI Innovation via
ecosystem platform for upskilling Data and Model Sharing (program director Michael Poz-
mantier), which are in year 1 of phase 2. This special issue
STEP UP is a collaboration among the track B phase 2 does not include an article on track C since there are no
projects with the long-term vision of providing a transla- strong linkages to AI in that track. The article by Baru et al.,
tional platform and a transition-to-practice ecosystem not on Enabling AI Innovation via Data and Model Sharing,
just for the NSF Convergence Accelerator efforts, but to provides an overview of all 18 projects in phase 1 of track
all fundamental researchers and stakeholders involved in D. Brief descriptions of projects currently in the 2019 and
Future of Work, including researchers in the NSF FW-HTF 2020 cohorts are available in the 2021 Convergence Accel-
(Future of Work at the Human Technology Frontier) core erator Portfolio Guide (NSF 2019b).
research program and those working in this area with other Tracks in the 2022 cohort are track E: Networked Blue
federal agencies. The goal of STEP UP is to not only transi- Economy (program director Dr. Aurali Dade) and track F:
tion use-inspired research to practice to have positive and Trust and Authenticity in Communications Systems (pro-
lasting impact on American workers but also to create a vir- gram director Michael Pozmantier) (NSF 2021b). Tracks
tuous cycle of fundamental research and transition to prac- in the 2023 cohort have been announced. They are, track
tice. STEP UP especially focuses on underserved groups G: Securely Operating Through 5G Infrastructure (NSF
and the missing millions in parts of the country that are 2022a) and track H: Enhancing Opportunities for Persons
often overlooked. with Disabilities, Track I: Sustainable Materials for Global
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 AI MAGAZINE

Challenges, and Track J: Food & Nutrition Security (NSF program also makes it different. The notion of “track inte-
2022b). gration,” where all projects within a track work together
to address the larger national or societal challenge to make
the whole greater than the sum of parts, is yet another new
Lessons learned aspect of the program whose impact is also closely being
followed and monitored.
The Convergence Accelerator program has taken off
extremely quickly on a steep trajectory. Indeed, the pro-
Looking ahead
gram team likes to say, “We are building the plane as we
are flying it!” With the first two tracks (tracks A and B)
The Convergence Accelerator is on an exciting path. It is a
not yet completed, it is too early to comment on the pro-
unique, young NSF program focusing on translational and
gram’s impact. However, one measure of success is the
use-inspired research and accelerating solutions toward
enthusiasm that the researchers have shown for this pro-
societal impact. This is an approach to which NSF is com-
gram, its approach, and its curriculum. Based on com-
mitted. NSF has long supported all facets of fundamen-
munity response to the program and informal researcher
tal resarch, from foundational, curiosity-driven research to
feedback, the program appears to be addressing important
use-inspired, solution-oriented research, and translational
needs in the research community. The program team has
research. NSF is actively learning from experiences in the
learned that not all research areas and topics are necessar-
first few years and using those experiences to continuously
ily suitable for track topics—each research area needs to
refine and enhance the program.
be at the right level of maturity to admit multidisciplinary
So far, more than 400 organizations from academia,
convergent approaches, attract multisector partnerships,
industry, government, nonprofit, and others have engaged
and identify tangible deliverables that can be delivered
in the program, including those represented by the project
in a 3-year time frame. While the funded researchers
Principal Investigators (PIs), co-PIs, and Senior Personnel,
have been enthusiastic about the program, the program
and partner organizations; and interest continues to grow
also resonates best with researchers interested in mak-
about the program. The first few tracks—A, B, and D—
ing an impact via “deliverables” in a relatively short term.
are directly related to AI. Track F focuses on misinforma-
The program also offers an exciting learning platform for
tion/disinformation and includes a number of projects that
graduate students and postdocs interested in translational
are using AI as a tool. As the program proceeds ahead,
research and transition to practice, that is, those interested
NSF anticipates almost every future track topic will have a
in working on use-inspired research and working toward
strong connection with AI, given the ubiquity and impor-
deliverables. Unlike a more “traditional” postdoc mentor-
tance of AI in nearly every science and engineering field
ing plan, the mentoring plan for Convergence Accelerator
and topic area.
projects is expected to emphasize training of postdocs in
the areas of translational and use-inspired research, team
science, and project management for deliverables. CONFLICT OF INTEREST
The program’s innovation curriculum and the mentor- There is no conflict of interest.
ing provided by coaches have been very popular with the
teams—providing effective paths to bringing the teams up REFERENCES
to speed on issues related to translational research and BDTC (Biomedical Data Translator Consortium 2019). 2019. “Toward
transition to practice. Importantly, the Convergence Accel- A Universal Biomedical Data Translator.” Clinical and Transla-
erator program is not a competitor to the long-running NSF tional Science, 12 (2): 86–90. https://doi.org/10.1111/cts.12591.
Cardenas-Navia, Isabel and Shalin Jyotishi. 2021. “Everything You’ve
Innovation Corps (I-Corps), Small Business Innovation
Ever Learned.” Issues in Science and Technology XXXVII(4).
Research (SBIR), and Small Business Technology Transfer https://tinyurl.com/ye24ez3k
(STTR) programs. On the contrary, it complements those CORD 2019. COVID-19 Open Research Dataset (CORD-19).
programs by being situated earlier in the research develop- www.semanticscholar.org/cord19.
ment path toward practice. Unlike those other programs, FWHTF. 2020. Future of Work at the Human-Technology Fron-
success in the Convergence Accelerator program is not tier. https://www.nsf.gov/news/special_reports/big_ideas/
necessarily about commercialization—although commer- human_tech.jsp.
HDR 2017. NSF Harnessing the Data Revolution Big Idea. www.nsf.
cialization may be a perfectly good and useful outcome.
gov/cise/harnessingdata/.
Projects can have an impact without becoming commercial
I2M. 2021. Stanford Idea-to-Market course. Link to online resource.
enterprises, such as open-source efforts, nonprofits, acqui- https://online.stanford.edu/professional-education/idea-market.
sition by a forprofit or nonprofit organization, and other NITRD. NITRD Open Knowledge Network Workshop, Bethesda,
avenues. In addition to the requirement for multidisci- MD, October 4–5, 2017. www.nitrd.gov/nitrdgroups/index.php?
plinarity and partnerships, the track-based structure of the title=Open_Knowledge_Network.
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AI MAGAZINE 15

NSCAI. 2021. Final Report National Security Commission on Aurali Dade is a Program Director leading Track E:
AI. www.nscai.gov/wp-content/uploads/2021/03/Full-Report- The Networked Blue Economy of the NSF Convergence
Digital-1.pdf. Accelerator. She is the Deputy Chief Research Officer
NSF. 2016. Growing Convergence Research. https://www.nsf.gov/od/
at George Mason University and previously served in
oia/growing-convergence-research/index.jsp.
research leadership roles for two decades in three large
NSF. 2019a. NSF 19–050 Dear Colleague Letter: NSF Convergence
Accelerator Pilot (NSF C-Accel), March 2019. https://www.nsf. research universities. She has a PhD in Environmental
gov/pubs/2019/nsf19050/nsf19050.jsp. Science and an MS and BS in Biology.
NSF. 2019b. NSF Convergence Accelerator –2021 Portfolio
Guide. https://das-prod-nsf-gov.s3.amazonaws.com/2021- Pradeep Fulay is a Program Director for the Conver-
09/NSF%20Convergence%20Accelerator%202021%20Portfolio%
gence Accelerator program at National Science Foun-
20Guide_08312021_Low-Res_Final_508.pdf.
dation where he provided leadership for several pro-
NSF. 2020a. NSF 20–061 Dear Colleague Letter: Request for Informa-
tion on Future Topics for the NSF Convergence Accelerator and grams including quantum technology and providing
Call for Future Topics Conference Proposals, March 2020. https: equity and accessibility to persons with disabilities. He
//www.nsf.gov/publications/pub_summ.jsp?ods_key=nsf20061. is an elected Fellow of the American Ceramic Society
NSF. 2020b. NSF 21-012, Dear Colleague Letter: Request for Informa- and an author of three textbooks and several archival
tion on Future Topics for the NSF Convergence Accelerator, Octo- journal publications and patents.
ber, 2020. https://www.nsf.gov/pubs/2021/nsf21012/nsf21012.jsp.
NSF. 2021a. NSF Convergence Accelerator 2021 Cohort, Phases 1 and
2, March 2021. https://www.nsf.gov/pubs/2021/nsf21572/nsf21572.
Alexander Loewi is an AAAS S&T Policy Fellow at
htm-reviewcrit. the NSF Convergence Accelerator. His work has shaped
NSF. 2021b. Announcement of 2021 Cohort, Track E and policy interventions in public health and education and
Track F projects. https://www.nsf.gov/news/special_reports/ received Best Methodology from the American Political
announcements/092221.jsp. Science Association. Previously, he co-founded a suc-
NSF. 2022a. Joint NSF/DOD Phases 1 and 2 for Track G: Securely cessful public opinion research startup. He holds a PhD
Operating Through 5G Infrastructure, January 10, 2022. https://
in Public Policy, an MS in Machine Learning, and a BS
beta.nsf.gov/node/29578.
from Stanford.
NSF. 2022b. Upcoming Topics for the 2022 NSF Convergence Acceler-
ator Solicitation, DCL NSF 22–036, January 11, 2022. https://beta.
nsf.gov/node/29579. Douglas Maughan is the inaugural Head of the NSF
Sanga, Sarath, and David Schwartz. 2020. Tear Down This Judicial Convergence Accelerator. During his more than 30
Paywall. Wall Street Journal 2020. (Online article). www.wsj.com/ years of government service, Maughan assisted more
articles/tear-down-this-judicial-paywall-11607900423 than 75 commercial and open-source information secu-
rity products to market. He held positions at DHS S&T,
DARPA, and NSA and has a PhD and MS in Computer
AU T H O R B I O G R A P H I E S Science and a BS in Computer Science and Applied
Statistics.
Chaitanya Baru is Distinguished Scientist, San Diego
Supercomputer Center, UC San Diego. From 2014 to Ibrahim Mohedas is a program director in the NSF
2018, he was Senior Advisor for Data Science at the Convergence Accelerator. Ibrahim earned his PhD in
National Science Foundation where he provided lead- Mechanical Engineering from the University of Michi-
ership for data programs including BIGDATA, Big Data gan focusing on the use of design ethnography within
Hubs, TRIPODS, Data Science Corps. From 2019 to the engineering design process. He currently leads
2021, he was Senior Advisor for the Convergence Accel- development and execution of the Convergence Accel-
erator and a member of the team that established the erator Innovation Curriculum.
program.
Linda K. Molnar is a Program Director for the
Lara Campbell is a Program Director for the NSF Con- NSF Convergence Acceleratort. She is a scientist and
vergence Accelerator which she helped launch in 2019. entrepreneur specializing in the translation of funda-
Lara’s expertise is in managing large, collaborative, and mental science into products and businesses. Linda has
often international research efforts including those in held positions in the federal government, large corpo-
NSF’s Office of International Science and Engineering. rations, start-ups, and venture capital. She has a BS in
She was also Director of the nonprofit CUBRC Cen- Chemistry and a PhD in Materials Science and Engi-
ter for International Science and Technology Advance- neering.
ment. She holds a PhD in chemistry.
 23719621, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12032, Wiley Online Library on [24/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16 AI MAGAZINE

Michael Pozmantier is a Program Director in the NSF toward commercialization. She has also held a wide
Convergence Accelerator. He has worked in the private range of creative roles in the hospitality industry
sector and government, including leading the Tran- including event management, operations, communica-
sition to Practice Program, focused on the commer- tions, and brand management.
cialization of federally funded cybersecurity research,
at the Department of Homeland Security, Science and Nicole Tehrani provides administrative support to the
Technology Directorate. Convergence Accelerator Office. Nicole is an MBA can-
didate at Baylor University. She has over 10 years of
Michael Reksulak is a faculty member at George combined experience working in research, academic,
Mason University Law School, currently on assignment and medical settings. Nicole holds a Bachelor of Science
with the NSF. He previously served as the Director in Neuroscience and Chemistry.
for Social Sciences, Arts and Humanities at the Qatar
National Research Fund and as Director of Research
Policy for all of Qatar Foundation R&D. He holds a J.D.
degree, a PhD and an MA in Economics. How to cite this article: Baru, C., Campbell, L.,
Dade, A., Fulay, P., Loewi, A., Maughan, D.,
Shelby Smith leads the communications and outreach Mohedas, I., Molnar, L., Pozmantier, M., Reksulak,
for the NSF Convergence Accelerator. Prior to NSF, M., Smith, S., and Tehrani, N. 2022. “The NSF
Ms. Smith advised government agencies in various out- Convergence Accelerator Program.” AI Magazine
reach campaigns, development of public-private part- 43: 6–16. https://doi.org/10.1002/aaai.12032
nerships, and marketing methods to assist research