What can the Semantic Grid do for Science and …

      What can the Semantic Grid do for Science and
                     Engineering?
                                       Jim Myers
                     Associate Director, Collaborative Cyberservices
                                          NCSA

Extended Abstract:

Scientists and Engineers have been happily performing research and Analyses for
hundreds of years without the Semantic Grid. What's changing in their world now that
would motivate them to look to the Semantic Grid? Which of their problems can it solve?
And how can we recognize the low-hanging fruit ­ the combinations of communities and
issues where introducing the Semantic grid now will create the most scientific value?

With advancing technologies, scientific productivity, roughly the number of experiments
per researcher, has been climbing rapidly. Instruments have become faster, more
accurate, more manageable, more automated, and have greater signal-to-noise. One result
is that traditional science can be done faster and more broadly (experiments on a broader
range of samples), leading to information overload. Combined with the increased ability
to analyze and share data, another result is researchers can perform new types of analyses
­ looking for higher dimensional relationships through data mining, feature detection,
correlation methods, clustering, etc. Thus, researchers are building community data stores
and analyzing colleagues' data. It has also become possible to work across communities ­
to do systems-science and ab-initio engineering. The issues involved can be understood
by `Considering a spherical cow'. A spherical cow model is adequate for example, for
studying cow metabolism. Expanding the model for use in studying cow locomotion
would add unnecessary complexity for studying cow metabolism and could lead to
requirements to take measurements (i.e. cow height) to be able to share data that would
not otherwise be required for the experiment being performed. Should we reach
consensus here? Or should researchers keep their current models and be able to integrate
data via assumptions (.e.g. assuming a standard cow shape and inferring height from the
mass in the spherical model)? What if a researcher in, for example, quantum mechanics,
is involved in communities modeling combustion, materials science, atmospheric science,
biology, etc. ­ reaching consensus suddenly involves all of science. Clearly a way to
describe models and the mappings between them is needed.

If one then looks at semantic technologies widely ­ i.e. at not only the stack of RDF,
OWL, etc., but at topic-based messaging, aspect-oriented computing, translation as a first
class operation, agents, etc. ­ the power of semantic technologies in science becomes
clear. It's useful to first ask `what can semantic technologies do in general' ­ make
information computer readable? Enable collaboration through standard models? Provide
decoupling between models and between models and software? Or standardization of
model representation allowing model manipulation? In fact, on the last two are unique to
semantic technologies ­ binary files are computer readable and can function as a shared
standard. Semantic technologies make it easier to discuss models separately from
implementation and allow the development of software that can visualize and analyze
multiple models.

A common syntax for describing semantics is clearly useful ­ analogous to the use of
common ways to describe mathematical equations. And it can be seen as decoupling ­
decoupling human discussion of semantics from the syntax used to describe them.
Beyond this, decoupling can be important in science and engineering in a number of
(arguably related) areas. Data described by models can be transformed more easily using
generic translator/parser software, providing `data virtualization' (beyond, for example,
data location virtualization). Similarly, one can use ontologies to simplify mapping event
messages, enabling easier assembly of modules into workflows or applications into
problem solving environments. At the highest level, one can describe the scientific
models being instantiated in the data and applications and consider mapping between
models of different phenomena to enable systems science/ab-initio engineering.

To put this in some context, consider the functionality related to `scientific logistics' ­
project organization, experiment planning, data organization, note-taking, process
logging, etc. These applications have shares concepts such as resource names and
hierarchical structures, but most also have more specialized or unique concepts ­ from
owners and access policies to notebook pages, quality/trust (e.g. in curation), etc. Since it
is important for coordination of this functionality ­ authoring notes about logged
processes used to make quality judgments for curation, it is tempting to consider an
object model ­ develop a `resource' object that has all the methods and attributes
required. However, an aspect/semantic model makes more sense, i.e. giving control of the
definition of attributes and methods to the developers on individual tools and using a
metadata (triple) store to capture the tools outputs in a common information space (and
then allowing, for example, a user to instruct a notebook to interpret the
project/experiment hierarchy created by a project planning tool as a notebook/chapter
hierarchy to synchronize these tools). The Architecture breakout group at the 2nd
International Data Provenance and Annotation Workshop (see the Background link from
the Semantic Data Grid project site given below) described a number of high-level
aspects of data and attempted to start mapping them to lower level shared
concepts/services. Although the details are outside the scope of this discussion, these are
some of the ideas we've been exploring in our Scientific Annotation Middleware and
Collaboratory for Multiscale Chemical Science, and Semantic Data Grid projects
(http://www.scidac.org/SAM/, http://cmcs.org/, http://collaboratory.pnl.gov/sdg/ ).

With this background, one can start to map the strengths of semantic grid technologies to
the issues facing science and engineering researchers through use cases. A common
syntax for describing semantics is useful in standardizing science and engineering
reference data ­ such as heat capacities for different substances ­ as well as the outputs of
standard analytical techniques such as mass spectroscopy. Data virtualization helps in
assembling/integrating reference data in `legacy' systems. Similarly, research
communities seeking to standardize experiment protocols (e.g. data analysis pipelines)
can benefit from communication virtualization, allowing different programs to be
assembled into workflows. In some cases, communities may be seeking to standardize
specific aspects of their work, i.e. ways of annotating data, and application virtualization
can help with the development of tools that can easily be integrated into environments.
Finally, researchers working across disciplines or across scales ­ systems science,
applications-driven research, consequence-based or ab-initio engineering, etc ­ can use
model-level virtualization to map between disciplinary world views. Note that for all of
these cases, although the semantic web aspect is highlighted in the description, the grid
concepts of distributed, independent resources and virtual-organization-level agreements
about appropriate policies and shared understandings are also critical.

A number of recent reports highlight these trends in science and information technology
requirements that match the semantic grid. In the US, the NSF CyberInfrastructure1 and
Long-Lived Data Collections2 reports, the DOE Data Management3 and National
Collaboratories4 reports (and presumably reports from upcoming workshops to define the
successor of the Scientific Discovery through Advanced Computing (SciDAC) program)
identify issues such as data heterogeneity, the growing need to share and integrate data,
the management of large, complex, dynamic analysis processes and coping with the
output of high-throughput techniques. In all of these areas, standardized syntax for
semantics, the decoupling inherent in the aspect/metadata approach, the scalability of the
web/web service approach, and the ability to scope resources, properties, and agreements
in virtual organizations can all simplify the development ­ and evolution ­ of solutions.

While the semantic grid may not be the only way to make progress on these issues, it
provides a very powerful model that addresses issues that are often under-appreciated.
Anyone who has served on a standards committee(s) knows why the joke about `the good
thing about standards is that there are so many of them' will continue to evoke wry
smiles. A quote from Alice In Wonderland gets to the heart of the matter:

       "When I use a word," Humpty Dumpty said, in rather a scornful tone, "it
       means just what I choose it to mean--nether more nor less."
       "The question is," said Alice, "whether you can make words mean so
       many different things."
       "The question is," said Humpty Dumpty, "which is to be the master--that's
       all."

Heterogeneity of understanding is central to creativity, and semantic grid technologies
promise to allow researchers (i.e.. science and engineering researchers, and computer
scientists) to remain masters of their domains while collaborating and coordinating at an
unprecedented scale.
1
  http://www.nsf.gov/cise/sci/reports/toc.jsp
2
  http://www.nsf.gov/nsb/meetings/2005/LLDDC_draftreport.pdf
3
  http://www.sc.doe.gov/ascr/Final-report-v26.pdf
4
  http://dsd.lbl.gov/Collaboratories/NCWorkshop/agenda.htm