Ontology Based Information Centric Tactical Edge Networking

Ontology Based Information CentricTactical Edge NetworkingThomas A. WamboldChristopher T. Cannon†Joseph B. Kopena†Duc NguyenMarcello BalducciniEmily C. LeBlancWilliam C. Reglitjkopena@bellerophonmobile.comFebruary 12, 2015

IntroductionBellerophon Mobile:Research and development in computer networking andinformation management for mobile & extreme environmentsStartup out of the Applied Informatics Group at Drexel UExtensive history with NSF and DoD applied researchIOver 10 years’ experience in networking and AIprojects for DARPA, CERDEC, ONR, NRL, DISAFour core competency areasInformation Management: Knowledge representation, automatedreasoning, and the Semantic Web for content curation & discoveryMobile Networking: Ad hoc, content-based, and delaytolerant protocols at network, transport, and application layersNetwork Testing and Evalution: Experimentation on mobilenetworks in emulation, simulation, and live testingMobile App Development: Android and HTML5 apps, includingconsumer facing products for Fortune 500 companies2/14

CBMEN THORReporting general experiences from just-concluded DARPA Content-BasedMobile Edge Networking (CBMEN) projectDrexel & Bellerophon responsible for content naming, search, discovery:THOR: Tactical Heterogeneous Ontology Representation3/14

Tactical Edge NetworksBattlefield networks are already prevalent and yet severely limitedSemi-autonomous groups of mobile users, constrained backbone linksSevere latency, disconnection, disruption are intrinsic domain featuresWide variety of devices, networks, apps, data, policiesLots of digital data already being generatedMostly exchanged only between missions and to/from command4/14

Changing ChallengesHistorically the dominant issues were impoverished nodes and linksReal issues right now are arguably disconnectivity and network scaleNodes have significant memory and processing resourcesNetwork shape is complex, can have substantial link bandwidthIHigh capacity at command & in edge groups, limited capacity betweenLink disruption and network partitioning are pervasive thoughScaling from handfuls of nodes to units to battlefield is a challenge5/14

Upcoming ProblemsMaturing and improving networks will raise challenges new to this domainNear-future critical issue is information managementBetter systems will result in network and users deluged with dataIIHave to make good decisions about routing and presenting contentHave to discover relevant content, not just return specific search results6/14

Information Centric NetworkingCBMEN addresses disconnects and limitedbandwidth via information centric networkingContent itself is the primary networkaddressable elementIApplicationFrontendAutogenNot node endpoints, e.g., servers/clientsEssentially every node acts as a cache andmay provide requested contentTHOR enhances this and prepares forinformation management challenge byaddressing and finding content via expresive,ontology-based metadataApplies Semantic Web technologies totactical edge networkingCBMEN curity7/14

Forwarding and CachingExpressive pub/sub & universal caching enable efficiency, robustnessAn application produces a message reporting observed activityLocal peer node has a matching, long-lived interest and receives reportRemote node also has interest, but reachback links constrainedIIDetailed matching of content eliminates unnecessary transfers“All SpotReport messages of activity A in region (L,T)–(R,B).”Later ad hoc query resolved with cached content, obviating transfers8/14

Forwarding and CachingExpressive pub/sub & universal caching enable efficiency, robustness?An application produces a message reporting observed activityLocal peer node has a matching, long-lived interest and receives reportRemote node also has interest, but reachback links constrainedIIDetailed matching of content eliminates unnecessary transfers“All SpotReport messages of activity A in region (L,T)–(R,B).”Later ad hoc query resolved with cached content, obviating transfers8/14

Forwarding and CachingExpressive pub/sub & universal caching enable efficiency, robustness?An application produces a message reporting observed activityLocal peer node has a matching, long-lived interest and receives reportRemote node also has interest, but reachback links constrainedIIDetailed matching of content eliminates unnecessary transfers“All SpotReport messages of activity A in region (L,T)–(R,B).”Later ad hoc query resolved with cached content, obviating transfers8/14

Forwarding and CachingExpressive pub/sub & universal caching enable efficiency, robustness?An application produces a message reporting observed activityLocal peer node has a matching, long-lived interest and receives reportRemote node also has interest, but reachback links constrainedIIDetailed matching of content eliminates unnecessary transfers“All SpotReport messages of activity A in region (L,T)–(R,B).”Later ad hoc query resolved with cached content, obviating transfers8/14

Metadata LanguagesCBMEN content is described in RDFGenerated by applications directly, orthrough shared Autogen user interfaceGenerally one piece of metadata perpiece of content, but it’s a truegeneric knowledge base and structurescan be built connecting content orcompletely independent dataSearches and proactive subscriptions arespecified as SPARQL queriesOWL-Lite ontologies define a few standardstructures and substantial taxonomies forthe domain that are applied in matching urn:registrar:mc#c5cb82.a2aa a messages:SpotReport;messages:contentType [ a provenance:ImageEntity ];messages:contentFormat [ a messages:JPG ];c2:latitude "38.165048" ;c2:longitude "-77.284355" ;rdfs:comment "From Olivia" .SELECT ?id ?format ?commentWHERE {?id a messages:SpotReport;messages:contentType ?ct .?ct a provenance:ImageEntity .OPTIONAL { ?id messages:contentFormat ?cf .?cf a ?format .MINUS { ?cf a ?cfsub .?cfsub rdfs:subClassOf ?format .}}OPTIONAL { ?id rdfs:comment ?comment . }}9/14

Metadata InferenceSemantic Web metadata supports both specificity and generalizationRDF metadata enables complex modelingof content descriptions and queriesGeoreferencing, units, roles,content summarization tags, etc.BrigadeinstanceOfis-asubEchelonOfis-aUnit4th BDEBattalioninstanceOfis-a1-44 BNsubEchelonOfCompanyinstanceOfsubEchelonOfA CoImpractical with flat or hierarchical labelingEng CoinstanceOfOWL-Lite enables just enough inferenceApply implicit knowledge from background ontologies (echelons, etc.)Derive implications and connections from context (task org, roles, etc.)ImageryVideoStillInfraredSubclass relationships in notionalhierarchy of imagery producer orconsumer typesA specific publication will be deliveredto a more general subscriberA specific request will not be matchedto a more general search10/14

MarkupCurrent demo apps generate metadata using THOR’s Autogen componentAutogen: An Android activity any app can invoke to construct RDFCaller specifies a base class and/or partially populated structureInterface queries ontologies for subclasses and fields, populating theinterface and presenting autocomplete options as user typesLiteral data fields may invoke other activities for inputIContacts for person, GPS or map for coordinate, calendar for date, etc.11/14

OntologyThe THOR ontology constructed by distilling doctrine and canonical appsRefinementMessage al ConceptAnalysisConcept ualDeconstructionDomainConceptsEvaluationFormal ConceptAnalysisConcept LatticeAcceptanceRefinementSeveral core structures and extensive taxonomies, organized into modulesStructures: Messages, force organization, missions, provenance, etc.Taxonomies: Nations, unit types, equipment, observations, tasks, etc.End product is 3700 classes, 100 properties12/14

ReasoningSearch and subscriptions are executed on all nodes by MasterchiefMobile-ready knowledge base built for CBMENSemantic Web logic implemented over SQLite for robustness;portability; and limited, constant-size primary memory utilizationTransformations, management, and interface in C & Go for portability13/14

LessonsA little semantics goes a long wayPotential stakeholders primarily interested in basic taxonomiesFairly difficult to get developers without KR experience up to speedProject apps didn’t get to point of utilizing capabilities forcollaboration, versioning, etc., offered by the underlying modelEvaluation of KR systems is extremely difficultPerformance is non-trivial but fairly straightforwardISidenote: What’s hard for network may not be hard for KR, & vice versaBut testing actual effectiveness and value requires complex yet realisticscenarios, revolves around metrics that are difficult to quantifySPARQL and RDF model aren’t quite the right tools for this taskSPARQL great for querying the KB, less ideal for fetching objectsIApps want all the metadata about content, resulting in massive queriesRDF SPARQL cumbersome when working with dynamic dataIE.g., “All reports within 3 miles of my current position.”14/14

Information Centric Networking CBMEN addresses disconnects and limited bandwidth via information centric networking Content itself is the primary network addressable element I Not node endpoints, e.g., servers/clients Essentially every node acts as a cache and may provide requested content THOR enhances this and prepares for