Download presentation
Presentation is loading. Please wait.
1
©2004, Philippe Cudré-Mauroux Sharing Pictures in Peer-DBMS MSRA, Image Retrieval Meeting Philippe Cudré-Mauroux Distributed Information Systems Laboratory (LSIR) Swiss Federal Institute of Technology, Lausanne (EPFL)
2
©2004, Philippe Cudré-Mauroux Outline I. My Background –Multimedia Servers –P2P Systems –Decentralized Semantics A drift from How to receive the data to How to receive the right data II. Sharing Pictures in a PDMS –Problem Definition –Sketch of a possible solution –Advantages of the solution
3
©2004, Philippe Cudré-Mauroux Multimedia Servers Media Delivery Architectures –Adaptive MPEG2 video system HP group, EPFL Keywords: Adaptive coding, RTP/RTCP, DirectShow, MMX –Composable Services for Multimedia servers IBM T.J. Watson Keywords: Composability, Access Control, XML technologies
4
©2004, Philippe Cudré-Mauroux P2P Architecures It was recognized in the 60's that only a decentralized and self-organizing infrastructure results in robust and scalable networks Today, the majority of multimedia content is being served by decentralized infrastructures –Akamai –Gnutella –Kazaa –BitTorrent –… Skype?
5
©2004, Philippe Cudré-Mauroux P-Grid (1) 00? 0?? 01? 1?? 10?11? 1 6 2 34 5 1 : 3 01 : 2 Stores data with key prefix 00 1 : 5 01 : 2 Stores data with key prefix 00 1 : 4 00 : 6 Stores data with key prefix 01 0 : 2 11 : 5 Stores data with key prefix 10 0 : 6 11 : 5 Stores data with key prefix 10 0 : 6 10 : 4 Stores data with key prefix 11 query(6, 100) query(5, 100) query(4, 100) 4 ??? Virtual binary shared tree
6
©2004, Philippe Cudré-Mauroux P-Grid (2) Original paper: P-Grid: A Self-Organizing Access Structure for P2P Information Systems Karl Aberer, Sixth International Conf. on Cooperative Information Systems (CoopIS 2001) Scalable solution: O(log N) even for heavily unbalanced trees Efficient Search in Unbalanced, Randomized Peer-To-Peer Search Trees Karl Aberer, EPFL Technical Report IC/2002/79 Self-organizing infrastructure P-Grid: A Self-organizing Structured P2P System Karl Aberer, Philippe Cudré-Mauroux, Anwitaman Datta, Zoran Despotovic, Manfred Hauswirth, Magdalena Punceva, Roman Schmidt SIGMOD Record, 32(2), September 2003. Media distribution in P-Grid A Decentralized Architecture for Adaptive Media Dissemination Philippe Cudré-Mauroux, Karl Aberer ICME2002 PIX-Grid: A Platform for P2P Photo Exchange Karl Aberer, Philippe Cudré-Mauroux, Anwitaman Datta, Manfred Hauswirth UMICS 2003
7
©2004, Philippe Cudré-Mauroux Shared Representation Isn't Enough Example: Searching Biological DBs –use ASCII (search text content) Searching for data on "anglerfish" –Easy –Results will be precise Searching for "leech" –Organism leech –Authors: "Bleech", "Leechman", … –Protein sequences: …MNTSLEECHMPKGD… Searching for "257" …
8
©2004, Philippe Cudré-Mauroux Shared Context Schemas provide context XML shared representation providing context for data But also context needs to be shared CBPH_LOPAM Roth Lophius americanus (American goosefish) (Anglerfish). MKQICSIVLL …
9
©2004, Philippe Cudré-Mauroux The Problem (1) Swissprot site at Geneva A lab at MIT A lab in Trondheim organism Query posted at EPFL species EMBLChange site at Cambridge organism EMBLChange peers species, … SwissProt peers authors, titles, organism, … other peers authors, … Not a toy-problem!
10
©2004, Philippe Cudré-Mauroux The Problem (2) How to obtain semantic interoperability among heterogeneous data sources without relying on pre-existing, global semantic models? Standards solutions (LAV / GAV) –static environments –views from/to global schemas
11
©2004, Philippe Cudré-Mauroux Emergent Semantics Francis Heylighen characterizing self-organization: “The basic mechanism underlying self-organization is the noise-driven variation which explores different regions in a system’s state space until it enters an attractor.” In our case: –States : all local communication states reached in consensus building. –Attractor : consistency of local agreements –The noise-driven variations : randomness of interactions and autonomy Semantics as agreement! Emergent Semantics Principles and Issues Karl Aberer, Philippe Cudré-Mauroux and Aris M. Ouksel (editors) Tiziana Catarci Mohand-Said Hacid, Arantza Illarramendi, Vipul Kashyap, Massimo Mecella, Eduardo Mena, Erich J. Neuhold, Olga De Troyer, Thomas Risse, Monica Scannapieco, Fèlix Saltor, Luca de Santis, Stefano Spaccapietra, Steffen Staab and Rudi Studer DASFAA 2004 Emergent Semantics Systems Karl Aberer, Tiziana Catarci, Philippe Cudré-Mauroux, Tharam Dillon, Stephan Grimm, Mohand-Said Hacid, Arantza Illarramendi, Mustafa Jarrar, Vipul Kashyap, Massimo Mecella, Eduardo Mena, Erich J. Neuhold, Aris M. Ouksel, Thomas Risse, Monica Scannapieco, Fèlix Saltor, Luca de Santis, Stefano Spaccapietra, Steffen Staab, Rudi Studer and Olga De Troyer ICSNW 2004
12
©2004, Philippe Cudré-Mauroux Outline of the solution A lab in Trondheim species EMBLChange site at Cambridge Swissprot site at Geneva A lab at MIT organism Query posted at EPFL organism EMBLChange peers species, … SwissProt peers authors, titles, organism, … other peers authors, … organism authors organism species species organism Local translations enabling global agreements
13
©2004, Philippe Cudré-Mauroux On Translations
14
©2004, Philippe Cudré-Mauroux Query Forwarding To whom shall we send the queries? –To peers susceptible of sending us a response… Simplistic solutions –Local Neighboring (same schema) Low recall –Query Flooding (entire network) Low precision, high network load Semantic Gossiping –Query forwarding by selecting the right peers –Query dependant PHBs (Per-Hop Behaviors) –Query / transformed queries analysis Intrinsic measures (syntactic distances) Extrinsic measures (semantic distances)
15
©2004, Philippe Cudré-Mauroux Similarity Measures Syntactic Similarity –Similarity measure between an original and a transformed query. –Iterative computation of information loss in selections / projections. Semantic Similarities –Probabilistic analysis (max. likelihood) upon the correctness of translations based on feedback received
16
©2004, Philippe Cudré-Mauroux References Semantic Gossiping A Framework for Semantic Gossiping Karl Aberer, Philippe Cudré-Mauroux, Manfred Hauswirth SIGMOD Record, 31(4), December 2002 Similarity Measures The Chatty Web: Emergent Semantics Through Gossiping Karl Aberer, Philippe Cudré-Mauroux, Manfred Hauswirth WWW2003 Self-Healing Semantic Network Start making sense: The Chatty Web approach for global semantic agreements Karl Aberer, Philippe Cudré-Mauroux, Manfred Hauswirth, Journal of Web Semantics, 1 (1), December 2003 Analyzing Semantic Interoperability in the Large A Necessary Condition for Semantic Interoperability in the Large Philippe Cudré-Mauroux, Karl Aberer ODBASE 2004
17
©2004, Philippe Cudré-Mauroux Context: The Semantic Web Providing machine-processable data to the Web
18
©2004, Philippe Cudré-Mauroux RDF/RDFS Overview RDF triple: RDF Schemas –Classes of resources –Classes of properties –Constraints on the subject (domain) or object (range) –Subclassing Extensible! Subject Object Property
19
©2004, Philippe Cudré-Mauroux Mapping Metadata into P-Grid 00?01?10?11? 0??1?? ??? 000010 100011 SELECT ?x WHERE (?x,, ?z) AND ! (?z eq ) <rdf:Description rdf:about="P- Grid://01100100011" John Doe User-defined meta-data (RDF triples) <rdfs:subPropertyOf rdf:resource=pgrid:PGridDataItemProperty/> User-defined categories (RDFS) <owl:equivalentProperty rdf:ID="mapping1“ rdf:resource=class2:créateur /> Category translations (OWL) RDQL queries
20
©2004, Philippe Cudré-Mauroux GridVine: a Peer-DBMS
21
©2004, Philippe Cudré-Mauroux GridVine Properties Principle of data independence –Physical vs. logical layer Physical layer –Scalable, robust, self- organizing indexing Logical layer –Fosters global semantic interoperability Semantic Gossiping Schema inheritance mechanisms Ref.: GridVine: Building Internet-Scale Semantic Overlay Networks Karl Aberer, Philippe Cudré-Mauroux, Manfred Hauswirth and Tim van Pelt ISWC04
22
©2004, Philippe Cudré-Mauroux Peer-DBMS Peer DataBase Management Systems –Individual peers contribute info. to a global system –Local semantics –Global vs. local views Is this a realistic setting? –"The good thing about standards is that there are so many to choose from" (A. Tanenbaum) –Extensible frameworks Semantic Web CreativeCommons Adobe XMP Microsoft WinFS
23
©2004, Philippe Cudré-Mauroux Adobe XMP Subset of RDF/S Already supported by –Adobe® Acrobat® –Adobe FrameMaker® –Adobe GoLive® –Adobe Illustrator® –Adobe InCopy® –Adobe InDesign® –Adobe LiveMotion™ –Adobe Photoshop® –Adobe Document Server –Adobe Graphics Server –Version Cue™
24
©2004, Philippe Cudré-Mauroux Example: Photoshop Schema in XMP
25
©2004, Philippe Cudré-Mauroux WinFS New file-system for Longhorn (NTFS +++ ) No more hierarchies (i.e., folders) but metadata Items – Attributes – Relationships – Schemas – Sub-Schemas (extensions) –Déjà vu?
26
©2004, Philippe Cudré-Mauroux Example: Image schema in WinFS
27
©2004, Philippe Cudré-Mauroux Hypotheses In a couple of years, most images will be annotated –Today: EXIF, Kazaa (non-extensible) –Tomorrow: XMP, SW, WinFS (extensible) Semantics will be (partially) local –Local schemas –Local resources (e.g., contacts – dangling relationships ) #images >> #peers >> #schemas Peer-DBMS scenario Hard problem (semantic interoperability) huge opportunities for image retrieval techniques
28
©2004, Philippe Cudré-Mauroux Emergent Semantics in Image PDBMS Semantic agreement –Local vs. global view on the system –Subset of peers / schemas Optimal precision and recall when submitting a query Emergent semantics: –States : all local communication states reached in consensus building. –Attractor : consistency of agreements, user feedback –Noise-driven variations : user queries
29
©2004, Philippe Cudré-Mauroux Scenario Peers insert image into the system with local, partial semantics –Infer semantic relationships with other schemas –Infer semantics based on low-level features Peers (potentially) annotate other images –Infer semantic relationships with other schemas Peers query the system and give feedback –Infer semantics –Infer semantic relationships Cross validation of inferred statements
30
©2004, Philippe Cudré-Mauroux Sketch of a solution (1) Inference based on mutual information schema, metadata Low-level features Low-level features metadata schema, metadata schema, metadata schema feedback metadata, schemas
31
©2004, Philippe Cudré-Mauroux Sketch of a solution (2) Cross-validation based on graph partitioning Ref.(1-2): Instance-based Schema Matching for Web Databases by Domain-specific Query Probing Jiying Wang, Ji-Rong Wen, Frederick H. Lochovsky, Wei-Ying Ma VLDB 2004
32
©2004, Philippe Cudré-Mauroux Approach Short-term goals –small experiments, Extending a (centralized) IR system ? Playing around with WinFS ? –initial heuristics Long-term goals –sound theoretical framework –full-fledged prototype –evaluations
33
©2004, Philippe Cudré-Mauroux Some Advantages Leverages metadata production by end-users Local communications / computations –Scalability Hopefully, better results than keywords / low- level analyses –Take advantage of context Implicit clustering of peers Images given local semantics Objectivity vs. subjectivity of interpretation becomes a measurable quality
34
©2004, Philippe Cudré-Mauroux Sharing Pictures in Peer-DBMS MSRA, Image Retrieval Meeting Philippe Cudré-Mauroux Distributed Information Systems Laboratory (LSIR) Swiss Federal Institute of Technology, Lausanne (EPFL)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.