`
`By: Charles R. Macedo (Reg. No. 32,781}
`Brian A. Con1ack{Reg. No. 45,343}
`Amster, Rothatein & Ebenstein LLP
`90 Park Avenue
`
`New York, NY 10016
`Telephone: (212) 336-80?-=1
`cmacedo@are1aw_com
`N l-Google- [P R[@areIaw.c om
`
`Gregory Dovei (admitted pro hnc vice)
`Dove] 3.5 Loner, LLP
`201 Santa Monica Bl'«.rd., Suite 600
`Santa Monica, CA 9040]
`
`Telephone: (310) 656-7066
`gr_eg{n:do*_u'elIaw.con1
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`GOOGLE INC.
`
`Petitioner
`
`‘U.
`
`NETWORK-I TECHNOLOGIES
`
`Patent Owner
`
`Cases IPR20I5-003-=13, IPR20I5-00345, IPR20I5-0034?, and IPREIH 5-00343
`
`Patents 3,640,179, 3,205,237, 3,010,983, and 8,656,441
`
`DECLARATION OF DR. GEORGE I-(ARYPIS
`
`NETWORK-I EXHIBIT 2005
`
`Google Inc. v. Network-I Technoiogies, Inc.
`
`IPR20l5-00345
`
`Page ] of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`Table of Contents
`
`Background to thy opinions in this Declaration. ........................................... ..I
`
`A.
`
`B. Assignment.
`
`C- Approach.
`
`D.
`
`Understanding ofthe law.
`
`Ii.
`
`Summary ofthe [PR Patents and asserted
`
`A.
`
`The IPR
`
`1.
`
`2.
`
`3.
`
`4.
`
`‘23? pater1t{Ex. 100] ‘23? IPR} ............................................... ..8
`
`‘988 patent{Ex. 100] "983
`
`‘|‘?9 patent{Ex. 1001 ‘H9 IPR) ............................................. ..ll
`
`‘~44! patent (Ex. 100] ‘-441 IPR} ............................................. ..l3
`
`B.
`
`The asserted
`
`I.
`
`2.
`
`3.
`
`4.
`
`5.
`
`Overview ofGhias—Ex. ltllt} (addressed in the ‘Z37’, ‘93S,
`
`‘W9, and ‘4-4] IPRs].
`
`-16
`
`Overview oflwamut-a—Ex. 1012 [addressed in the ‘Z3? and
`
`‘Q88 lPRs). .............................................................................. .. 19
`
`Overview ofConwe1i~—Ex. I009 (addressed in the ‘ I ‘I9 and
`‘£141
`
`22
`
`Overview of Phi|1vaw—Ex. I014 {addressed in the ‘ 1?‘) and
`544] |PRs as a secondary reference}. ...................................... ..24
`
`Overview ofChen—Ex. 1008 (addressed in the ‘23?‘' [FR as a
`secondary ret'erence}.
`
`III.
`
`IV.
`
`General
`
`General concerns with the IPR Petitions and Dr. Moulin's
`
`Declarations (Eats. 1004 in each
`
`ClaimConstructions............
`
`A.
`
`suh-linear {‘23'.=' patent).
`
`1.
`
`2.
`
`3.
`
`the words used in the construction: “size ofthe dataset" ........4l
`
`‘E3?
`
`Petitioner’s
`
`B.
`
`non-exhaustive search (‘Z31 ‘Q88, ‘N9. and ‘44l patents) ............. ..54
`
`i
`
`Page 2 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`l.
`
`2.
`
`The Board’s preliminary construction of “non-exltaustive
`search” is consistent with the understanding ofone ofordinary
`skill in the art: “a search that locates a match without a
`
`comparison ofall possible matches.”
`
`The Board properly rejected Petitioner’s assertion that a “non-
`exhaustive search" should be construed as “a search that locates
`
`a match without conducting a brute force comparison ofall
`possible matches. and all data within all possible matches.” ...56
`
`C.
`
`D.
`
`neighbor search I identifying a neighbor I neighbor 2’ near neighbor
`(‘23'i', “J88, “ I79, and ‘r-141
`
`approximate nearest neighbor search {*2}? patent}.
`
`1.
`
`2.
`
`VI.
`
`“identifying a close match that is not necessarily the closest
`match"
`
`“sublinear"
`
`A.
`
`‘Z37 Ground l: The instituted claims ofthe ‘Z37 patent are ttot
`anticipated by lwamura.
`
`l.
`
`2.
`
`3.
`
`4.
`
`5.
`
`sub-linear time seat'ch (claims elements lib] and 5(b.2}} ....... .156
`
`approximate nearest neighbor search {claim elements 9{b} and
`
`nonexhaustive search {claim element 25[b)}.
`
`identify a neighbor it near neighbor {claims elements l[b}, 5[b),
`and
`
`sublinear approximate nearest neighbor search (claim element
`
`13.
`
`‘Z3’? Ground 2: The instituted claims ofthe ‘23T patent at'e not
`anticipated by Ghias.
`
`I.
`
`2.
`
`sublinear time search [claim elements lib) and 5(b.2}]. ........I34
`
`approximate nearest neighbor search (claim elements 9{b1I and
`l3I[b.2}] .................................................................................. .. I 55
`
`C.
`
`‘237 Ground 3: The instituted claims ofthe ‘237 patent are not
`obvious over lwamura and
`
`I66
`
`‘it'll.
`
`‘938patent.
`
`A.
`
`‘993 Grottnd l: The instituted claims ofthe ‘988 Patent at'e not
`
`anticipated by Ghias.
`
`[68
`
`ii
`
`Page 3 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and IPR20l5-00348
`
`Declaration of George Karypis
`
`l.
`
`2.
`
`3.
`
`non-exhaustive search [claim elenient l5{b}}.........................169
`
`search identifying a neighbor [claim element I5(b]}I....._......_. I36
`
`determining an action based on the identification [claim
`element l5[c}). ...................................................................... .. [90
`
`‘£3138 Ground 2: The instituted claims ofthe ‘998 patent are not
`obvious over Ghias.....-......-... --
`..
`
`I 93
`
`‘983 Ground 3: The instituted clairns ofthe ‘998 patent are not
`anticipated by Ii.va1nura.l94
`
`B.
`
`C.
`
`I.
`
`2.
`
`non-exhaustive search {claim I5(b)} ..................................... .. I 94
`
`identifying a neighbor {claim
`
`VIII.
`
`‘ IT‘?patent.
`
`A.
`
`‘I TO Ground I: The instituted claims ofthe ‘W9 patent are not
`anticipated by
`
`I.
`
`2.
`
`neighbor search {claims I. I3,
`
`non-exhaustive search [claims I, I3.
`
`B.
`
`‘ I79 Ground 2: The instituted claims ofthe ‘ 179 Patent are not
`
`obvious in view of Ghias and
`
`I.
`
`2.
`
`non-exhaustive search [claims I, I3,
`
`neighbor search (claims I, I3,
`
`IX. "4-4]
`
`A.
`
`‘44I Ground 1: The instituted claims oftlie ‘dill Patent are not
`
`anticipated by Conweil .................................................................... .256
`
`I.
`
`2.
`
`neighbor search {claims I. I3.
`
`non-exhaustive search [claims I, I3.
`
`B.
`
`‘44l Ground 2: The instituted claims of the ‘MI Patent are not
`
`obvious over Ghias and
`
`I.
`
`2.
`
`non-exhaustive search [claims I, I3,
`
`neighbor search (claims I, I3,
`
`iii
`
`Page 4 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345. IPRZGIS-0034?, and lPR2{}l5-00348
`
`Declaration of George Karypis
`
`L George Karypis, declare:
`
`[ am making this Declaration at the request of Patent Owner Network-l
`
`Technologies. Inc. in the Fo1|owing!mcr!’rrrre.s- Reviews ofU.S. Patent Nos.
`
`3,2fl5,237 {‘237 patent}, 3,[llfl,988 ["988 patclttl, 8,640,179 (‘I79 patent], and
`
`3,656,441 {‘4-4| patent} (collectively the “]PR Patents”):
`
`-
`
`-
`
`I
`
`Ir
`
`IPR2o1s-oo34s P23? patent},
`
`[PR2{1|I5-lJt]34? (‘983 patent},
`
`IPRZDIS-00343 If I 79 patent], and
`
`IPRZDIS-00348 (‘A41 patent].
`
`(collectively the “lPRs"], all initiated by petitioner Googie lnc. [“Petitioner").
`
`I.
`
`Background to my opinions in this Declaration.
`
`A.
`
`Expertise.
`
`l.
`
`I am a Professor in the Department of Computer Science and
`
`Engineering at the University of Minnesota.
`
`I hold a Ph.D. in Computer Science
`
`from the University of Minnesota. granted in 19%.
`
`I began my post-graduate
`
`school career as a Research Associate in my current department.
`
`I became an
`
`Assistant Professor in I999. an Associate Professor in 2004. and a Professor in
`
`2009.
`
`I teach courses in Algorithms and Data Structures. Parallel Programming.
`
`and Data Mining. among other subjects.
`
`I
`
`Page 5 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`2.
`
`I am a inember ofthe Editorial Boarcl ofa number ofacaclemic
`
`joumais, and I have chaired a number ofacademic conferences.’
`
`I am a co-author
`
`of the books iiiti-ridut-rim: to Priro.-'.-’ei if ‘riiiipiitmg, and .-"nrror.~’ric'iioii to .“rrrrih'c.-’
`
`{'riiiipntiiig.' f)e.rign and x€l'.vr,1.-’_v.si.s ofxi .-’gtJriH'mi.s'.
`
`1 am an author of more than 80
`
`published journal papers, and tnoI'e than I 15 published conference papers:
`
`Representative academic conferences include:
`
`I Program Connnittee co-C hair of the AC M Recoininender Systems
`
`Conference (RecSvs‘ 13), Hong Kong, China {Z013};
`
`a Program Committee co-Chair of the l3"‘Inten1ational Conference on Data
`
`Mining [ICD1'v“I), Da|1as,Texas(Decetnber 2013); and
`
`a Program Coinrnittee Co-Chair ofthe international Conference on Data
`
`Science and Advanced Analytics (DSAA 2014), Shanghai, China, {November
`
`2'31-=1}.
`
`l'\—l
`
`Representative papers include:
`
`Ir
`
`"i.2Kniig.'
`
`i"'ti.st .-";'xrir.'t K-Nearest Neighhtir (irrnili f'riii.s'ti'm;'Iitiii will: L2-
`
`Norm .-"i-iniiiig " David C. Anastasia and George Karypis, 24"‘ ACM
`
`International Conference on Infonnation and Kiiowledge Management
`
`tCIKM). Melbourne, Australia (2015).
`
`2
`
`Page 6 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`3.
`
`I have also developed a number of SUfiWfll‘E svstents for a variety of
`
`fiinctions, including software for analyzing high-diinensionai data sets. A copy of
`
`my curriculum vitae is attached as Exhibit A to this Declaration.
`
`it contains a
`
`more complete listing ofmy professional activities and background.
`
`B.
`
`Assignment.
`
`4.
`
`[ have been retained by Patent Owner Network-I Technologies. Inc.
`
`as a technical consultant.
`
`I am being compensated for my time at my standard
`
`consulting rate of $350 per hour.
`
`1 am not receiving any compensation that
`
`depends on the outcome ofthe IPRs.
`
`5.
`
`This declaration addresses the validity of:
`
`I
`
`“i'.2Al’.'
`
`I-it.i'! I ’o.i-me .'~1‘iii1i.-’rm'i_1r .'s'corci'i with l’refi.r 12-2 Norm B.1’Jt‘.1'll‘d.fl‘" David
`
`Anastasiu and George K}11‘}’[JiS, 30"‘ IEEE lntemational Conference on Data
`
`Engineering (ICDE), pp. 784—T95 (ZDI-4].
`
`I
`
`“If 'rJnipr.rri.i'tm of'I)e.sc*r.iprrJr S_nac‘e.v_fi*.i1" ('heiii.r'cau" ( 'rni:r_nr.innt1' Retrieval avid
`
`(‘!o.~r.o)1'ic'oiitiri“ Nil-cii Wale and George Karypis, IEEE International
`
`Conference on Data Mining [ICDML pp. 6'i'8—68‘-) (2006).
`
`I
`
`“t':'mpiricct.-’ rind i'iI'ieov'e!.ic'o1' f 'mnpctri.i'tm.-.' tgfSe1'eci'eta’ {"ri!critJit .-'-i1tnc!i'oii.i'_fi)t'
`
`.’Jo1:.'mnen1'I".-’i.1.-.'1'r:1ri1'1g” Ting Zhao and George Karvpis, Machine Learning, 55,
`
`pp. 311-331 12004}.
`
`3
`
`Page 7 of292
`
`
`
`IPRZDIS-00343, [P112015-00345, IPRZGIS-0034?, and lPR20l5-00348
`
`Declaration of George Karypis
`
`-I
`
`claims 1, 3-5, T-9, 11-13, 15, 16, 21-2?, 29, 30, 33, 34, 35, 37', and 38 of
`
`the ‘Z3? patent;
`
`claims 15—l?, 21-23, 31-33, 51, and S2 ofthe ‘998 patent;
`
`claims 1-3, I5, 8-14, 18, 1'3‘, 21-2?, 29-31, and 34-3? ofthe ‘ H9 patent; and
`
`claims 1-3, 6, 8-14, 13, 1'9, 21-2?, 29, and 30 ofthe ‘£141 patent-
`
`C.
`
`Approach.
`
`6.
`
`To develop my opinions, I have read:
`
`the four [PR Patents [the ‘23'?, ‘938, ‘I39, and ‘44] patents];
`
`the four Petitions for Inter P-:.r.*'.*r»:.s Reviews;
`
`the exhibits accoinpanyirig the Petitions, including the four Declarations of
`
`Dr. Pierre Mou1in{Eas. 1004 in each {PR}:
`
`the four Decisions instituting the IPRs; and
`
`the testirnony oi‘Dr. Pien'e Mouliu, dated August 19-20, 2015 (Ex. zooms}.-‘
`
`Ir
`
`-I
`
`I
`
`-
`
`Ir
`
`I
`
`1-
`
`-
`
`[n this Declaration, I identify the specific Petition, Declaration, and Decision
`
`that [ ant citing by including the corresponding patent abbreviation in a
`
`parenthetical. For example, I refer to the Petition addressing the ‘23? patent as Pet.
`
`(‘E37’) at X; and the Moulin Declaration addressing the ‘ 1 T9 patent as Moulin
`
`Decl. 1," [T9] 11X. Because there is only one Dr. Moulin Deposition transcript for all
`
`4
`
`Page 8 of292
`
`
`
`IPRZDIS-00343. IPRZUIS-00345. [P112015-UU34'r'. and lPR20l5-00348
`
`Declaration c-fGeorge Karypis
`
`T.
`
`In addition, I relied oI1 my personal knowledge and experience with
`
`both research and development in the technology underlying the IPR Patents and
`
`the art asserted against the IPR Patents.
`
`D. Understanding ofthe law.
`
`8.
`
`My understanding regarding the law as applicable to this Declaration
`
`is based on my discussions with counsel.
`
`I have included in the text of my
`
`Declaration quotations from or references to certain legal cases or statutes that
`
`were provided to me by counsel to provide me with an understanding of the
`
`relevant law.
`
`E.
`
`9.
`
`Person of ordinary skill in the art.
`
`Through my education, experience and training, in academia and
`
`industry, and my analysis ofthe [FR Patents, I am familiar with the knowledge ofa
`
`person ofordinary skill in the field of the IPR Patents at the time ofinvention in
`
`2000.
`
`It}.
`
`For the purposes ofthis Declaration, I am ofthe opinion that a person
`
`ofordinary skill in the art with respect to the [PR Patents is a person with a
`
`Bachelor’s degree in computer science, mathematics, or a similar discipline and
`
`two to three years of relevant experience. or a graduate degree in the same area.
`
`four lPRs (Ex. 2006], I simply refer to Dr. Mouliifs deposition testimony as
`
`Moulin Depo. Z.
`
`5
`
`Page 9 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?, and lPR20l5-00348
`
`Declaration of George Karypis
`
`In determining what would be the level ofordinary skill in the field as ofthe 2000
`
`time frame, 1 considered the following:
`
`{a}
`
`the educational level ofthe inventor, [ngemar J. Cox (it is my
`
`understanding that Dr. Cox has a bachelor's degree in electronics and
`
`computer science from University College London {I930} and a PhD.
`
`from Oxford {[9831};
`
`[b]
`
`the type of problems encountered in the art—i.c., how to identify a digital
`
`work without modifying the work (see e.;:., ‘23';', 1:30-36);
`
`{C}
`
`the prior art solutions to those problems {see e.g., ‘23':', 1:3?-4:4, and the
`
`prior art asserted by the Petitioner in the IPRs addressing related problems
`
`involving searching, matching, and identifying melodies, audio files, and
`
`other digital files within clatabases—Conwell, Ghias, Iwamura, Chen, and
`
`Philyaw};
`
`{d} the rapidity with which innovations are made [based on my observations
`
`over the past 2U plus years, major innovations in content identification
`
`occur about every 5 to 10 years];
`
`(e) the sophistication of the technology {developing content identification
`
`solutions is a moderately sophisticated technology); and
`
`6
`
`Page 10 of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`{ii} the educational level of workers in the field (workers in the field generally
`
`had have at least a bachelor’s degree in computer science, mathematics or
`
`a similar discipline and at least two to three years ofrelevant experience].
`
`I 1.
`
`Based on these factors, it is my conclusion that a person ofordinaly
`
`skill in the art at the time would have been a person with a Bache1or’s degree in
`
`computer science, mathematics, or a similar discipline and two to three years of
`
`relevant experience, or a graduate degree in the same or related area.
`
`I2.
`
`I note that Dr. lvloulin suggests that the person of ordinary skill in the
`
`art “would have been highly skilled, and typically would have possessed at least an
`
`MS. in computer science, electricai engineering, or mathematics; knowledge of
`
`video and audio processing techniques; and 1-2 years of experience i11 audio,
`
`video, or image processing." See .-:-.'._r:., Pvioulin Decl. (‘E37’) 117'; Pet. {‘23?'] at 4. Dr.
`
`1‘v’Ioulin‘s opinion as to the person of ordinary skill in the art is similar to mine with
`
`respect to the degrees and years of experience, but I note that: Ifl] Dr. Moulin does
`
`not provide any rational underpinnings for his opinion; and (2) the phrase “highly
`
`skilled” used by Dr. Moulin in his description is a relative term and Dr. lvioulin
`
`does not provide the context for this phrase.
`
`7'
`
`Page ll of 292
`
`
`
`[PFt20l5-00343; lPR20l5-00345; IPR201S-0034?; and lPR20l5-00348
`
`Declaration of George Karypis
`
`II.
`
`Summary of the [PR Patents and asserted art.
`
`l3.
`
`In this Declaration:
`
`an
`
`I use the term “v_vp$" to mean the item (e_;;;., a digital audio or image tile} to
`
`be identified using the search {see e.g., ‘Z31’, 6:5 1-56; ‘988, 'l':l T-20; ‘ I T9,
`
`6:18-21; 544}; 6:-49-52];
`
`-
`
`I use the term " ’ to mean one ofthe units in the reference database
`
`that the extracted features of the work may be compared to {see
`
`‘23?',
`
`6:16-20; ‘E188; 6:46-50; ‘ [T9, 6:21-24; ‘-441; 6: I5-I 3}; and
`
`-
`
`[ use the term ‘ “J,”or to mean the collection of
`
`all records to be sea1'elied(.-we e.g., “23'r', 6:23-30; ‘988, 6:50-60; ‘I79, 6:30-
`
`36; ‘-441, 6:24-30}.
`
`A.
`
`The IPR Patents.
`
`I4.
`
`Each IPR Patent (the ‘23'?, ‘ 1 T0, *938, and ‘£141 patents} involves a
`
`search that compares features from a given work to records in a reference database
`
`of potential matches to identify an action to be taken.
`
`I.
`
`‘Z3? patent (Ex. I001 ‘23? IPR}.
`
`15.
`
`The independent claims ofthe ‘Z37 patent include the following
`
`elements:
`
`[I] receiving or obtaining Features extracted f'rom a work;
`
`8
`
`Page 12 of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`[2] identifyittg the work using the extracted features. to perform a search of
`
`the database, where the search is:
`
`I
`
`II
`
`II
`
`Ir
`
`a sub—]inear time search to identify a neighbor [claims I and 5);
`
`an approximate nearest neighbor search {claims 9 and 13};
`
`a non-exhaustive search
`
`to idetttify a near neighbor [claim 25}; or
`
`a sublinear approximate nearest neighbor search [claim 33}; and
`
`[3] either ti) transmitting infonnation about the identified work to the client
`
`device, or (ii) determining an action based on the identity ofthe work.
`
`16.
`
`The invention claimed in the ‘Z3? patent includes two key features:
`
`1?,
`
`Feature 1: Although the language varies among the claims. each
`
`claim requires that the *‘identifying“ he performed based on a search that has two
`
`properties:
`
`[ I ) a suh-linear or non-exhaustive property [reflected in the underlined
`
`language):
`
`I
`
`I
`
`sub-linear tinte search
`
`to identify a neighbor {ciaims I and 5};
`
`aggroximate nearest neighbor search {ciaims 9 and 13];
`
`I non-exhaustive search
`
`to identify a near neighbor {claim 25}; and
`
`Ir
`
`sublinear approximate nearest neighbor sea1'ch (claim 33).
`
`(2) a neighbor property {reflected in the underlined language}:
`
`9
`
`Page 13 of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`I-
`
`identify a neighbor [claims 1 and 5):
`
`I approximate nearest neighbor search {ciaims 9 and 13}:
`
`It non-exhaustive search
`
`to iIIiEI]tif}" a near neighbor {claim 25]; and
`
`-0
`
`sublineat approximate nearest neighbor search (claim 33).
`
`18.
`
`Feature 2: The system tnust either determine an “action" based on the
`
`identification {claims 25 and 33); or transmit information about the identified
`
`media work to a “client device" {claims 1, 5, 9, and 13).
`
`It is not sufficient to
`
`simply identify a match. Rather, an action must also be identified or information
`
`about the identified work must be transmitted to the client device.
`
`2.
`
`‘988 ]1atent{Ex. H10] ‘988 IPR).
`
`I9.
`
`The independent claims ofthe “J38 patent include the following
`
`elements:
`
`[I] extracting features from a work;
`
`[2] identifying the work based on the extracted features by perfonning “a
`
`non-exhaustive searclt identifying a neighbor;”'
`
`[3] detemtining an action based on the identity oftlte work; and
`
`[4] performing the action.
`
`20.
`
`The invention claimed in the ‘988 patent includes two relevant
`
`distinguishing features:
`
`ID
`
`Page 14 of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`[1] the “ide11tifying" must be peifornied using a “non-exhaustive search
`
`identifying a neighbor;” and
`
`(2) the system must “deterrnin[e] an action” and “perFonn[] the action“
`
`based on the identity ofthe work.
`
`It is not sufficient to identify a match.
`
`Rather, “an action” associated with the match Inust be “detern1in[ed]”
`
`and “perform[ed]." “.988, claim 15.
`
`2].
`
`I note that the Board did not institute trial for independent ciaini I of
`
`the ‘.983 patent and any claims dependent on claim I. Accordingly, I do not
`
`address these claims in this Deciaration.
`
`3.
`
`‘I79 patent (Ex. 100] ‘I79 IPR).
`
`22.
`
`The independent claims ot'the ‘H9 patent [claims I, 13, and 25}
`
`inciude the following five elements for identifying a work and perfonning a
`
`corresponding action:
`
`[I] a database comprising: {a] electronic representations of works; and {I3}
`
`eiectronic data related to an action corresponding to works;
`
`[2] obtaining extracted features of an unknown work;
`
`[3] iCiBfltif}"il1g the unknown work by comparing the extracted features and
`
`electronic representations using a “non-exhaustive neighbor search ;"
`
`1 I
`
`Page 15 of 292
`
`
`
`IPRZUIS-00343, IPRZUIS-00345, IPR20l5-0034?, and lPR2{}15-00348
`
`Declaration of George Karypis
`
`[4] determining an appropriate action based on the electronic data related to
`
`an action; and
`
`[5] associating the determined action with the identified work.
`
`‘ [T9, claims i, I3. and 25.
`
`23.
`
`The ciaimeti steps are illustrated in Figure I:
`
`”5"""~’“E
`E5(TFt.F.t2T|DN
`oPEi=..I.TIoH:5
`
`FEATURE TO
`woatt In
`-,_.,,_-_,G,H,5
`PER,-WW3.
`Di.‘-|'loB-I‘-\5E
`I5-EP£R.fl.I'|DH
`GPERATEPIH
`
`I-hAT|.J'£h
`[\I'ECTDR]t.|I.TRAET1Dfi>l
`UPER-5-TIIUNEEJ
`
`FEATLRE
`If"|"E|iiTC|'H.] LDDi4.U1'
`DPERATIONISJ
`
`I
`
`I
`"‘
`'
`''''I
`rrwrunsret micron:
`
`.
`
`_
`
`J.
`
`'-t.
`
`"'~-
`
`'«"°‘l"-*350'3lfiTED
`|N'FUHH.N.T1DH LCQKUP
`|3"EH‘-"-\T|CiN[5-]
`
`1411
`
`135
`
`|:u.t.I.a-L55
`otneiutton
`Qpffigrggpqgi
`
`""-...__l
`
`-.._‘I.-~-..
`
`mama 11"“ 111
`
`ACTION
`|NiTlM'|DN
`DFERATIDIHSJ
`
`m
`FIGURE 1
`
`Figure 1 iilustrates {“for work @t2"}:
`
`Iv “feature {vector} extraction operation[s}”{14fl] that extract features from the
`
`work {‘ l'i'9, 6145-437}:
`
`12
`
`Page 16 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`I
`
`“feature [vector] lockup operatiouts)" [150] that identify the work by
`
`searching for a matching feature vector {‘ I79, 6:50-52};
`
`Ir
`
`“work-associated infonnatioo lookup operation(s}” (160) that retrievets)
`
`associated infonnation, such as an action {‘ 1 T9, 6:55-58}; and
`
`II
`
`“action initiation operationtsl" IE 170} that performtsl some action based on
`
`the associated information F179, s:s3-so).
`
`24.
`
`The invention claimed in the ‘ITS’! patent includes two relevant
`
`distinguisliing features:
`
`[I] the “identifying,” must be performed by comparing the extracted features
`
`to the electronic representations using a “non-exhaustive neighhor
`
`search:” and
`
`[2] the system must determine or associate an ‘‘action‘‘ based on the
`
`identified work.
`
`It is not sufficient to simply identify a tnatch. Rather,
`
`“an action” associated with the match must be “determined” or
`
`“associated."
`
`‘I79, claims 1, I3, and 25.
`
`4.
`
`‘MI patent (Ex. lllfll ‘44l IPR}.
`
`25.
`
`The independent claims of the ‘-441 patent (claims 1, 13, and 25}
`
`include the following five elements for identifying a work and performing a
`
`13
`
`Page 1? of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, [P112015-0034?, and lPR2015-00348
`
`Declaration of George Karypis
`
`corresponding action:
`
`[1] a database with (a) first data related to records, and {is} second data
`
`related to action information corresponding to the records:
`
`[2] extracting features from a work;
`
`[3] identifying the work by comparing the extracted features and the data
`
`related to the records using “a norkexltaustive neighbor search;"
`
`[4] determining an action based on the identity ofthe electronic work; and
`
`[5] perfonning the action.
`
`‘44] , claims 1, I3, and 25.
`
`26.
`
`The claimed steps are illustrated in Figure 1:
`
`l4
`
`Page 18 of 292
`
`
`
`IPRZUIS-00343, IPRZUIS-00345, IPR20l5-0034?, and lPR2{}l5-D0348
`
`Declaration of George Karypis
`
`FEATURE
`;'..'E|!.'.'TfiI"¥;|E.'|.‘|'|'-'f.-'ul3-"|'l|L'Ir-i
`UFLHAIIDNL-3|
`
`FE-I‘-‘.'|'LIFlE
`I
`I
`°'s:;a¢a":-t
`L
`
`FEATURE
`Eutnamiou
`‘3"E"-“"'¢'-"'55
`
`FEATURE to
`“MK,”
`Tm“
`-oPEFu.tIoh-:5
`
`l.‘nh'I'.i.P.F$E
`EE HERA‘-l1|I;|N
`oasnatuorcsp
`
`'I'I"||}
`IHFDRHATIDH
`
`1»
`Fehruncrsmcctoaz 4 ~I|.|''''‘'1_12
`“- 1.35
`I'.I.il.TAfl‘.fi.'5.E
`
`woar:..-ussncunsn D .;~,g_..EH,.L.,.:.,.
`ll'l:UFl|\l.AT‘.|I'.‘IH LOIDILUF
`upEfi_.|._-hams]
`1.-noaac
`DFEFt.-1TI-:1MI$I
`:.‘l_
`cisoauatnon _
`
`lfivq
`....
`m nssuocuteonuronsutnouug ncnom qu"-- in
`
`-a
`
`action
`INITIATICIN
`l3F'EF|.fi.T||.".|H|51
`
`mg
`
`Figure 1 illustrates (“for work @t2’’}:
`
`1- “feature ( vector} extraction operation[s}" (140) that extract{s} features from
`
`the work (441 , 6:39-41};
`
`Ir “feature {vector} Iookup operationfs)" (150) that identify the work by
`
`searching for a matching feature vector {‘-441, 6:44-48};
`
`- “work-associated infonnation lockup operation[s}" (160) that retrie-acts)
`
`associated information, such as an action (‘-441 , 6:49-51}; and
`
`II
`
`“action initiation operation{s}" ( ITO} that perforrn[s) some action based on
`
`the associated information [‘44l, 6:52-54].
`
`15
`
`Page 19 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345. IPRZGIS-0034?. and lPR2{}l5-00348
`
`Declaration of George Karypis
`
`2?.
`
`The invention claimed in the ‘£141 patent includes two relevant
`
`tiistinguishing features:
`
`(I ) the "identifying" must be performed by comparing the extracted features
`
`to the electronic representations using a “non-exhaustive neighbor
`
`search?’ and
`
`(2) the system must determine or associate an “action" based on the
`
`identified work.
`
`It is not sufficient to simply identify a match. Rather.
`
`“an action" associated with the match must be “determined" or
`
`“associated."
`
`‘£141. claims I, I3. and 25.
`
`B.
`
`The asserted art.
`
`28.
`
`The four lPRs address three primary references and two secondary
`
`reterences.
`
`I address each reference in turn. starting with the pmuary references
`
`and then turning to the secondary references.
`
`I.
`
`Overview of Ghias—Es. ]l]I[l {addressed in the ‘Z37. ‘983,
`‘I79. and "44! {PR5}.
`
`29. Ghias [Patent No. 5,S':'4,o86) discloses “an apparatus [for] searching
`
`melodies.” Gliias, Abstract. As illustrated in Figure l ofGhias, a “tune 12 is
`
`hummed by a person [3 into a microphone 20.” Ghias, 2:4 I -42.
`
`16
`
`Page 20 of 292
`
`
`
`[PR2Di5-00343, [PR2{JI5-00345, [P112015-0034?, and lPR2{}15-00348
`
`Declaration of George Karypis
`
`cQ|,qP|_|TEfi
`
`|’LEI'W_P7A§§' F ‘ ‘ ‘ ‘ ‘-
`|
`¢‘I..IEI!C
`I
`————— ——
`}w.wELE'rI '..F."-_TEE_II SPLINE I
`I
`I
`' TFi:iii:{iiIin ' ————— ——. I""$‘I’i?:'LtuETi
`""" " LSEEJEI twsesi
`
`
`QUERY ENGINE
`
`AUTDDGRRELATION
`BHSED PITCH
`DETECTOR
`
`
`
`
`
`
`RANKED LIST
`GF MATCHING
`H:EI.OC.1|E.5
`
`%:'_'IG.I
`
`The data from the microphone is Fed into “a pitch tracking module 22 in computer
`
`16” which extracts “a contour representation” ofthe melody (23). Ghias, 2:41-50.
`
`The computer uses a “query engine 24” which “searches the melody database I4."
`
`Ghias, 2:50-52. The disclosed search can produce a ranked list ofmatching
`
`meIodies——-“ranked by how well they matched the query" {Ghias, 6:60-63} as
`
`illustrated at 26.
`
`30.
`
`As I explain below in detail, all searches disclosed in Ghias are linear
`
`[not sub-iinear} with respect to the size of the data set being searched.
`
`In
`
`addressing “the problem of approximate string matching," Ghias identifies the
`
`running times ofseveral algorithms:
`
`1'?
`
`Page 2! of292
`
`
`
`[PR2Dl5-00343, [PR2{Jl5-00345, [P112015-0034?, and lPR20l5-00348
`
`Declaration of George Karypis
`
`1-Several Algorithms have been developed that address the
`pniliiern of approximate string matching. Running times
`:5 have ranged [rum Uimfi for the brute force algorithm to
`
`|
`
`ttfltfi orililogirni. where "ill" means "on the orderof." m I
`
`is the number of pitch difiierences in the query, and_ .
`E
`
`Ghias, 6:23-28.
`
`In each identified instance, the running time ofthe search is not
`
`sub-linear with respect to the data set. As clarified in this passage from Ghias (and
`
`as I address in detail below]:
`
`I
`
`“in is the number of pitch differences in the query” corresponding to the
`
`length of the query (highlighted in green in the passage above]; and
`
`-
`
`“n is the size ofthe string {song]" corresponding to the size ofa record being
`
`searched [highlighted in orange in the passage above}.
`
`3 l.
`
`The disclosed searches may be sub-linear with respect to the length of
`
`the query being searched “m
`
`the nttmber ofpiteh differences in the query.“
`
`Speeificaliy, the referenced search with a running time of‘O(niog(n1}) is subiinear
`
`with respect to “of” because it is a function of 1og(rn}), The disclosed searches,
`
`however, are never sub-linear with respect to “n. . the size of the string (song]" or
`
`the size ofthe data set {N} [i'.e., the number ofsongs to be compared}. Rather, the
`
`search time will grow linearly with each additional song to be searched and the
`
`length ofthe song.
`
`32. Also as I describe in detail below, the searches discioseti in Ghias are
`
`exhaustive rather than “ttonexhaustive." The “query engine 24” compares the
`
`18
`
`Page 22 of 292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`work {user input 23] to “all the songs" in the melody database I4 {the library].
`
`Ghias. 5:66-6:2. After searching all possible matches. the system “output[s] a
`
`ranked list ofapproximately matching melodies.” Ghias, 2:50-53.
`
`33,
`
`Finally, as I describe in detail below, the searches disclosed in Ghias
`
`aI'e not “neighbor” searches because the searches always necessarily identify the
`
`exact or closest match—they are guaranteed to identify an exact match or the
`
`closest match. Ghias does not identify any search in which an exact or the closest
`
`match is not guaranteed to be identified.
`
`2.
`
`Overview ol‘Iwan1ura—Ex. ]l]I2 [addressed in the ‘Z3? and
`"9S8 IPRs}.
`
`34.
`
`lwamura [Patent No. fi,l83,Dli}} discloses a “method to enable one to
`
`search for a song title when only its melody is known." Iwamura. Abstract. “A
`
`remote music database with melody information is searched for the melody entered
`
`by the user. using for example. a peak or differential matching algorithm.”
`
`lwamura, Abstract. Figure 1 illustrates “an example of a search interface”
`
`[lwamura. 2:45-46]:
`
`19
`
`Page 23 of 292
`
`
`
`lPR2Dl5-00343. lPR2{Jl5-00345, [P112015-0034?, and lPR2{}l5-00348
`
`Declaration of George Karypis
`
`FIG. 1
`
`:£.|llflI'I|1Il||'.|'l
`
`Ilalnap Search
`
`Trafimsrei Ft. scliumann
`
`35.
`
`Iwatnura discloses a searching algorithm that is designed to be more
`
`efficient than alternatives by matching up peak notes from the work to be identified
`
`with the peak notes of the records in the database when comparing the notes from
`
`the work to be identified with the notes in the records. “Peak notes are also
`
`detected and marked when the data base is built.” lwamura, 6:59-60. “A fast
`
`search is performed by using a peak or differential matching algorithm.“ Iwamura.
`
`I2: 1-2.
`
`36.
`
`As I explain in detail below, the search disclosed in Iwamura is
`
`exhaustive rather than the claimed “non-exhaustit.ie.“ “sublinear,“ or “approximate
`
`nearest neighbor“ search. While the individual comparisons ofa work and a
`
`record in the library can be more efficient using the “peak note“ approach
`
`20
`
`Page 24 of292
`
`
`
`IPRZDIS-00343, IPRZUIS-00345, IPRZGIS-0034?. and lPR20l5-00348
`
`Declaration of George Karypis
`
`disclosed in lwantura [ search speed can be increased]. in doing so each record in
`
`the library is searched as part ofthe disclosed algorithm and “[t]he reference
`
`melody that gives the least difference is returned as a search result.” lwamura.
`
`7:53-55.
`
`37. Moreover, the Boyer-Moore algorithm referenced in Iwamura
`
`searches “word by word from the beginning of the database to the end.” lwamura,
`
`9:5 I-55. As a result. as I explain in detail below. while the Boyer-Moore
`
`algorithm may be sublinear with respect to the length ofthe query {the work to be
`
`ide11tified],4 it is not sub-linear with respect to the relevant size ofthe dataset being
`
`searched.
`
`If the query pre-processing step of the Boyer-Ivloore algorithm is included as
`
`part ofthe execution time. then the algorithm may be linear in terms ofthe length
`
`of the query. If the query is used repeatedly, however, the pre-processing
`
`execution time will only be incurred once. One can think ofconcatenating all the
`
`database strings to give as an aggregate length of n. If“m” is the query length.
`
`then the worse-case complexity is Thetatm) (){n], which is linear with respect to
`
`both the database It and the query length m.
`
`2]
`
`Page 25 of 2.92
`
`
`
`[PR2Dl5-00343, lPR2{J]5-00345, [P112015-0034?, and lPR2{}l5-00348
`
`Declaration of George Karypis
`
`3.
`
`Overview of Conwell—Ex. 1l]l]9 (add ressed in the ‘I79 and
`‘=14! IP'Rs}.
`
`38.
`
`Conwell {Patent No. 6,970,366} discioses associating media content,
`
`such as MP3 files, with identifiers and URLs. Conwell, Abstract. As illustrated in
`
`Figure 3, the identifiers {e.g., "(J3-'-1”} are associated with corresponding URLs (e.g.,
`
`'!'.I-
`“www.songrnusic.co1nfcatalogf05o34.htm] }:
`
`E w
`
`anu.:son1n11u5ic.I:um!:a1ald-gJ'UEllJ14.htrrl
`www.:iupaflracks.oon1Jindexu‘ariis.tsu'ta.1fior.h'lt1I
`
`nnuI.u'.|DUd&'y'B.oom!rap!199'9M$?55fS4E.htm|
`
`vwvw.amu3io.oo1'rIl'Dé§Eié53r:.pdf
`
`www.cdw.oorrutmus|c.tEaaturad_CDsa"-ndax.html
`
`E vMw.sony1nu5ic.contF::ata|ogI{lI}231.hufl
`E wutnnI.pol5rgrarn.oomffranklirtFadf_‘2‘34.hh'n
`
`3‘-E}.
`
`Conwell discloses two approaches to identifying a work: {I}
`
`assigning identifiers, or (2) implicitly generating identifiers derived from the data
`
`using a hashing algorithm. Conwell, Abstract.