`571–272–7822
`
`Paper 6
`Entered: June 23, 2015
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_______________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`_______________
`
`GOOGLE INC.
`Petitioner,
`
`v.
`
`NETWORK-1 TECHNOLOGIES, INC.,
`Patent Owner.
`_______________
`
`Case IPR2015-00345
`Patent 8,205,237 B2
`_______________
`
`
`
`Before KEVIN F. TURNER, LYNNE E. PETTIGREW, and
`JON B. TORNQUIST, Administrative Patent Judges.
`
`TURNER, Administrative Patent Judge.
`
`DECISION
`Institution of Inter Partes Review
`37 C.F.R. § 42.108
`
`
`
`
`
`
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 1 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`I. INTRODUCTION
`Google Inc. (“Petitioner”) filed a Petition (Paper 1, “Pet.”) requesting
`inter partes review of claims 1–27, 29, 30, 32–35, 37, 38, and 40 of U.S.
`Patent No. 8,205,237 B2 (Ex. 1001, “the ’237 Patent”). Network-1
`Technologies, Inc. (“Patent Owner”) filed a Preliminary Response to the
`Petition. Paper 5 (“Prelim. Resp.”). We have jurisdiction under 35 U.S.C.
`§ 314, which provides that an inter partes review may not be instituted
`“unless . . . there is a reasonable likelihood that the petitioner would prevail
`with respect to at least 1 of the claims challenged in the petition.” 35 U.S.C.
`§ 314(a).
`For the reasons set forth below, on this record, Petitioner demonstrates
`a reasonable likelihood of prevailing in showing the unpatentability of
`claims 1–16, 21–27, 29, 30, 33–35, 37, and 38 of the ’237 Patent.
`Accordingly, we institute inter partes review as to claims 1–16, 21–27, 29,
`30, 33–35, 37, and 38 of the ’237 Patent on the grounds specified below.
`
`A.
`Related Proceedings
`The parties inform us that the ’237 Patent is the subject of the
`following lawsuit: Network-1 Technologies, Inc. v. Google Inc. and
`YouTube, LLC, Case No. 1:14-cv-02396 (S.D.N.Y.). Pet. 1. YouTube, LLC
`is a subsidiary of Petitioner, and is acknowledged as a real party-in-interest.
`Id. In addition, three additional patents, U.S. Patent Nos. 8,2640,179,
`8,010,988, and 8,656,441, all issuing from applications related to the ’237
`Patent, are subject to requests for inter partes review, namely IPR2015-
`00343, IPR2015-00347, and IPR2015-00348, respectively, where institution
`of those proceedings are being decided concurrently.
`
`2
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 2 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`B.
`The ’237 Patent
`The ’237 Patent relates to identifying a work, such as a digital audio
`or video file, without the need to modify the work. Ex. 1001, 1:31–36,
`4:25–31. This identification can be accomplished through the extraction of
`features from the work, and comparison of those extracted features with
`records of a database or library. Id. at Abstract. Thereafter, an action may
`be determined based on the identification determined. Id. at 4:24–25. Patent
`Owner refers to Figure 1 as illustrating the steps of the claimed computer-
`implemented methods (Prelim. Resp. 4):
`
`Fig. 1 of the ’237 Patent illustrating the claimed method
`
`
`
`3
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 3 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`4
`
`C.
`Illustrative Claim
`Claim 1 is independent, is considered representative of the claims
`
`challenged, and is reproduced below:
`1. A computer-implemented method comprising:
`a) receiving, by a computer system including at least one
`computer, features that were extracted from a media work by a
`client device;
`b) determining, by the computer system, an identification
`of the media work using the received features extracted from
`the media work to perform a sub-linear time search of extracted
`features of identified media works to identify a neighbor; and
`c) transmitting, by the computer system, information
`about the identified media work to the client device.
`Ex. 1001, 25:2–13.
`
`D.
`Evidence of Record
`Petitioner relies on the following references and Declaration (see
`Pet. 2–3):
`Reference or Declaration
`Sunil Arya, et al., “An Optimal Algorithm for
`Approximate Nearest Neighbor Searching in Fixed
`Dimensions” Journal of the ACM, 45(6), 891–923 (1998)
`(“Arya”)
`Christian Böhm, et al., “Efficient Similarity Search in
`Digital Libraries” IEEE Advances in Digital Libraries,
`193–199 (2000) (“Böhm”)
`U.S. Patent No. 7,444,353 (“Chen”)
`U.S. Patent No. 5,874,686 (“Ghias”)
`U.S. Patent No. 6,597,405 ("Iggulden")
`U.S. Patent No. 6,188,010 ("Iwamura")
`U.S. Patent No. 6,505,160 (“Levy”)
`U.S. Patent No. 7,743,092 (“Wood”)
`Declaration of Dr. Pierre Moulin
`
`
`Ex. 1008
`Ex. 1010
`Ex. 1011
`Ex. 1012
`Ex. 1013
`Ex. 1015
`Ex. 1004
`
`Exhibit No.
`Ex. 1006
`
`Ex. 1007
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 4 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`
`
`E.
`Asserted Grounds of Unpatentability
`Petitioner asserts that the challenged claims are unpatentable on the
`following grounds (see Pet. 2):
`Claims Challenged
`1, 3–5, 7–9, 11–13, 15, 16, 21–25, 29,
`30, 33, 37, and 38
`1–25, 29, 30, 32, 33, 37, 38, and 40
`25, 32, 33, and 40
`1–3, 5–7, 9–11, 13–15, and 21–24
`9–16, 23, and 24
`26, 27, 34, and 35
`26, 27, 34, and 35
`26, 27, 34, and 35
`
`Basis Reference(s)
`§ 102(e)
`Iwamura
`
`Levy and Arya
`§ 103
`Iggulden and Böhm
`§ 103
`§ 102(b) Ghias
`§ 102(e) Wood
`§ 103
`Iwamura
`§ 103
`Iwamura and Chen
`§ 103
`Levy, Arya, and Chen
`
`II. ANALYSIS
`A. Claim Construction
`The claims of an unexpired patent are interpreted using the broadest
`reasonable interpretation in light of the specification of the patent in which
`they appear. See 37 C.F.R. § 42.100(b); In re Cuozzo Speed Techs., LLC,
`778 F.3d 1271, 1281–82 (Fed. Cir. 2015). On this record and for purposes
`of this decision, we determine that only the claim terms addressed below
`require express construction.
`
`1.
`Nonexhaustive
`Independent claim 25 of the ’237 Patent recites that the “determining”
`step occurs through performance of “a nonexhaustive search of reference
`extracted features.” Petitioner argues that the term “nonexhaustive search”
`should be construed to mean “‘a search that locates a match without
`conducting a brute force comparison of all possible matches, and all data
`
`5
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 5 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`within all possible matches.’” Pet. 5. Patent Owner substantially agrees, but
`argues that the last clause, “and all data within all possible matches,” would
`render the claim term to include any search that does not compare “all data”
`in each record, even if the search was a brute force comparison of each
`record in the database. Prelim. Resp. 6. Patent Owner continues that the
`latter portion of Petitioner’s proposed construction is inconsistent with how
`the term “nonexhaustive” is used in the Specification of the ’237 Patent and
`not part of the ordinary meaning. Id. at 7–9. We largely agree with Patent
`Owner.
`On this record, we are persuaded that the broadest, reasonable
`construction, consistent with the Specification, of “nonexhaustive search” is
`“a search that locates a match without a comparison of all possible matches.”
`Patent Owner points out (id. at 6) that the Specification of the ’237 Patent
`discusses both exhaustive and nonexhaustive searches in its section
`Exemplary Techniques for Matching Extracted Features with Database
`Entries (Ex. 1001, 8:27–9:36). Patent Owner also supplies an example it
`contends shows the ordinary meaning of “exhaustive search” or “brute-force
`search.” (Ex. 2001). Neither discussion mentions the evaluation of all data
`within each possible match. By contrast, Petitioner provides no other
`support for its construction other than its declarant, Dr. Pierre Moulin. See
`Ex. 1004 ¶ 43. To the extent that Dr. Moulin testifies that a nonexhaustive
`search “encompassed anything other than a ‘brute force’ search” (id.), we
`agree, but we do not find that his testimony supports the latter portion of
`Petitioner’s construction, i.e., “and all data within all possible matches.”
`
`6
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 6 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`As such, based on the instant record, we are persuaded that a “non-
`exhaustive search” should be construed as “a search that locates a match
`without a comparison of all possible matches.”
`
`2.
`Sub-Linear
`Independent claims 1 and 5 recite “a sub-linear time search,” and
`independent claim 33 recites “a sublinear approximate nearest neighbor
`search.” Petitioner argues that “sub-linear,” or “sublinear,” should be
`construed to mean “a search whose execution time has a sublinear
`relationship to database size,” and provides examples of the same. Pet. 6–7.
`Patent Owner offers a construction of “a search whose execution time scales
`with a less than linear relationship to the size of the data set to be searched,
`assuming computing power is held constant.” Prelim. Resp. 9. We are
`persuaded that these definitions are largely consistent and we construe a
`“sub-linear search” as “a search whose execution time scales with a less than
`linear relationship to the size of the data set to be searched.” One example
`of such a sub-linear search would be a search with an execution time
`proportional to the logarithm of the size of the data set (“N”), where a
`doubling of N would lead to an execution time proportional to log(2N).
`
`3.
`Neighbor / Near Neighbor / Nearest Neighbor
`Each of the independent claims of the ’237 Patent recite one of the
`following: “identify a neighbor,” “identify a near neighbor,” and “nearest
`neighbor search.” Petitioner argues that each term “should be construed to
`mean ‘identify a close, but not necessarily exact, match.’” Pet. 6. Patent
`Owner largely agrees with that construction, arguing, with respect to
`
`7
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 7 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`“neighbor” and “near neighbor,” that those claim terms should be defined as
`“identifying a close, but not necessarily an exact or the closest, match of a
`feature.” Prelim. Resp. 10–11. Patent Owner provides further refinements
`as to what “close” and “feature” mean in the context of the Specification of
`the ’237 Patent (id.), but we are not persuaded that such refinements are
`necessary. Patent Owner argues that “close” should mean a distance or
`difference that falls within a defined threshold, but we are persuaded such an
`alteration would incorporate the element of a threshold into the independent
`claims, where none is specifically required. Although we acknowledge that
`the Specification addresses such a threshold (Ex. 1001, 6:36–40), and we
`may evaluate closeness by such an evaluation, we are not persuaded that
`such an implicit limitation should be applied to the claim construction of
`“neighbor search.” Similarly, Patent Owner urges that “feature” means a
`“feature vector” (Prelim. Resp. 10–11), but we are persuaded also that
`characterizing a “feature” as a “feature vector,” would unduly limit the
`scopes of the independent claims.
`Thus, on this record, we are persuaded that identifying a “neighbor”
`and “near neighbor” should be construed as identifying “a close, but not
`necessarily exact or closest, match.”
`With respect to “nearest neighbor,” however, we agree with Patent
`Owner that the claim term cannot have the same construction because
`different claim terms are presumed to have different meanings. Id. at 11–12.
`The modifiers “near” and “nearest” have related meanings, but those
`meanings are not identical. Patent Owner argues that “nearest neighbor” is
`“the closest, but not necessarily exact match.” Id. We agree and we
`construe “nearest neighbor” as “a closest, but not necessarily exact match.”
`
`8
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 8 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`4.
`Approximate Nearest Neighbor Search
`Patent Owner also seeks a specific construction of the claim term
`“approximate nearest neighbor search,” asserting that Petitioner did not
`provide a specific construction of that term. Prelim. Resp. 12. Patent Owner
`continues that the “Petition should be rejected at the threshold with respect
`to claims 9 and 13 (and claims dependent on these claims) that include[] this
`phrase.” Id. We do not agree, as a petition may rely on the plain, ordinary
`meaning of claim terms, i.e., without specific construction, and still set forth
`how the challenged claim is to be construed. See 37 C.F.R. § 42.104(b)(3).
`Patent Owner continues that the Specification of the ’237 Patent provides
`that the claimed “approximate nearest neighbor search” is “a sub-linear
`neighbor search that does not always find the closest point to the query.”
`Prelim. Resp. 12.
`We largely agree with Patent Owner’s construction, but note that the
`Specification refers to “[o]ne example of a sub-linear time search is an
`approximate nearest neighbor search” (Ex. 1001, 9:12–14), such that we are
`not persuaded that an “approximate nearest neighbor search,” must be a sub-
`linear search, as that term has been construed above. As such, we are
`persuaded that the proper construction of “approximate nearest neighbor
`search” is “identifying a close match that is not necessarily the closest
`match.”
`
`
`B.
`Asserted Grounds of Unpatentability
`1. Alleged Anticipation of Claims 1, 3–5, 7–9, 11–13, 15, 16, 21–25,
`29, 30, 33, 37, and 38 by Iwamura
`Petitioner argues that claims 1, 3–5, 7–9, 11–13, 15, 16, 21–25, 29,
`30, 33, 37, and 38 are anticipated by Iwamura. Pet. 7–16. Iwamura
`
`9
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 9 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`discloses a system for identifying a melody input by a user, whereby the
`system searches a remote music database for the melody. Ex. 1012,
`Abstract. After the search, the database server sends back the results in the
`form of a web page. Id. Iwamura discloses the extraction of features from
`an electronic work, with the melody input through a microphone, converted
`into an electronic signal, and analyzed by a Fast Fourier Transform (“FFT”)
`to obtain frequency spectrum information from the waveform data. Id. at
`4:4–14. The extracted features are used to search the remote music database
`“using a peak or differential matching algorithm” (id. at 12:1–2), although
`other matching algorithms may be applied, including a Boyer-Moore
`algorithm. Id. at 9:57–10:3. The web page that the database server sends
`back can include a link to an on-line music shop, which can enable a user to
`make a purchase, or can include a sound file to be played by the PC client.
`Id. at 12:9–13.
`In addition, Iwamura discloses that “[t]o accelerate the search,
`computation of the total absolute difference can be stopped when it exceeds
`a certain limit.” Id. at 7:56–57. Petitioner identifies this as an example of
`Iwamura’s non-exhaustive searching, in that not all records in the remote
`music database are necessarily searched. Pet. 10. Also, Iwamura provides
`for input fault tolerance so that the “search engine will find the closest
`melody from the database” (Ex. 1012, 9:24–25), where no initial exact result
`is obtained, which Petitioner argues is equivalent to the claimed
`“identify[ing] a neighbor,” “identify[ing] a near neighbor,” and “to perform
`an approximate nearest neighbor search.” Pet. 10–16.
`Patent Owner argues that the search process in Iwamura is exhaustive,
`rather than “nonexhaustive,” “sub-linear,” or “approximate nearest neighbor
`
`10
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 10 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`search,” as recited in the claims. Prelim. Resp. 13. Patent Owner
`acknowledges that the individual comparisons of a work and a record in the
`library can be more efficient using the “peak note” approach disclosed in
`Iwamura, but notes that each record is searched, making the search
`exhaustive. Id. at 13–14.
`We note that all of the independent claims of the ’237 Patent utilize
`“comprising” language, such that those claimed methods and apparatuses do
`not exclude additional, unrecited elements or method steps. See Mars Inc. v.
`H.J. Heinz Co., 377 F.3d 1369, 1376 (Fed. Cir. 2004). Thus, the scope of
`independent claim 25 can include an exhaustive search, as long as it
`performs a nonexhaustive search as well. Thus, even if Patent Owner is
`correct and a particular search in Iwamura is exhaustive, that does not end
`the inquiry. In addition, we note that no claim in the ’237 Patent requires the
`searching, in the determining aspect of the claims, to be both nonexhaustive
`and sub-linear, such that a sub-linear search of the data within the records,
`even if every record is searched, can potentially teach the aspect of
`independent claims 1 and 5 which recite “perform[ing] a sub-linear time
`search of extracted features.” Although the Specification of the ’237 Patent
`discloses that a sub-linear search is performed on the records of the database
`and not information within the records, the claims do not specify that the
`sub-linear search must be performed on a subset of all of the records, and not
`information within individual records.
`Patent Owner also argues that Iwamura’s computational limit does not
`create a nonexhaustive search because “it does not state or suggest that all
`records in the music library are not use[d] in the comparison.” Prelim. Resp.
`18. We do not agree. If, in Iwamura, the computational limit is reached, the
`
`11
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 11 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`search is stopped, even if not all of the records have been searched. Per our
`construction of “nonexhaustive search,” i.e., “a search that locates a match
`without a comparison of all possible matches,” we are persuaded on this
`record that the process of Iwamura, with the computational limit, would
`prevent all of the records of the remote music database from being searched,
`but would ultimately provide a match because of the input fault tolerance
`process, discussed above. See Ex. 1012, 7:56–57, 9:20–34.
`Patent Owner also argues that Iwamura’s use of the “‘Boyer-Moore
`algorithm’ searches all items in the database and therefore is not sublinear.”
`Prelim. Resp. 18–19. As discussed above, we are not persuaded that this is a
`deficiency with respect to the instant claims. With respect to “approximate
`nearest neighbor search,” Patent Owner argues that the input fault tolerance
`capability of Iwamura cannot teach the same because it does not state or
`imply “that all records in the music library are not used in the comparison as
`required in an ‘approximate nearest neighbor search.’” Id. at 19–20. Our
`construction of “approximate nearest neighbor search” to be “identifying a
`close match that is not necessarily the closest match” does not require that
`all of the records in the library are not used, so we also do not find this to be
`a deficiency of the ground.
`We also have reviewed Petitioner’s arguments and claim chart with
`respect to the remaining limitations of the independent claims and the
`dependent claims challenged on this ground. On the present record, we find
`that Petitioner has demonstrated a reasonable likelihood of prevailing in
`showing that claims 1, 3–5, 7–9, 11–13, 15, 16, 21–25, 29, 30, 33, 37, and
`38 are anticipated by Iwamura.
`
`
`12
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 12 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`2. Alleged Obviousness of Claims 1–25, 29, 30, 32, 33, 37, 38, and 40
`over Levy and Arya
`Petitioner argues that claims 1–25, 29, 30, 32, 33, 37, 38, and 40 are
`rendered obvious over Levy and Arya. Pet. 16–30. Levy relates to
`identification of media objects based on identifiers embedded into those
`objects or their containers. Ex. 1013, Abstract. An identifier can be
`extracted from an object and mapped to an action, which can be executed,
`including returning metadata, redirecting a request, or identifying the media
`object. Id. The identifying characteristics of the object may be a unique
`identifier or encoded metadata of the object, and preferably the system may
`use a hash algorithm to create statistically-unique “fingerprints” of the
`content of the object. Id. at 2:29–37, 9:40–61.
`By contrast, Arya is directed to optimal algorithms for approximate
`nearest neighbor searching. Ex. 1006, Abstract. Arya discloses that its
`nearest neighbor searching is non-exhaustive and can provide significant
`improvements over brute-force searching. Id. at 19–21. Petitioner argues
`that one of ordinary skill in the art would need to look for more than the
`statistically unique identifier (i.e., fingerprint of the work) in Levy, due to
`slight discrepancies in the generated identifiers, and would be interested in
`also finding near matches. Pet. 20–21. Petitioner argues that Arya’s
`disclosure of approximate nearest neighbor searching would have allowed
`for such a variation in Levy’s identification processes and would have
`motivated the combination of Levy and Arya. Id.
`We have reviewed Petitioner’s assertions and supporting evidence,
`and Patent Owner’s opposing arguments, assertions and evidence, and we
`are not persuaded that Petitioner demonstrates a reasonable likelihood of
`
`13
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 13 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`prevailing in showing that claims 1–25, 29, 30, 32, 33, 37, 38, and 40 are
`rendered obvious over Levy and Arya.
`Patent Owner argues that the “fingerprints” produced in Levy, based
`on a hash function, would be unique, but that minor changes to the content
`would result in widely varying fingerprints. Prelim. Resp. 30–31. As a
`result, outputs of hash functions that are “close,” i.e., differing by a few bits
`from each other, provide no indication that the underlying contents are close
`or similar. Id. Patent Owner continues that “incorporating the neighbor
`search algorithm of Arya into the exact-match system of Levy will create
`erroneous meaningless results.” Id. at 36. We agree. Finding the
`approximate nearest neighbors of objects, as taught by Arya, listed by their
`“fingerprints” or unique identifiers, per Levy, would not provide a result
`commensurate with the independent claims. The discovered “neighbors”
`would not be related to the obtained identifier, as the unique identifiers in
`the database are indexed only by the hashed extracted features. Therefore,
`based on this record, we agree with Patent Owner that there would be no
`benefit to apply the approximate nearest neighbor searching of Arya to the
`fingerprinting process of Levy.
`Although we acknowledge Petitioner’s argument “that disparities in
`the source of a fingerprinted work led to slight discrepancies in generated
`fingerprints,” and the testimony of Petitioner’s declarant, Dr. Pierre Moulin,
`that ordinarily skilled artisans would want near matches, as well as exact
`fingerprint matches (Pet. 21, citing Ex. 1004 ¶ 86), we are not persuaded that
`this would be sufficient to motivate the combination, given the differences
`discussed above. On this record and at this stage of the proceeding, we are
`persuaded Petitioner has not demonstrated a reasonable likelihood of
`
`14
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 14 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`prevailing in showing that claims 1–25, 29, 30, 32, 33, 37, 38, and 40 are
`rendered obvious over Levy and Arya.
`
`3. Alleged Obviousness of Claims 25, 32, 33, and 40 over Iggulden
`and Böhm
`Petitioner argues that claims 25, 32, 33, and 40 are rendered obvious
`over Iggulden and Böhm. Pet. 30–39. Iggulden relates to real-time
`identification and alteration of a television broadcast signal. Ex. 1011,
`Abstract. The system includes a “selective muting unit” which is connected
`to a television by the television's audio and video output jacks and an
`infrared link, which the selective muting unit can use to transmit a mute
`signal to the television. Id. at 23:8–19. The identification occurs through a
`“signature pattern” of a television segment, such as “a bit string hash code
`representative of whether each of a set of selected lines of a selected frame
`of a received segment has an average luminance level that exceeds a
`predetermined threshold.” Id. at 9:55–67, 18:6–65. The signature pattern is
`compared to stored signature patterns and the video is muted if the identified
`signature pattern is “one of the selected segments.” Id. at 4:54–62.
`Böhm discloses a method for “search[ing] for similar objects” in
`“digital libraries” of complex data, such as “high-resolution images or
`molecular protein structures.” Ex. 1007, Abstract. Petitioner argues that
`Böhm’s searching methods are non-exhaustive and sublinear. Pet. 33–34.
`Petitioner also argues that a person of ordinary skill in the art would have
`been motivated to combine Iggulden with Böhm because the algorithm in
`Böhm “consistently outperforms the . . . most efficient scan-based access
`method for high dimensional data sets [previously] proposed,” and such
`
`15
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 15 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`skilled artisans would have understood that the same principles apply to the
`search of a video database. Pet. 36 (citing Ex. 1007, 199; Ex. 1004 ¶ 105).
`Patent Owner argues that:
`Applying Böhm’s neighbor search (“similarity queries”)
`algorithm to Iggulden’s hashed extracted features would
`identify “k nearest neighbors” of the hashed audio signal.
`Böhm, 7. But that result is not a useful neighboring match to
`the digital audio signal; rather it identifies a neighbor to the
`hashed digital signal, which means it is no closer than a random
`selection from the library.
`Prelim. Resp. 48.
`This argument is analogous to that raised against the combination of
`Levy and Arya, and we find it to be equally as persuasive. The sections of
`Iggulden cited in the Petition make clear that the unique signature may be
`any unique characteristic of the broadcast segment, including a hash code
`representative of that segment, as well as alternative signatures. Pet. 37
`(citing Ex. 1011, Abstract, 9:55–67, 18:6–44). The cited processes all
`provide signature patterns that are related to their television segments, but
`“neighbors” to a specific signature pattern in the database of stored signature
`patterns would not necessarily have any relation to other television segments
`related to the original. Therefore, there would be no reason to identify
`neighbors as taught by Böhm. The Petition, as well as Dr. Moulin’s
`Declaration (Ex. 1004 ¶ 105), provides no rationale that would overcome
`this incongruity.
`On this record and at this stage of the proceeding, we are persuaded
`Petitioner has not demonstrated a reasonable likelihood of prevailing in
`showing that claims 25, 32, 33, and 40 are rendered obvious over Iggulden
`and Böhm.
`
`16
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 16 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`4. Alleged Anticipation of Claims 1–3, 5–7, 9–11, 13–15, and 21–24
`by Ghias
`Petitioner argues that claims 1–3, 5–7, 9–11, 13–15, and 21–24 are
`anticipated by Ghias. Pet. 39–47. Ghias relates to searching for melodies.
`Ex. 1010, Abstract. The system of Ghias receives a melody input through a
`microphone, converts it into a digitized representation based on relative
`pitch differences between successive notes, and searches a database of such
`representations for an approximate match. Id. Ghias also provides that a
`preselected error tolerance may be applied to the search. Id. at 2:50–59.
`The results of the search are presented as a ranked list of approximately
`matching melodies, or alternately just one best match. Id. at 2:50–59, 6:60–
`63. Ghias also discloses that it is desirable to perform key-searching within
`the database using “an efficient approximate pattern matching algorithm,”
`where different algorithms have various running times dependent on the
`number of entries in the database, including search algorithms that are
`substantially faster than “brute force” searches, with execution times that are
`proportional to the logarithm of the size of the data set. Id. at 6:7–11, 6:23–
`35.
`
`Ghias also discloses that “a cubic-spline wavelet transform or other
`suitable wavelet transform” may be used to improve the speed and
`robustness of the pitch-tracking algorithm. Id. at 5:13–20. As well, Ghias
`provides that the melody database “may be located apart from computer 16
`and suitably connected thereto for communicating.” Id. at 6:13–20. We are
`persuaded that Petitioner has demonstrated a reasonable likelihood of
`prevailing in showing that claims 1–3, 5–7, 9–11, 13–15, and 21–24 are
`anticipated by Ghias.
`
`17
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 17 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`
`Patent Owner argues that the search in Ghias is exhaustive, detailing
`that the query engine compares the work to “all the songs” in the melody
`database, and thus cannot teach a “sublinear” or an “approximate nearest
`neighbor search,” per the independent claims. Prelim. Resp. 20–25 (citing
`Ex. 1010, 5:62–6:2). Patent Owner also argues computational time that it
`would take to perform a search grows linearly in Ghias, as opposed to
`sublinear growth recited in the claims, where Patent Owner continues that
`one example of a sub-linear time search is an approximate nearest neighbor
`search. Id. at 21–22 (citing Ex. 1001 9:12–19). Patent Owner continues that
`Petitioner quotes Ghias out of context, because Ghias’s approximate string
`matching algorithm is employed in the comparison of a work with a single
`record of the database. Id. at 23–24. We do not agree.
`Ghias provides that “[t]he number of matches that the database 14
`should retrieve depends upon the error-tolerance used during the key-
`search,” and “the user can perform a new query on a restricted search list
`consisting of songs just retrieved. This allows the user to identify sets of
`songs that contain similar melodies.” Ex. 1010, 6:63–65, 7:5–8 (emphases
`added). Thus, Ghias makes clear that the search need not be exhaustive, as
`Patent Owner has argued, and will act to “identify[] a close, but not
`necessarily exact or closest, match,” per our claim construction.
`Additionally, as we found above, the claims do not specify that the sub-
`linear search must be performed on a subset of all of the records, and not
`information within individual records. As such, we are persuaded that the
`sub-linear approximate string matching, in Ghias, satisfies the claimed
`recitation of “using the received features extracted from the media work to
`
`18
`
`Fontem Ex. 2026
`R.J. Reynolds Vapor Company v. Fontem Holdings 1 B.V. IPR2016-01268
`Page 18 of 23
`
`
`
`IPR2015-00345
`Patent 8,205,237 B2
`
`perform a sub-linear time search of extracted features of identified media
`works to identify a neighbor.”
`We also have reviewed Petitioner’s arguments and claim chart with
`respect to the remaining limitations of the independent claims and the
`dependent claims challenged on this ground. On the present record, we find
`that Petitioner has demonstrated a reasonable likelihood of prevailing in
`showing that claims 1–3, 5–7, 9–11, 13–15, and 21–24 are anticipated by
`Ghias.
`
`5. Alleged Obviousness of Claims 26, 27, 34, and 35 over Iwamura
`Alone or in Combination with Chen
`Petitioner argues that claims 26, 27, 34, and 35 are obvious over
`Iwamura alone, or in combination with Chen. Pet. 53–57. Both grounds are
`based on the alleged ground of claims 1, 3–5, 7–9, 11–13, 15, 16, 21–25, 29,
`30, 33, 37, and 38 being anticipated by Iwamura, discussed above. Pet. 53.
`Based on the di