Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 1 of 12
`
`UNITED STATES DISTRICT COURT
`SOUTHERN DISTRICT OF NEW YORK
`
`
`
`NETWORK-1 TECHNOLOGIES, INC.
`
`
`Plaintiff,
`
`
`
`
`
`
`
`
`- against -
`
`
`GOOGLE LLC and YOUTUBE, LLC
`
`Defendants.
`
`14 Civ. 2396 (PGG-SN)
`
`14 Civ. 9558 (PGG-SN)
`
`DEFENDANTS’ RESPONSE TO PLAINTIFF’S SUPPLEMENTAL BRIEF
`REGARDING DEFENDANTS’ MOTION FOR SUMMARY JUDGMENT
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 2 of 12
`
`
`
`Pursuant to the Court’s Order of July 11, 2022 (Dkt. No. 266), Google respectfully submits
`
`the following response to the Supplemental Brief in Opposition to Defendants’ Motion for
`
`Summary Judgment filed by Network-1 on September 23, 2022 (Dkt. No. 274).
`
`I.
`
`INTRODUCTION
`
`Behind Network-1’s supplemental brief lies this fundamental truth: it remains undisputed
`
`that Google’s Content ID system uses linear searching that cannot infringe the patent claims at
`
`issue here that require a sublinear search. Despite its heated rhetoric and inflammatory charges,
`
`Network-1 fails to raise any genuine issue of material fact for trial because Google’s system
`
`operates just as Google explained in its summary judgment papers in performing a non-infringing
`
`linear search. That was true before the single change made by Google’s engineers in 2020 that is
`
`the focus of Network-1’s supplemental brief, and it is true today. And the outrageous charges of
`
`the supplemental brief aside, the theoretical changes to Google’s Content ID system that formed
`
`the basis of Network-1’s attempt to avoid summary judgment in 2020 remain as “hypothetical”
`
`today as they were then. The changes that Network-1’s expert hypothesized “would result in
`
`sublinear scaling” were never implemented by Google, and they remain wholly irrelevant
`
`speculation about a system that does not exist. Summary judgment should be granted.
`
`Network-1’s supplemental brief is predicated on a single change made by Google software
`
`engineers in 2020 to the “Siberia version” of the company’s Content ID system. The change in
`
`question entailed
`
` as part of a
`
`search of an audio reference index—one of many indices used by Google in its Content ID system.
`
`Before the change, each lookup of the audio reference index involved scanning a fixed fraction
`
`
`
`for summary judgment, all of those searches were linear, rather than “sublinear.” Dkt. No. 224
`
`(Defs.’ Mem. in Supp. of Mot. for Summ. J.) at 16–21; see also Dkt. No. 240 (Pl.’s Opp. to Mot.
`
`. As Google explained in its motion
`
`
`
`1
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 3 of 12
`
`
`
`for Summ. J.) at 12 (“Network-1 agrees that ‘if additional references were added to the existing
`
`shard/partition structure, the [index lookup] portion of the search would scale linearly.’”). The
`
`Siberia version of the Content ID system therefore does not infringe the asserted claims of the ’988
`
`patent or the ’237 patent, all of which require a “sublinear” search, i.e., “[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.” Dkt. No. 146 (Am. Claim Constr. Chart) at 2.
`
`Following the change referenced in Network-1’s supplemental brief, each lookup of the
`
`audio reference index involved
`
`
`
`. See, e.g., Dkt. No. 274-2 (Ex. 86) at -409. By the same indisputable logic,
`
`searches of
`
` are also linear, rather than
`
`sublinear. That is because a search of a predetermined fraction of a feature space is a linear search
`
`under the agreed-upon construction, irrespective of whether
`
`
`
`. See, e.g., Dkt. No. 158-7 (Decl. of Michael D. Mitzenmacher
`
`in Supp. of Network-1’s Reply Claim Constr. Br.) ¶ 19 (“There are examples of searches that
`
`compare to less than all records in a data set that scale linearly, such as those that compare to the
`
`same percentage of a random selection of records in the data set regardless of data set size.”). The
`
`Siberia Version of the Content ID system did not perform a “sublinear” search in 2020 (or any
`
`preceding year), and it does not perform that kind of search in 2022 either.
`
`Network-1’s sudden interest in this human-implemented change to
`
`
`
` is as puzzling as it is misplaced. Although this particular
`
`modification was made several months after the close of fact discovery, Network-1 unquestionably
`
`knew during discovery that Google’s engineers could and sometimes did make this kind of change.
`
`For example, when a Google engineer was asked at her deposition in 2019 whether the number in
`
`
`
`2
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 4 of 12
`
`
`
`question has “always been
`
` as long as that index system has been used,” she explained
`
`that “we have tried many different things,” and “many different numbers are used in different
`
`indexes.” Dkt. No. 240-5 (Pasula Dep. Tr. (Oct. 4, 2019)) at 83:17–23. That deposition testimony
`
`has been on the docket since 2020, when it was filed with the summary judgment bundle. Even
`
`though all of this was apparent to Network-1 at the time, Network-1’s expert did not offer any
`
`opinions in 2019 or 2020 that were predicated on actual or potential changes to
`
`
`
` in the Siberia version of the Content ID system. See Section II supra. And
`
`for good reason: Any such changes were and are immaterial to Network-1’s claims of patent
`
`infringement, because a fixed-fraction search
`
` scales linearly—and therefore cannot
`
`infringe Network-1’s patents directed to “sublinear” searches.
`
`Although that legal issue is the only one presented by Network-1’s supplemental brief, the
`
`following section nevertheless begins by recounting the background leading to Network-1’s
`
`supplemental submission. While this additional detail may not be strictly necessary to understand
`
`why the Siberia version of the Content ID system does not perform the claimed “sublinear” search,
`
`it is set forth below because Network-1 makes the preposterous claim that “Google’s summary
`
`judgment arguments were false and misleading when Google made them.” Pl.’s Supp. Br. at 4.
`
`As the following background makes clear, that assertion is flat wrong, and frankly beyond the
`
`bounds of acceptable advocacy. Neither Network-1’s baseless attacks nor its eleventh-hour
`
`attempt to create a factual dispute have any bearing on Google’s motion for summary judgment.
`
`II.
`
`FACTUAL AND PROCEDURAL BACKGROUND
`
`Google’s summary judgment brief (Dkt. No. 224) set forth multiple grounds on which
`
`summary judgment should be entered with respect to all of the patent claims asserted by Network-
`
`1. Most of those patent claims are invalid (see id. at 15–16, 32–36), and all of the claims are not
`
`
`
`3
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 5 of 12
`
`
`
`infringed by either the older LSH version of Google’s Content ID system or the newer Siberia
`
`version of the system (see id. at 16–32, 36–40).
`
`As Google explained in further detail in its summary judgment papers, the Siberia version
`
`of the Content ID system includes indices that correspond to certain copyrighted video, audio, and
`
`melody content. See, e.g., id. at 9. The indices in question contain, among other things,
`
`
`
`
`
`. Id. at 10–11. A lookup of those indices involves, among other things,
`
`. Id. at 11–12. In the
`
`index of copyrighted video content that was used as an exemplar by Network-1’s expert and in the
`
`summary judgment briefing, the
`
`. That
`
`was the case during fact discovery in 2019, and when the summary judgment motion was filed in
`
`2020. See, e.g., id. at 9, 11; Dkt. No. 225 (Defs.’ Stmt. of Material Facts) ¶ 48 (“In the index for
`
`reference video content, for example,
`
`
`
`.”).
`
`The purported basis for Network-1’s supplemental brief is that the Google software
`
`engineers who designed and are continuously improving the Content ID system can and
`
`occasionally do change the
`
` by modifying the
`
`pertinent computer code. As noted, the change that Network-1 relies on involved
`
`
`
`.1 See, e.g.,
`
`Dkt. No. 274-2 at -409. While that kind of change does not happen every day, Network-1 was
`
`
`1 Although the change referenced by Network-1 was not made to the video reference index that
`was the primary source of the specific examples in both Network-1’s expert’s report and Google’s
`summary judgment papers, it would make no difference if the same change had also been made to
`that index.
`
`
`
`4
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 6 of 12
`
`
`
`well aware that it can and does occur—and not just because computer code as a general matter can
`
`be and sometimes is modified by humans. For example: When the lead engineer for this portion
`
`of the system was asked at her deposition in October 2019 whether the number in question had
`
`“always been
`
` as long as that index system has been used,” she responded that “we
`
`have tried many different things,” and “many different numbers are used in different indexes.”
`
`Dkt. No. 240-5 at 83:17–23. In response to follow-up questions, she explained that “where new
`
`YouTube videos are being looked up against the copyright index, we ask for
`
`she then described an example of a different index where the lookup always
`
`,” and
`
`
`
`and another instance where “we did not always
`
`.” Id. at 83:24–85:21. That deposition
`
`occurred before any expert reports were served, and the transcript was filed with the parties’
`
`summary judgment papers. In short, Google never suggested that the
`
`was somehow incapable of being changed through deliberate human intervention.
`
`As the materials filed by Google in 2020 make clear, the fraction
`
`
`
` is
`
`“fixed” because the value is specified in the computer code. See, e.g., Dkt. No. 224 at 21. For any
`
`given search of any given index, the fraction does not vary based on a factor such as the
`
`
`
`
`
`words, unless a human decides to modify the pertinent computer code, the system will continue
`
`.2 In other
`
`
`2 See, e.g., Dkt. No. 225 ¶ 120 (“The Siberia version of the Content ID system outputs a fixed
`fraction
`
`
`, regardless of the distance or difference between the
` performed by the
`.”); id. ¶ 121 (“In the search of
`Siberia version of the Content ID system in the index for reference video content, the
`
`, regardless of
`.”).
`
`the distance or difference between the
`
`
`
`5
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 7 of 12
`
`
`
`.
`
`
`
`Even though it was apparent during fact and expert discovery that Google’s engineers could
`
`and sometimes did make changes to the
`
`,
`
`Network-1 never offered any evidence or analysis suggesting that could somehow render the
`
`search “sublinear.” As Google explained in its summary judgment brief, Network-1’s expert
`
`“generally agree[d]” that “if additional references were added to the existing shard/partition
`
`structure,” then the “portion of the search” discussed here “would scale linearly” instead of
`
`sublinearly. Dkt. No. 224 at 16–17 (quoting Dkt. No. 226-6 (Expert Rep. of Dr. Michael
`
`Mitzenmacher re. Google’s Infringement (Dec. 20, 2019) ¶ 229)). The theory of infringement
`
`offered by Network-1’s expert was that Google would modify the structure of the index, such that
`
`“as the size of a reference index increases, so would the number of shards and partitions,” which
`
`purportedly “would result in sublinear scaling.” Dkt. No. 226-6 ¶¶ 230, 238. In describing one of
`
`the reasons why that theory fails as a matter of law, Google explained that Network-1’s expert’s
`
`contention that Google “might ‘increas[e] the number of partitions per shard as the size of the data
`
`set increases’ in order to create ‘sublinear scaling’” is premised on “a hypothetical change that
`
`Google might make to its system.” ECF No. 227 (Defs.’ Reply in Supp. of Mot. for Summ. J.) at
`
`4. That statement was and is accurate in every respect. The change addressed in Network-1’s
`
`supplemental brief has nothing to do with modifying the number of partitions that exist on each
`
`shard; rather, it involves a change to
`
`
`
`.3
`
`
`3 Network-1’s supplemental brief asserts that its expert “explained in his initial expert report that
`… the search of the Siberia system was designed to adapt to increases in the size of the data set by
`allowing it to adjust the portion of the index that was searched to preserve sublinear scaling.” Pls.’
`
`
`
`6
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 8 of 12
`
`
`
`A few months after Google filed its motion for summary judgment, Google notified
`
`Network-1 that as of January 2021, the portion of the Content ID system addressed here was no
`
`longer operating in any data centers located in the United States, which extinguished any ongoing
`
`claim of infringement. See, e.g., Dkt. No. 249 (Joint Ltr. Mot. (Apr. 7, 2021)) at 3–4. In response
`
`to that disclosure, Network-1 served additional requests for production and conducted additional
`
`depositions that were concerned primarily with the geographic relocation completed in January
`
`2021. See Dkt. No. 256 (Joint Stip. (May 7, 2021)).
`
`Network-1 did not suggest to Google or the Court that it believed the supplemental
`
`discovery had any bearing on Google’s motion for summary judgment until July 2022—more than
`
`a year after the document describing the change in question was produced, and more than seven
`
`months after Network-1 introduced it as a deposition exhibit. See ECF No. 265 (Joint Ltr. (July
`
`7, 2022)) at 2; Pl’s. Supp. Br. at 4. Even then, Network-1 did not indicate which subject it intended
`
`to address through supplemental briefing, let alone intimate that Google’s summary judgment
`
`submission was somehow “false and misleading,” as it now contends. Pl’s. Supp. Br. at 4. Those
`
`baseless assertions were made for the first time in Network-1’s supplemental brief.
`
`III. LEGAL ARGUMENT
`
`Two of the three patents asserted by Network in this case require, among other things, a
`
`“sublinear” search, which is “[a] search whose execution time scales with a less than linear
`
`
`Supp. Br. at 2. Network-1’s brief does not cite its expert’s initial report for that proposition, and
`the initial report does not say anything about “adjust[ing] the portion of the index searched to
`preserve sublinear scaling.” This can be confirmed by comparing paragraphs 229–240 of Dr.
`Mitzenmacher’s initial report from December 2019 (Dkt. No. 226-6) with paragraphs 17–26 of his
`supplemental report from August 2022 (Dkt. No 274-4). Google’s motion for summary judgment
`properly focused on the theories that Network-1 actually disclosed because “[t]he burden remains
`with the patentee to prove infringement, not on the defendant to disprove it.” Welker Bearing Co.
`v. PHD, Inc., 550 F.3d 1090, 1095 (Fed. Cir. 2008).
`
`
`
`7
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 9 of 12
`
`
`
`relationship to the size of the data set to be searched, assuming computing power is held constant.”
`
`ECF No. 146 at 2. There is no dispute that each individual search performed by the Siberia Version
`
`of the Content ID system is linear, rather than sublinear. Even “Network-1 agrees that ‘if
`
`additional references were added to the existing shard/partition structure, the [index lookup]
`
`portion of the search would scale linearly.’” Dkt. No. 240 at 12 (brackets in original). As indicated
`
`above and described in greater detail in the summary judgment briefs filed in 2020, Network-1 has
`
`not identified any factual dispute concerning whether additional references are added to “the
`
`existing shard/partition structure” of a pertinent index. See, e.g., Dkt. No. 227 at 4–6. Neither
`
`Network-1’s supplemental brief nor its expert’s supplemental report offers any support for its
`
`original hypothesis that “as the size of a reference index increases, so would the number of shards
`
`and partitions.” Dkt. No. 226-6 ¶ 230. Instead, Network-1 contends that an entirely different form
`
`of human-directed change—namely,
`
` … as the size
`
`of the data set increases’”—purportedly “results in sublinear scaling.” Pl’s. Supp. Br. at 4. The
`
`particular human intervention relied on by Network-1 is the decision to “
`
`
`
`” when performing a lookup of the “
`
`” as part of a
`
`.” Dkt. No. 274-2 at -406, 409.
`
`Network-1’s theory fails to create a triable issue because it fundamentally misapprehends
`
`what the patents’ claims require. All of the claims asserted in this case are directed to particular
`
`methods. Claim 33 of the ’237 patent, for example, is directed to “[a] computer-implemented
`
`method comprising,” among other things, the step of “determining, by the computer system, an
`
`identification of the media work using the media work extracted features to perform a sublinear
`
`approximate nearest neighbor search of reference extracted features of reference identified media
`
`works.” Dkt. No. 226-4 at 28:5–16 (emphasis added). Under governing law, it “is axiomatic that
`
`
`
`8
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 10 of 12
`
`
`
`a method claim is directly infringed only if each step of the claimed method is performed.” Lincoln
`
`Nat’l Life Ins. v. Transam. Life Ins., 609 F.3d 1364, 1370 (Fed. Cir. 2010) (emphasis in original);
`
`see ePlus, Inc. v. Lawson Software, Inc., 700 F.3d 509, 521 (Fed. Cir. 2012) (distinguishing “the
`
`capability of the accused system” from “an actual act of infringement” (emphasis in original)).
`
`Network-1 has not identified any particular instance in which the Content ID system
`
`actually performed a “sublinear” search of the audio reference index (or any other index). Before
`
`the change in question, the system performed a vast number of automated look-ups of the audio
`
`reference index that involved a scan of the contents of
`
`. There is
`
`no question that each individual search of this predetermined fraction of the index was linear. E.g.,
`
`Dkt. No. 226-6 ¶ 229 (report of Network-1’s expert) (“[S]everal of Defendants’ technical witnesses
`
`testified that if additional references were added to the existing shard/partition structure, the [index
`
`lookup] portion of the search would scale linearly. I generally agree with this notion.”). After
`
`Google’s engineers made the change referenced by Network-1, the system performed many more
`
`automated look-ups of the audio reference index, with each involving a scan of the contents of
`
`. Again, each of those searches is unquestionably linear. See id.
`
`Network-1’s theory is not that as a result of this one change in 2020 that Google somehow
`
`algorithmically adjusts
`
` each time new references are added to an index,
`
`as its own brief contends that “the index is always growing.” Pl’s. Supp. Br. at 6. Network-1 also
`
`does not argue that only the single search of the audio reference index performed immediately after
`
` was “sublinear” as a consequence of the human-imposed
`
`change. And it cannot be that the potential of one day manually changing the
`
`
`
` converts every linear search into a sublinear one. See, e.g., Cybersettle, Inc. v. Nat’l
`
`
`
`9
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 11 of 12
`
`
`
`Arbitration Forum, Inc., 243 F. App’x 603, 607 (Fed. Cir. 2007) (“A party that does not perform
`
`a claimed step does not infringe a method claim merely because it is capable of doing so.”).
`
`Instead of identifying the specific form of search that the claims require, Network-1
`
`contends that “the amount of the index interrogated in a particular individual search … is the wrong
`
`frame,” and the question here is “whether, over time, the work to be done grows in a linear fashion
`
`along with growth of the index.” Pl.’s Supp. Br. at 6. Network-1 offers no authority in support of
`
`this interpretation, but even taking the framing at face value, it is undisputed that the work to be
`
`done does grow in a linear fashion with the growth of the index. As Network-1’s expert puts it,
`
`“doubling the size of a reference index by simply adding those references to the existing shards …
`
`could result in the [index lookup] portion of the search taking approximately twice as long, or an
`
`(unnecessary) double of computing resources could be consumed for that portion of the search.”
`
`Dkt. No. 226-6 ¶ 230. Network-1 merely argues that in the face of this “linear scaling,” engineers
`
`can and sometimes do make other efficiency-enhancing changes, such as
`
` (e.g., from one that linearly
`
`
`
`
`
`, to one that linearly
`
`). That kind
`
`of human-implemented change does not alter the dispositive point that the very next search
`
`performed by the Siberia version of the Content ID system is still one in which “the work to be
`
`done grows in a linear fashion along with growth of the index.” Pl’s. Supp. Br. at 6.
`
`IV. CONCLUSION
`
`There is no genuine dispute as to any material fact concerning whether the Siberia version
`
`of Google’s Content ID system infringes the asserted claims of the ’988 and ’237 patents, all of
`
`which require a “sublinear” search. For this reason and the other reasons set forth in Google’s
`
`summary judgment filings in these cases, Google respectfully requests that the Court enter
`
`judgment in its favor.
`
`
`
`10
`
`
`

`

`Case 1:14-cv-02396-PGG-SN Document 279 Filed 09/30/22 Page 12 of 12
`
`
`
`Dated: September 30, 2022
`
`
`
`
`
`Respectfully Submitted,
`
`
`
`
`
`
`
`
`
`
` /s/ Andrew V. Trask
`Thomas H. L. Selby (pro hac vice)
`Samuel Bryant Davidoff
`Andrew V. Trask
`Melissa Collins (pro hac vice)
`Graham W. Safty (pro hac vice)
`WILLIAMS & CONNOLLY LLP
`680 Maine Avenue, S.W.
`Washington, DC 20024
`Phone: (202) 434-5000
`Fax: (202) 434-5029
`tselby@wc.com
`sdavidoff@wc.com
`atrask@wc.com
`mcollins@wc.com
`gsafty@wc.com
`
`For Matters in New York:
`WILLIAMS & CONNOLLY LLP
`650 Fifth Avenue, Suite 1500
`New York, NY 10022
`
`Kevin Hardy (pro hac vice)
`QUINN EMANUEL URQUHART &
`SULLIVAN, LLP
`1300 I Street, NW, Suite 900
`Washington, DC 20005
`Phone: (202) 538-8000
`Fax: (202) 538-8100
`kevinhardy@quinnemanuel.com
`
`Attorneys for Google LLC and
`YouTube, LLC
`
`11
`
`
`