`
`UNITED STATES DISTRICT COURT
`SOUTHERN DISTRICT OF NEW YORK
`
`
`
`
`NETWORK-1 TECHNOLOGIES, INC.,
`
`
`
`
`
`Plaintiff,
`
`
`
`v.
`
`
`
`
`
`14 Civ. 2396 (PGG)
`
`14 Civ. 9558 (PGG)
`
`
`
`
`
`
`GOOGLE LLC and YOUTUBE, LLC,
`
`
`
`
`
`
`
`
`Defendants.
`
`
`
`
`
`
`DECLARATION OF PROFESSOR MICHAEL D. MITZENMACHER IN SUPPORT OF
`PLAINTIFF NETWORK-1 TECHNOLOGIES, INC.’S
`OPENING CLAIM CONSTRUCTION BRIEF
`
`
`
`
`
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 2 of 32
`
`TABLE OF CONTENTS
`
`
`
`I.
`
`II.
`
`INTRODUCTION .............................................................................................................. 1
`
`QUALIFICATIONS ........................................................................................................... 1
`
`III. MATERIALS CONSIDERED ........................................................................................... 2
`
`IV.
`
`V.
`
`LEVEL OF ORDINARY SKILL IN THE ART ................................................................ 3
`
`LEGAL PRINCIPLES INVOLVED IN MY ANALYSIS ................................................. 3
`
`A. Claim Construction ............................................................................................................. 3
`
`B. Claim Definiteness .............................................................................................................. 4
`
`VI.
`
`BACKGROUND ................................................................................................................ 5
`
`A. Feature Extraction ............................................................................................................... 7
`
`B. Building the Databases of Reference Works ...................................................................... 9
`
`C. Comparing an Unknown Work with the Reference Works in the Database .................... 11
`
`VII. PARTICULAR CLAIM TERMS ..................................................................................... 16
`
`A. “non-exhaustive [. . .] search” ........................................................................................... 16
`
`B. “correlation information” .................................................................................................. 24
`
`C. “extracted features” and “extracting features” .................................................................. 26
`
`
`
`
`
`
`
`
`
`
`
`i
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 3 of 32
`
`I.
`
`INTRODUCTION
`
`1.
`
`I have been retained as an expert in the above-captioned case by counsel for
`
`Plaintiff Network-1 Technologies, Inc. I understand that Network-1 is currently asserting three
`
`patents in this case: U.S. Patent Nos. 8,010,988 (“the ‘988 patent”), 8,205,237 (“the ‘237 patent”),
`
`and 8,904,464 (“the ‘464 patent”) (collectively the “patents-in-suit”).
`
`2.
`
`These patents relate generally to systems and methods for identifying media content
`
`and performing actions associated with the identified content. All three were invented by Professor
`
`Ingemar J. Cox and all claim their priority to a provisional patent application filed on September
`
`14, 2000.
`
`II.
`
`QUALIFICATIONS
`
`3.
`
`I am currently employed as a Professor of Computer Science at Harvard University.
`
`Specifically, I am the Thomas J. Watson, Sr. Professor of Computer Science in the School of
`
`Engineering and Applied Sciences. I joined the faculty of Harvard as an Assistant Professor in
`
`January 1999. I was promoted to Associate Professor in 2002 and to Professor in 2005. In 2010,
`
`I began a three-year term as Area Dean, which is essentially equivalent to what other schools call
`
`Department Chair, of Computer Science, and held that position through June 2013. I am currently
`
`serving as Area Co-Chair of Computer Science for the 2018-2019 academic year. My work
`
`address is 33 Oxford Street, Cambridge, MA 02138. My primary research interests include design
`
`and analysis of algorithms, networks and data transmission, and information theory.
`
`4.
`
`I received my undergraduate degree in Mathematics and Computer Science from
`
`Harvard College in 1991. I received a Certificate of Advanced Study in Mathematics from
`
`Cambridge University in 1992. I received a Ph.D. in Computer Science from the University of
`
`California at Berkeley in 1996. From August 1996 to January 1999, I was employed as a Research
`
`
`
`1
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 4 of 32
`
`Scientist at Digital Systems Research Center, where my work included projects on algorithms for
`
`the Internet.
`
`5.
`
`I am listed as an inventor or co-inventor on 19 issued patents, and am the co-author
`
`of a textbook entitled “Probability and Computing” published by Cambridge University Press. I
`
`am a Fellow of the Association for Computing Machinery (ACM).
`
`6.
`
`I regularly serve on program committees for conferences in networking, algorithms,
`
`and communication. For example, I have served on the program committee multiple times for the
`
`SIGCOMM conference, which is the flagship annual conference of the ACM Special Interest
`
`Group on Data Communication (SIGCOMM). I have also served on numerous program
`
`committees related to algorithms, including the ACM Symposium on the Theory of Computing,
`
`the International Colloquium on Automata, Languages, and Programming, and the International
`
`Conference on Web Search and Data Mining.
`
`7.
`
`The field of endeavor at issue in this case is identification of electronic content
`
`(such as video or audio content) using algorithmic search techniques. I have published over 200
`
`research papers in computer science and engineering conferences and journals, many of which
`
`have explored algorithms and data structures for algorithmic search techniques, including both
`
`mathematical analysis and applications.
`
`8.
`
`A copy of my curriculum vitae is attached as Exhibit A to this Declaration. It
`
`contains a more complete listing of my professional activities and background.
`
`III. MATERIALS CONSIDERED
`
`9.
`
`In forming my opinions set forth in this declaration, I have reviewed, considered,
`
`and/or had access to the patent specifications and claims and their prosecution histories. I have
`
`also considered the parties’ respective proposed claim constructions. In addition, I have relied
`
`
`
`2
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 5 of 32
`
`upon my professional and academic experience, as well as a number of references (including
`
`academic papers, other patents, and other publications) identified in the body of this declaration.
`
`I reserve the right to consider additional materials or information as I become aware of them and
`
`to revise my opinions accordingly in light of such additional information.
`
`IV.
`
`LEVEL OF ORDINARY SKILL IN THE ART
`
`10.
`
`It is my understanding that analysis of claim interpretation is to be undertaken from
`
`the perspective of a person of ordinary skill in the art to which the patents are directed at the time
`
`of the invention, here in September 2000. The patents-in-suit are directed to the field of
`
`identification of electronic content (such as video or audio content) using algorithmic search
`
`techniques. In my opinion, a person of ordinary skill in this art would have 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.
`
`V.
`
`LEGAL PRINCIPLES INVOLVED IN MY ANALYSIS
`
`A.
`
`11.
`
`Claim Construction
`
`I have been informed that in connection with patent claim interpretation, a Court’s
`
`analysis begins with the language of the claims themselves and that the words of a claim are
`
`generally given their ordinary and customary meaning, i.e., the meaning that the term would have
`
`to a person of ordinary skill in the art in question at the time of the invention, i.e., as of the effective
`
`filing date of the patent application.
`
`12.
`
`I have further been informed that the person of ordinary skill in the art is deemed
`
`to read the claim term not only in the context of the particular claim in which the disputed term
`
`appears, but in the context of the entire patent, including the specification and its prosecution
`
`history.
`
`
`
`3
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 6 of 32
`
`13.
`
`In general, I understand that the Court looks to the sources available to the public
`
`that show what a person of skill in the art would have understood the disputed claim language to
`
`mean in the context of the specification, and that those sources include the words of the claims
`
`themselves, the remainder of the patent specification and prosecution history, as well as extrinsic
`
`evidence concerning relevant scientific principles, the meaning of technical terms, and the state of
`
`the art.
`
`B.
`
`Claim Definiteness
`
`14.
`I also understand that the defendants in this case assert that some terms of the claims
`in the patents-in-suit render certain claims indefinite.
`
`15.
`
`I have been informed that a claim is indefinite only if the claim, read in light of the
`
`specification delineating the patent, and the prosecution history, fails to inform, with reasonable
`certainty, those skilled in the art about the scope of the invention.
`
`16.
`I have also been informed that the definiteness requirement mandates clarity, while
`recognizing that absolute precision is unattainable.
`
`I have used these understandings in providing my opinions as set forth in this
`
`17.
`declaration.
`
`
`
`
`
`
`
`4
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 7 of 32
`
`VI.
`
`BACKGROUND
`
`18.
`
`There are three patents-in-suit about which I have been requested to provide my
`
`expert opinion:
`
`• U.S. Patent No. 8,010,988 (“the ’988 patent”, Exhibit 11).
`
`• U.S. Patent No. 8,205,237 (“the ’237 patent”, Exhibit 2).
`
`• U.S. Patent No. 8,904,464 (“the ’464 patent”, Exhibit 3).
`
`Collectively, I refer to all three of these patents as the “patents-in-suit.”
`
`19.
`
`The patents-in-suit generally share a common specification, so for ease of
`
`reference, unless otherwise indicated, I will refer to the ’988 patent specification in citing to the
`
`inventor’s description of his invention. I recognize that there are, however, some differences in
`
`the specifications of the patents-in-suit. The “Summary of the Invention” (Section 2 in the text of
`
`the specification) varies somewhat among the patents. Also, there are some additional discussions
`
`of certain material from the literature that was incorporated by reference in the earlier specification
`
`of the ’988 patent that is set forth in greater detail in the later patents, such as, for example, in
`
`Section 4.2.1.1.3 of the specification. In the ’237 and ’464 patents, for example, the specification
`
`contains more details drawn from one of the Yianilos references that was incorporated by
`reference. ’237 patent at 9:7-19; ’464 patent at 9:1-14.
`
`20.
`
`As set forth in the Detailed Description section of the patents-in-suit (Section 4),
`
`the patents describe systems and methods “for identifying works without the need of embedding
`
`signals therein. Once identified, such information can be used to determine a work-related action.”
`
`’988 patent at 5:39-42. In simple terms, these systems can analyze an unknown digital “work”
`
`
`1 All Exhibit references are to Exhibits that are attached to the Declaration of Amy E. Hayden in
`Support of Plaintiff Network-1 Technologies, Inc.’s Opening Claim Construction Brief, which I
`understand is to be filed concurrently with my declaration.
`5
`
`
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 8 of 32
`
`such as a piece of content (like an audio and/or a video file) using characteristics of that work, and
`
`then compare it, using those characteristics, to a collection of known references to determine if the
`
`unknown content matches one of the known references. See, e.g., id. at 6:58-62. If it does, the
`system can then take actions based on that identification.
`
`21.
`
`For example, assume that the unknown work is a video that includes the song
`
`“White Christmas” by Bing Crosby. The patents-in-suit describe an apparatus and methods to (1)
`
`process a video file to extract features, such as by sampling the audio portion of that file or
`
`performing calculations based on information within the file, and (2) then compare those extracted
`
`features to a database that includes features related to many songs (for our example, assume that
`
`the database includes “White Christmas” as well as other songs). If the comparison process
`
`(implemented by a matching algorithm) finds that the extracted features from the unknown video
`
`match the features extracted from the reference file for “White Christmas,” then the system looks
`
`to see what action the database provides for works that match “White Christmas.” That action is
`
`then associated with the unknown video. Such actions might include providing or displaying an
`
`advertisement, displaying a weblink, dialing a number, or performing an e-commerce transaction.
`See id. at 9:59-10:1.
`
`22.
`An example of how this process may be carried out is illustrated in Figure 1 of the
`patens-in-suit:
`
`
`
`6
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 9 of 32
`
`
`
`The patents-in-suit explain: “FIG. 1 is a process bubble diagram of operations that may be
`
`performed . . . in which intra-work information [information derived from the work itself, as
`
`opposed to information added or appended to the work] is used to identify the work.” Id. at 6:34-
`37.
`
`A.
`
`23.
`
`Feature Extraction
`
`The process of Figure 1 begins with a feature extraction operation that can be used
`
`to identify a known reference work. See id. at 7:11-8:2 (§ 4.2.1.1.1). The patents-in-suit explain
`
`that examples of a work can include “an image, an audio file or some portion of an audio signal or
`
`may be one or more frames or fields of a video signal, or a multimedia signal.” Id. at 7:18-20. In
`
`the example of the music video, the reference work could be the audio file and/or portions of the
`audio signal of Bing Crosby’s “White Christmas.”
`
`
`
`7
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 10 of 32
`
`24.
`
`The patents-in-suit explain that “[t]he purpose of the feature extraction operation is
`
`to derive a compact representation of the work that can subsequently be used for the purpose of
`
`recognition.” Id. at 7:20-23. The patents recognize that electronic works can be represented with
`
`shorter “sketches” or “fingerprints” that require far less space to store in computer memory, and
`
`far less computing resources to compare, but must be sufficiently complex that each sketch or
`
`fingerprint represents the underlying content (the primary work) with a low likelihood that two
`
`different primary works will have the same sketch or fingerprint. The patents refer to these so-
`
`called “sketches” or “fingerprints” as “compact electronic representations,” “feature vectors,” or
`“extracted features.”
`
`25.
`
`The patent specification teaches numerous ways in which feature extraction can be
`
`accomplished. See id. at 7:11-8:2 (§ 4.2.1.1.1). The specification explains that feature extraction
`
`operations derive a representation of the work by, for example, using “a pseudo-random sample of
`
`pixels” from a frame of a video. Id. at 7:23-26. In addition, feature extraction can be accomplished
`
`through the use of a variety of mathematical operations including Fourier, wavelet, or cosine
`
`transforms/decompositions or statistical methods like principle component analysis. See id. at
`
`
`
`7:26-43.
`
`
`
`
`
`8
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 11 of 32
`
`B.
`
`26.
`
`Building the Databases of Reference Works
`
`In the example illustrated in Figure 1, the patents-in-suit contemplate assembling a
`
`database of these sketches or fingerprints of reference works and associated actions that are
`
`connected to the individual reference works. See id. at 8:4-59 (§ 4.2.1.1.2). This process is
`illustrated in the following excerpt of Figure 1:
`
`
`
`In the example referenced earlier, “WORK @t1” could be the reference work, Bing Crosby’s
`
`“White Christmas.” This reference is used in generating the reference database. In Step 122, one
`
`or more “feature extraction operation(s)” are performed. These operations extract features from
`
`the reference work (in this example “White Christmas”) to generate one or more sketches or
`fingerprints that can be used to identify this work.
`
`27.
`
`In Step 124, the extracted features can also be associated to a work identification
`
`number or reference called a “work id.” Each work id is associated with a specific work, but may
`
`also be associated with one or more extracted features from the “WORK @t1.” The feature(s)
`
`(vector) and work ids are tied together to form the database 110 referred to as “WID Information”
`in Figure 1.
`
`
`
`9
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 12 of 32
`
`28.
`
`As part of the same or a different database, the work ids can be linked with
`
`“associated information” 136 (such as an action to be performed with respect to any works linked
`to the work id). This process is shown in the following excerpt also from Figure 1:
`
`
`
`Thus, the patents describe a system in which the operator could create/maintain a database of
`
`known references. For example, this database might include popular songs from major record
`
`companies, and television programs from major studios. The full versions of the reference items
`
`in the database might be very large (for example, the electronic file for a single song might be
`
`more than 1 megabyte (1 million bytes of data) and a film might require more than 1 gigabyte (1
`
`billion bytes of data)). Storing such references in their full length would require huge amounts of
`
`storage for a large database. Even more problematic, searching by comparing a reference to the
`
`entirety of a reference work would be very difficult and require significant computing resources
`and time.
`
`29.
`
`The patents contemplate a system where each reference work in the database is
`
`represented by a “sketch” or “fingerprint.” Although these sketches or fingerprints may be far
`
`more compact than the complete media file, they still need to be sufficiently complex that
`
`numerous different reference works will each be very unlikely to have the same sketch or
`fingerprint.
`
`
`
`10
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 13 of 32
`
`C.
`
`30.
`
`Comparing an Unknown Work with the Reference Works in the Database
`
`To compare an unknown video to the database of reference works, the patents
`
`explain that one can obtain a sketch or fingerprint of the unknown video and then search for a
`
`match in the database. The following excerpt from Figure 1 illustrates an example of this process
`with reference to “WORK @t2:”
`
`
`
`In the working example discussed earlier, the “WORK @t2” could be an uploaded video
`
`containing at least a portion of the song “White Christmas.” Thus, in this portion of the process,
`
`the same feature extraction operations that were run on the reference works used to generate the
`
`WID Information database 110 could now be run in Step 140 on the unknown work (“WORK
`
`@t2”)—the unknown video containing White Christmas in the example above. The extracted
`
`
`
`11
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 14 of 32
`
`features from the unknown work can then be matched with features in the WID Information
`database to identify a matching work.
`
`31.
`
`The patents explain that the matching of these sketches or fingerprints is not
`
`equivalent to looking up a word in a dictionary. Id. at 8:63-9:5. Looking up a word in a dictionary
`
`is a search for an exact, or identity, match in an ordered set of data. All of the words in the
`
`dictionary have been pre-processed by organizing them in alphabetical order. If one is searching
`
`for a particular word in the dictionary, it is possible to search for an exact match very efficiently
`
`because of the pre-processing organization of the data set. As the patents explain, the kind of
`
`matching involved here is different both because the comparisons are not looking for exact
`
`matches, and because the data set involves high dimensional data that is not ordered in the way a
`dictionary can be ordered.
`
`32.
`
`As explained in the patents, its comparisons are not necessarily looking for an exact
`
`match because there can be, for example, noise or distortions in the unknown video. Id. This
`
`could be a consequence of using imperfect recording technology, recording a video from a signal
`
`that had some static in it (like a weak television broadcast), recording a video using a video
`
`recorder pointed at a television showing a program, or many other possible reasons. Similarly, the
`
`unknown video might be altered slightly from the reference. For example, a reference television
`
`program might be 22 minutes and 30 seconds long, but an uploaded recording of the same
`
`television program might have started recording a few seconds early so that the recording is 22
`
`minutes and 45 seconds long. In each of these cases, the unknown video would not be identical to
`
`the reference work, but it is still desirable for the system to identify the two files as matching.
`
`Likewise, if one is searching for a song, an unknown sample might include only a portion of the
`song in it, but it still might be desirable to identify the unknown song as a match to the reference.
`
`
`
`12
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 15 of 32
`
`33.
`
`The system needs to be designed to recognize the two exemplary similar videos
`
`discussed above as a match, even though they are not identical. Comparing the entirety of the files
`
`would be very difficult. For example, a single video or audio file might be many millions of bytes
`
`(megabytes) long. Comparing all of the millions of individual data bytes in the entire file to all of
`
`the millions of individual data bytes of another (reference) file to see how similar they are would
`
`be extremely time consuming and difficult. The patents discuss using feature vectors of, compact
`
`electronic representations of, and sets of features extracted from the files to help simplify the
`
`comparisons. Shorter representations of the electronic works are easier to compare than the entire
`
`files, but the representations must be sufficiently complex that each compact representation is very
`
`unlikely to be the same for more than one primary work. For example, if one wanted to represent
`
`a song, a measure of the tempo, such as the beats per minute of the song, would be a simple way
`
`to represent the song, but many songs could have the same tempo, so that representation by itself
`
`would not be very helpful for use in comparisons to identify an unknown work. Rather, one could
`
`capture snapshots of or “sample” particular values (for example the pitch and intensity values2) at
`
`multiple times during the song. See id. at 7:11-42 (describing various feature extraction
`
`methodologies); see also infra ¶¶ 63-65. Each of these independent features could be compared
`
`to the same features of a reference work. If the values for many of the features were close to the
`
`same features for a reference work, it might be possible to infer that the two primary works (the
`
`works represented by the compact representations) were also similar. One would say that the two
`
`representations are close both in the feature space (the compact representations are similar) and
`close in the primary space (the works represented by the compact representations are also similar).
`
`
`2 Although these exemplary features are “human-recognizable features,” in practice the features
`may be defined via a mathematical process so they are recognizable to a computer processor, but
`are not necessarily recognizable by a human.
`
`
`
`13
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 16 of 32
`
`34.
`
`The patents explain that the matching discussed could use such things as a statistical
`
`comparison of the compact representations to determine similarity. ’988 patent at 9:3-5. The
`
`patents give several examples of such statistical comparisons including linear correlation,
`
`correlation coefficients, mutual information, Euclidean distance, and Lp-norms. Id. at 9:5-8. Each
`
`of these examples are types of comparisons that can be done between two feature vectors to try to
`
`measure not only whether they are the same, but whether they are sufficiently similar to
`
`characterize them (and the underlying works that the sketches or fingerprints represent) as
`matching.
`
`35.
`
`As noted above, the patent points out that these types of comparisons are different
`
`than simply looking up a word in a dictionary. Words in a dictionary have been pre-processed so
`
`that they are an ordered data set. Likewise, the search for a word in the dictionary is an identity
`
`search. When looking up a particular word (for example, the word “banjo”), the search is for an
`
`exact match, and the data set can be pre-processed (i.e., put in alphabetical order) to make the
`
`search straightforward so that one can quickly locate the word without looking at every page of
`
`the dictionary or every word on every page. If, however, one were searching for something
`
`different, a different kind of search might be required. First, consider the same word, “banjo,” but
`
`instead of looking for the word in the dictionary (a well-ordered data set), consider looking for it
`
`in the text of a novel—a more random, non-ordered data set that is not pre-processed. It would no
`
`longer be possible to simply turn to the pages where words beginning with the letter “b” were
`
`located. A different kind of search, such as looking at every word, might be necessary in that kind
`
`of search. Another, still different kind of search might be needed if one were to search for not only
`
`exact matches to the word “banjo,” but also words that were “similar” by some statistical measure.
`
`For example, consider a measure of similarity where a word might be considered similar to “banjo”
`
`
`
`14
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 17 of 32
`
`when it has approximately the same number of letters and the proximity of each letter to other
`
`letters on a keyboard is also considered. In this example, the (hypothetical) word “darko” might
`be considered similar when each letter is considered as shown in the table below:
`
`“banjo”
`b
`a
`n
`j
`o
`
`
`This kind of similarity search would not be simplified by the kind of pre-ordering of the data set
`
`“darko”
`D
`A
`R
`K
`O
`
`keyboard distance
`3 (b to d)
`0
`4 (n to r)
`1 (j to k)
`0
`
`that was utilized in the dictionary. Organizing the list of words alphabetically does not necessarily
`
`allow for easy pruning of the data set to limit the number of comparisons that need to be made.
`
`36.
`
`In this example, the total character distance between the two words could be
`
`represented as 7 (2 + 0 + 4 + 1 + 0). If it was decided that two words with a total character distance
`
`of 10 or less were sufficiently similar to be called a “match,” then these two words would be a
`
`match. Alternatively, one could decide that a total character distance of no more than 4 was needed
`
`to identify a match, so “darko” would not be a match, but the (hypothetical) word “banko” would
`
`be close enough. To find this kind of a match, even in a dictionary, is far more complicated than
`simply looking up the word “banjo,” and a different search strategy would be needed.
`
`37.
`
`The patents further explain that other information can be stored with the reference
`
`works. For example, additional information can be connected with a reference work about actions
`
`that are to be performed in connection with that work. Id. at 6:34-60. This kind of action
`
`information could be stored in a single database as part of the same record for the reference work,
`
`or as part of a separate database identified with a common “key” that is also connected to the
`
`record for the reference work. Id. These actions could include various things like displaying an
`
`
`
`15
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 18 of 32
`
`advertisement, displaying a weblink, or initiating an e-commerce transaction. See, e.g., id. at 9:65-
`10:1.
`
`38.
`
`The patents go on to explain that when an unknown work is identified as a “match”
`
`to a reference work in the database, then the action that is connected with the reference work can
`
`also be performed in connection with the newly-identified uploaded work. Id. at 9:59-10:4. Thus,
`
`for example, if an advertisement or weblink (or both) are to be displayed with a particular reference
`
`work, then those same actions can be performed with the identified uploaded work once it is
`identified.
`
`VII. PARTICULAR CLAIM TERMS
`
`A.
`
`39.
`
` “non-exhaustive [. . .] search”
`
`I understand that Defendants contend this term is indefinite, and Network-1 offers
`
`the following construction: “a search designed to locate a [near] neighbor without comparing to
`
`all possible matches (i.e., all records in the reference data set), even if the search does not locate a
`
`[near] neighbor.” I further understand that this claim term appears in the claims asserted from the
`
`’988 patent (claim 17) and the ’464 patent (claims 1, 8, 10, 16, 18, 25, 27, and 33).
`
`40.
`
`The term “non-exhaustive search” is a term that is well understood by skilled
`
`artisans in the field of the patents, and was well understood at the time of the invention. In my
`
`view, it is not indefinite.
`
`41.
`
`In my expert opinion, Network-1’s proposed definition accurately reflects the
`
`ordinary meaning of this term that would have been understood by persons of ordinary skill in the
`
`art at the time of the patents. Skilled artisans then and now understand that the difference between
`
`a search being exhaustive and non-exhaustive is not a question of whether or not the search can
`
`find a match, but rather as turning on the methodology employed in the search itself. Specifically,
`
`
`
`16
`
`
`
`
`
`Case 1:14-cv-02396-PGG-MHD Document 148-1 Filed 05/30/19 Page 19 of 32
`
`the difference between exhaustive and non-exhaustive searches turns on the number of
`
`comparisons that must be performed between the query and the reference data set to be searched.
`
`The patents discuss this concept in the specification, for example at columns 8-9 and 21-22 of the
`
`’988 patent.
`
`42.
`
`Essentially, the patents explain that an exhaustive search may involve comparing a
`
`query to every record in the data set to be searched until a match is found. The patents note that
`
`on average, where you can halt a search after the match is found (and there is only a single match
`
`for each query), this would require N/2 comparisons (where N is the number of records in the data
`
`set). See, e.g., id. at 9:24-28. This means that, if you can stop after finding a match, over a number
`
`of searches, you will, on average, find the match after comparing the query to half of the total
`
`number of records in the set. This stands to reason (assuming that the queries are random in
`
`relation to the data set) because some matches might be found after being compared to only a small
`
`percentage of the set, while others might require comparisons to a much higher percentage of the
`
`set before finding matches. Collectively, though, the average is expected to be N/2. As a
`
`simplified example, we can consider a hypothetical search of a database of people’s names and
`
`social security numbers. Assume that the database is sorted only by the social security numbers,
`
`and that there are no repeated names in the database. If one has a name and is looking for the
`
`social security number associated with that name, the search might need to compare the name to
`
`every name in the database until a match is found. The first search might find a name after only a
`
`small number of comparisons, but the next search might not find a matching name until it had
`
`comp