`
`
`
`
`
`
`
` Exhibit 29
`
`
`
`
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/2CbgwflfifébltopY
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 2 of 219
`
`UNITED STATES DISTRICT COURT
`
`FOR THE SOUTHERN DISTRICT OF NEW YORK
`
`NETWORK-1 TECHNOLOGIES,
`
`INC.
`
`Plantiff
`
`
`
`vs.
`
`Case No. 14—CV-2396
`
`GOOGLE,
`
`INC., and YOUTUBE, LLC
`
`Defendant
`
`CONFIDENTIAL
`
`DEPOSITION OF SHUMEET BALUJA, Ph.D.
`
`June 19, 2015
`
`Michele E. Edd Notar Public
`393685
`Y'
`Y
`
`1972 ®
`
`BARKLEY
`
`barkley.com
`
`(858) 455-5444 San Diego
`(949) 955-0400 Irvine
`(415) 433-5777 ‘San Francisco
`(310) 207-8000 Los Angeles
`(951) 686-0606 Riverside
`(760) 322-2240 Palm Springs
`(408) 885-0550 San Jose
`(916) 922-5777 Sacramento
`(518) 490-1910 Albany
`(347) 821 -4611 Brooklyn
`(212) 808—8500 New York City
`(818) 702—0202 Woodland Hills
`(702) 366-0500 Las Vegas
`(312) 379-5566 Chicago
`(914) 510-9110 White Plains
`(516) 277-9494 Garden City
`00+1+800 222 1231 Paris
`00+1+800 222 1231 Dubai
`001+1+800 222 1231 Hong Kong
`
`
`
` 1 UNITED STATES DISTRICT COURT
`
` 2 FOR THE SOUTHERN DISTRICT OF NEW YORK
`
` 3
`
` 4 NETWORK-1 TECHNOLOGIES, INC.
`
` 5 Plantiff
`
` 6 vs. Case No. 14-CV-2396
`
` 7 GOOGLE, INC., and YOUTUBE, LLC
`
` 8 Defendant
`
` 9 _______________________________/
`
`10
`
`11 CONFIDENTIAL
`
`12 The deposition of SHUMEET BALUJA, Ph.D., was
`
`13 held on the 19th day of June, 2015, commencing at 9:09
`
`14 a.m., at the Law Offices of Finnegan, Henderson, Farabow,
`
`15 Garrett & Dunner, LLP, Two Freedom Drive, Reston,
`
`16 Virginia, before Michele E. Eddy, Notary Public.
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25 REPORTED BY: MICHELE E. EDDY, RPR, CRR, CLR
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 3 of 219
`
`2
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
` 1 APPEARANCES:
`
` 2
`
` 3 ON BEHALF OF THE PLAINTIFF:
`
` 4 BRIAN D. LEDAHL, ESQUIRE
`
` 5 Russ August & Kabat
`
` 6 12424 Wilshire Boulevard, 12th Floor
`
` 7 Los Angeles, California 90025
`
` 8 Telephone: (310) 826-7474
`
` 9 Facsimile: (310) 826-6991
`
`10 Email: Bledahl@raklaw.com
`
`11
`
`12
`
`13 ON BEHALF OF THE DEFENDANTS:
`
`14 DOUGLAS R. NEMEC, ESQUIRE
`
`15 Skadden, Arps, Slate, Meagher & Flom, LLP
`
`16 4 Times Square
`
`17 New York, New York 10036
`
`18 Telephone: (212) 735-2419
`
`19 Facsimile: (917) 777-2419
`
`20 Email: Douglas.nemec@skadden.com
`
`21
`
`22 ALSO PRESENT: James Sherwood, In-house Counsel, Google
`
`23 Daniel Holmstock, Videographer
`
`24
`
`25
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 4 of 219
`
`3
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
` 1 INDEX
`
` 2 Deposition of SHUMEET BALUJA, Ph.D.
`
` 3 June 19, 2015
`
` 4 Examination By: Page
`
` 5 Mr. Ledahl 7
`
` 6 Exhibit No. Marked
`
` 7 Exhibit 68 E-mail chain; top e-mail 96
`
` 8 dated 12-8-06 from Shumeet
`
` 9 Baluja to Jay Yagnik;
`
`10 GOOG-NETWORK-00656181-186
`
`11 Exhibit 69 Content ID Moma Posting; 114
`
`12 GOOG-NETWORK-00598291-92
`
`13 Exhibit 70 Paper titled "Waveprint: 118
`
`14 Efficient Wavelet-based
`
`15 Audio Fingerprinting" by
`
`16 Shumeet Baluja and Michele
`
`17 Covell;
`
`18 GOOG-NETWORK-00610574-87
`
`19 Exhibit 71 Abstract titled "Permutation 153
`
`20 Grouping: Intelligent Hash
`
`21 Function Design for Audio &
`
`22 Image Retrieval" by Shumeet
`
`23 Baluja, Michele Covell, and
`
`24 Sergey Ioffe;
`
`25 GOOG-NETWORK-00610856-59
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 5 of 219
`
`4
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
` 1 INDEX (Continued)
`
` 2 Deposition of SHUMEET BALUJA, Ph.D.
`
` 3 June 19, 2015
`
` 4
`
` 5 Exhibit No. Marked
`
` 6 Exhibit 72 PowerPoint titled "Audio and 158
`
` 7 Video Snippet Identification"
`
` 8 by Michele Covell, Jeff Faust,
`
` 9 Sergey Ioffe, Dave Marwood,
`
`10 Shumeet Baluja, and Jay Yagnik;
`
`11 GOOG-NETWORK-00633973-4049
`
`12 Exhibit 73 U.S. Patent No. US 2009/0052785; 165
`
`13 GOOG-NETWORK-00696867-890
`
`14 Exhibit 74 Abstract titled "Learning to 176
`
`15 Hash: Forgiving Hash Functions
`
`16 and Applications" by Shumeet
`
`17 Baluja and Michele Covell;
`
`18 GOOG-NETWORK-00613418-464
`
`19 Exhibit 75 U.S. Patent No. 8,184,953; 188
`
`20 GOOG-NETWORK-00697356-76
`
`21
`
`22
`
`23
`
`24
`
`25
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 6 of 219
`
`5
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
` 1 PROCEEDINGS,
`
` 2 THE VIDEOGRAPHER: This is video number
`
`09:09 3 1 in the video recorded deposition of Shumeet
`
`09:09 4 Baluja taken by the plaintiff in the matter of
`
`09:09 5 Network-1 Technologies, Inc., Plaintiff, versus
`
`09:09 6 Google, Inc., and YouTube, LLC, Defendants,
`
`09:09 7 pending before the United States District Court
`
`09:09 8 for the Southern District of New York, Case Number
`
`09:09 9 14-CV-2396.
`
`09:09 10 This deposition is being held at the law
`
`09:09 11 offices of Finnegan Henderson at 11955 Freedom
`
`09:09 12 Drive in Reston, Virginia, on June 19th, 2015.
`
`09:09 13 The time on the video screen is 9:09 a.m.
`
`09:10 14 My name is Daniel Holmstock. With me is
`
`09:10 15 Michele Eddy. We are in association with Barkley
`
`09:10 16 Court Reporters, Inc., located at 1875 Century
`
`09:10 17 Park East, Suite 1300, in Los Angeles, California.
`
`09:10 18 For the record now will counsel please
`
`09:10 19 introduce themselves and whom they represent.
`
`09:10 20 MR. LEDAHL: Brian Ledahl, from Russ
`
`09:10 21 August & Kabat, on behalf of the plaintiff
`
`09:10 22 Network-1.
`
`09:10 23 MR. NEMEC: Doug Nemec, from Skadden
`
`09:10 24 Arps, on behalf of the defendants Google and
`
`09:10 25 YouTube, and with me is Jim Sherwood, in-house
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 7 of 219
`
`6
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:10 1 counsel for Google.
`
`09:10 2 THE VIDEOGRAPHER: Will the court
`
`09:10 3 reporter please administer the oath.
`
`09:10 4 Whereupon,
`
`09:10 5 SHUMEET BALUJA, Ph.D.,
`
`09:10 6 having been duly sworn, testified as follows:
`
`09:10 7 EXAMINATION BY MR. LEDAHL:
`
`09:10 8 Q Good morning, Mr. Baluja.
`
`09:10 9 A Hi.
`
`09:10 10 Q Thank you for coming in today. At the
`
`09:10 11 outset I'm going to want to cover a couple, sort
`
`09:10 12 of, preliminary matters, but before I do that,
`
`09:10 13 have you ever had your deposition taken before
`
`09:10 14 today?
`
`09:10 15 A No, this is the first.
`
`09:10 16 Q Okay. So I'm going to cover a couple
`
`09:10 17 things that hopefully will make the process a
`
`09:10 18 little smoother for both us and for the court
`
`09:10 19 reporter. First of all, do you understand that
`
`09:11 20 the oath you took a moment ago is the same oath
`
`09:11 21 you would take if you were testifying live in a
`
`09:11 22 court in front of a judge and a jury?
`
`09:11 23 A I do.
`
`09:11 24 Q You understand that that oath carries
`
`09:11 25 the same penalties of perjury as if you were
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 8 of 219
`
`7
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:11 1 testifying live in court?
`
`09:11 2 A I do.
`
`09:11 3 Q Now, as you may have noticed, we have a
`
`09:11 4 court reporter and a video recording of the
`
`09:11 5 deposition proceedings today. As a consequence of
`
`09:11 6 that, there are a couple of things I'm going to
`
`09:11 7 ask that we try to observe to make the process a
`
`09:11 8 little smoother and to make sure we have a clear
`
`09:11 9 record of the proceedings.
`
`09:11 10 Because the court reporter takes down
`
`09:11 11 what we say, it's important that we do our best to
`
`09:11 12 try not to speak at the same time since she can
`
`09:11 13 only record one of us at a time. So, to that end,
`
`09:11 14 I'll do my best to wait until you finish answering
`
`09:11 15 one of my questions to ask another. Likewise, if
`
`09:11 16 you can wait until I finish my question to start
`
`09:11 17 your answer, it will make things a lot clearer and
`
`09:11 18 simpler for the court reporter. Is that
`
`09:11 19 acceptable?
`
`09:11 20 A Yes.
`
`09:11 21 Q Similarly, because it's very hard for
`
`09:12 22 her to capture things like nods or shakes of the
`
`09:12 23 head or words like uh-hmm and huh-uh clearly, I'm
`
`09:12 24 going to ask that you do your best to remember to
`
`09:12 25 give clear verbal responses like yes or no so that
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 9 of 219
`
`8
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:12 1 we have a clear record. Is that acceptable?
`
`09:12 2 A Yes.
`
`09:12 3 Q Now, after the deposition, the court
`
`09:12 4 reporter will prepare a transcript and you will
`
`09:12 5 have an opportunity to review that and make
`
`09:12 6 corrections or changes if you believe they are
`
`09:12 7 necessary. However, I want to caution you that to
`
`09:12 8 the extent you change the substance of your
`
`09:12 9 testimony, I will have the opportunity at some
`
`09:12 10 point, at a trial, perhaps, to comment on whether
`
`09:12 11 that may affect or implicate your credibility. So
`
`09:12 12 to the extent possible, I'm going to ask you to do
`
`09:12 13 your best to give your best and clearest
`
`09:12 14 recollection today. Do you understand?
`
`09:12 15 A Yes.
`
`09:12 16 Q And is there any reason, such as a
`
`09:12 17 medical reason or medication, that would prevent
`
`09:12 18 you from providing your best and most accurate
`
`09:12 19 recollections today?
`
`09:12 20 A No.
`
`09:12 21 Q During the course of the day, if at any
`
`09:12 22 point you don't understand my question, please let
`
`09:12 23 me know. I'll do my best to try to clarify it for
`
`09:13 24 you. Otherwise if you can provide an answer, I'll
`
`09:13 25 assume you understood the question. Okay?
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 10 of 219
`
`9
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:13 1 A Okay.
`
`09:13 2 Q Similarly, as we go along, if at any
`
`09:13 3 point you need a break, please let me know. I'll
`
`09:13 4 ask that we don't take breaks while we have a
`
`09:13 5 question to you pending but otherwise we'll do our
`
`09:13 6 best to accommodate that as soon as we can.
`
`09:13 7 A Okay.
`
`09:13 8 Q Now, Mr. Baluja, where are you currently
`
`09:13 9 employed?
`
`09:13 10 A Google.
`
`09:13 11 Q How long have you been at Google?
`
`09:13 12 A Almost 12 years.
`
`09:13 13 Q So since about 2003?
`
`09:13 14 A Correct.
`
`09:13 15 Q I would like to start a little bit with
`
`09:13 16 your, sort of, educational background. So
`
`09:13 17 starting with any college or university education,
`
`09:13 18 where, if at all, did you attend university?
`
`09:13 19 A I went to University of Virginia, and
`
`09:13 20 then I went to Carnegie Mellon for my Ph.D.
`
`09:13 21 Q So at UVA you -- what degree did you
`
`09:13 22 obtain there?
`
`09:13 23 A Computer science and philosophy.
`
`09:13 24 Q Was that a bachelor's degree?
`
`09:13 25 A Correct.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 11 of 219
`
`10
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:14 1 Q What year did you graduate and receive
`
`09:14 2 your degree from UVA?
`
`09:14 3 A I think '92, I think. '91 or '92. I
`
`09:14 4 don't remember.
`
`09:14 5 Q Then you said you went to Carnegie
`
`09:14 6 Mellon from there?
`
`09:14 7 A Correct.
`
`09:14 8 Q And that was for, I think you said, a
`
`09:14 9 Ph.D.?
`
`09:14 10 A Correct.
`
`09:14 11 Q What was the area of study for your
`
`09:14 12 Ph.D.?
`
`09:14 13 A Computer science and robotics.
`
`09:14 14 Q And did you complete your Ph.D. studies
`
`09:14 15 there?
`
`09:14 16 A Yes.
`
`09:14 17 Q What year did you receive your degree?
`
`09:14 18 A '96.
`
`09:14 19 Q Did you receive any interim degrees such
`
`09:14 20 as a master's degree along the way?
`
`09:14 21 A No.
`
`09:14 22 Q Do you have any other postgraduate
`
`09:14 23 education other than the Ph.D. from Carnegie
`
`09:14 24 Mellon?
`
`09:14 25 A No.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 12 of 219
`
`11
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:14 1 Q And did I understand correctly that you
`
`09:14 2 went essentially straight through from UVA to
`
`09:14 3 Carnegie Mellon?
`
`09:14 4 A Yes.
`
`09:14 5 Q Who was your -- well, strike that.
`
`09:14 6 Did you have a faculty advisor or Ph.D.
`
`09:15 7 advisor at Carnegie Mellon?
`
`09:15 8 A Yes.
`
`09:15 9 Q Who was that?
`
`09:15 10 A Dean Pomerleau and Tom Mitchell.
`
`09:15 11 Q Can you spell Pomerleau for me?
`
`09:15 12 A Yes, P-O-M-E-R-L-E-A-U.
`
`09:15 13 Q And the other person you said was?
`
`09:15 14 A Tom Mitchell.
`
`09:15 15 Q Tom Mitchell. That, I can probably
`
`09:15 16 spell.
`
`09:15 17 After receiving your Ph.D. at Carnegie
`
`09:15 18 Mellon, did you have employment shortly
`
`09:15 19 thereafter?
`
`09:15 20 A Yes.
`
`09:15 21 Q Where did you work first after your
`
`09:15 22 Ph.D.?
`
`09:15 23 A Justsystem Pittsburgh Research Center.
`
`09:15 24 Justsystem, J-U-S-T.
`
`09:15 25 Q Got it.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 13 of 219
`
`12
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:15 1 A Yes.
`
`09:15 2 Q And what was the business or activity of
`
`09:15 3 Justsystem?
`
`09:15 4 A They were basically the Microsoft of
`
`09:15 5 Japan so they did all the same stuff that
`
`09:15 6 Microsoft does but in Japan.
`
`09:15 7 Q What did you do at Justsystem?
`
`09:16 8 A Research.
`
`09:16 9 Q In any particular area or focus?
`
`09:16 10 A General computer science. I was all
`
`09:16 11 over the place.
`
`09:16 12 Q How long did you work at Justsystem?
`
`09:16 13 A Two years.
`
`09:16 14 Q So roughly until 1998?
`
`09:16 15 A Yes, right around then, yes.
`
`09:16 16 Q Where did you go after working at
`
`09:16 17 Justsystem?
`
`09:16 18 A Lycos, Incorporated.
`
`09:16 19 Q Is that L-Y-C-O-S?
`
`09:16 20 A Correct.
`
`09:16 21 Q What did you do at Lycos?
`
`09:16 22 A I was a chief scientist.
`
`09:16 23 Q What was Lycos' business, generally?
`
`09:16 24 A Web search.
`
`09:16 25 Q How long were you at Lycos?
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 14 of 219
`
`13
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:16 1 A About a year and a half or so.
`
`09:17 2 Q After Lycos, where did you go next?
`
`09:17 3 A It was called eCompanies.
`
`09:17 4 Q That would have been in about 2000 or
`
`09:17 5 thereabouts?
`
`09:17 6 A Yes, right around there. I don't
`
`09:17 7 remember the exact dates.
`
`09:17 8 Q What did you do at eCompanies?
`
`09:17 9 A I was the senior vice president of
`
`09:17 10 research.
`
`09:17 11 Q What was eCompanies' business generally?
`
`09:17 12 A They're an incubator.
`
`09:17 13 Q Did you focus on particular companies or
`
`09:17 14 developing companies within eCompanies?
`
`09:17 15 A Data mining and generally the technology
`
`09:17 16 side of companies.
`
`09:18 17 Q So was your role as providing some kind
`
`09:18 18 of research or scientific and technical support to
`
`09:18 19 various companies that were incubated at
`
`09:18 20 eCompanies?
`
`09:18 21 A Yeah, it was much less scientific. When
`
`09:18 22 you're incubating companies, it's more technical,
`
`09:18 23 yes.
`
`09:18 24 Q Fair enough.
`
`09:18 25 Do you know approximately how many
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 15 of 219
`
`14
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:18 1 companies, if any, spun out from that incubator
`
`09:18 2 during the time you were there?
`
`09:18 3 A Yes, I think we had three or four IPOs
`
`09:18 4 or buyouts.
`
`09:18 5 Q Do you remember the names of any of the
`
`09:18 6 companies?
`
`09:18 7 A Sure. Business.com, Jamdat Mobile, USB
`
`09:18 8 X. I think that was one. I don't remember the
`
`09:18 9 other one.
`
`09:18 10 Q How long were you at e- --
`
`09:19 11 A Oh, Boingo.
`
`09:19 12 Q Boingo.
`
`09:19 13 How long were you at eCompanies?
`
`09:19 14 A A year and a half or so.
`
`09:19 15 Q Where did you go next after working at
`
`09:19 16 eCompanies?
`
`09:19 17 A Jamdat Mobile.
`
`09:19 18 Q What did you do at Jamdat?
`
`09:19 19 A I was the CTO, chief technology officer.
`
`09:19 20 Q And, generally speaking, what was
`
`09:19 21 Jamdat's business at that time?
`
`09:19 22 A Mobile games.
`
`09:19 23 Q How long did you stay there?
`
`09:19 24 A Two years, I think.
`
`09:19 25 Q If I've got the timing correct, from
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 16 of 219
`
`15
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:19 1 there did you go to Google?
`
`09:20 2 A I took a year off.
`
`09:20 3 Q Why did you leave Jamdat?
`
`09:20 4 A My dad had health issues in Virginia so
`
`09:20 5 I left there.
`
`09:20 6 Q After you -- so was it after this year
`
`09:20 7 off that you began working at Google?
`
`09:20 8 A Correct.
`
`09:20 9 Q What led you to join Google?
`
`09:20 10 A A lot of my friends were there.
`
`09:20 11 Q Did you work at Google? Like where
`
`09:20 12 physically, geographically?
`
`09:20 13 A I went to Mountain View.
`
`09:20 14 Q When you first started at Google, what
`
`09:20 15 was your position or role there?
`
`09:20 16 A Researcher.
`
`09:20 17 Q Was there any particular focus or area
`
`09:21 18 that you were involved in?
`
`09:21 19 A No. Research is pretty open.
`
`09:21 20 Q Was your -- I take it there is a
`
`09:21 21 research department or research group at Google;
`
`09:21 22 is that right?
`
`09:21 23 A Correct.
`
`09:21 24 Q Was there any kind of group organization
`
`09:21 25 within that such that there were particular focus
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 17 of 219
`
`16
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:21 1 areas of any kind?
`
`09:21 2 A There were, but I was not a part of
`
`09:21 3 that.
`
`09:21 4 Q I see.
`
`09:21 5 Did at you focus personally on anything
`
`09:21 6 in particular when you started?
`
`09:21 7 A It would be -- okay, so, caveats on this
`
`09:21 8 is this was 10, 12 years ago so I don't remember.
`
`09:21 9 The first project, I can tell you what I worked
`
`09:21 10 on. I started the wireless group at Google.
`
`09:21 11 Q How long were you working on that
`
`09:21 12 project?
`
`09:22 13 A So I'll say I worked two or three years
`
`09:22 14 on it, but then that wasn't full-time, right, so
`
`09:22 15 you pick up projects as you go along.
`
`09:22 16 Q What was, in a general sense, the
`
`09:22 17 wireless group's focus?
`
`09:22 18 A Your cell phone that you have, we
`
`09:22 19 started that.
`
`09:22 20 Q And was that in the context of wireless
`
`09:22 21 communication or operating systems for phones or
`
`09:22 22 some other focus?
`
`09:22 23 A Yes.
`
`09:22 24 Q Both?
`
`09:22 25 A Everything.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 18 of 219
`
`17
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:22 1 Q Then what was the next major, sort of,
`
`09:22 2 project or area that you focused on?
`
`09:22 3 A Kid Safe Search. So I tried to keep
`
`09:22 4 kids safe on the Internet.
`
`09:23 5 Q And did that work start sometime during
`
`09:23 6 that two to three years or so that you mentioned?
`
`09:23 7 A Exactly, yes.
`
`09:23 8 Q What was the next, sort of, major
`
`09:23 9 project or area that you focused on?
`
`09:23 10 A Two, image processing and advertising --
`
`09:23 11 oh, and graphs.
`
`09:23 12 Q What was the image processing area that
`
`09:23 13 you mentioned?
`
`09:23 14 A Everything from face recognition to
`
`09:23 15 analyzing images.
`
`09:23 16 Q What was the advertising project that
`
`09:23 17 you mentioned?
`
`09:23 18 A Social -- all I can say is social
`
`09:23 19 advertising.
`
`09:23 20 Q Can you give me just a very high-level
`
`09:24 21 explanation of what that means?
`
`09:24 22 A Sure. You're familiar with the
`
`09:24 23 advertising you see on Facebook?
`
`09:24 24 Q Sure.
`
`09:24 25 A So this is on our social networks.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 19 of 219
`
`18
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:24 1 Q So, essentially, connecting advertising
`
`09:24 2 in some way to content on someone's social
`
`09:24 3 network --
`
`09:24 4 A Sure.
`
`09:24 5 Q -- or relating thereto?
`
`09:24 6 A I can accept that.
`
`09:24 7 Q Okay. And what was the graphs project
`
`09:24 8 or area that you mentioned?
`
`09:24 9 A So, like I mentioned, you know, Facebook
`
`09:24 10 is a big social graph. We also have social graphs
`
`09:24 11 as well as connectivity between web pages, the
`
`09:24 12 graphs of the whole web, so graph algorithms.
`
`09:24 13 Q What's the purpose of these kind of
`
`09:24 14 graph algorithms?
`
`09:24 15 A In general, it falls under
`
`09:24 16 understanding -- understanding the content of a
`
`09:24 17 web page.
`
`09:24 18 Q So are you referring generally to, sort
`
`09:24 19 of, the kind of complex interconnection and
`
`09:25 20 linkages that occur between multiple websites and
`
`09:25 21 locations and --
`
`09:25 22 A Correct.
`
`09:25 23 Q -- recognizing those relationships?
`
`09:25 24 A Yes. So you're familiar -- and the
`
`09:25 25 big -- the big one is PageRank, as I'm sure you've
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 20 of 219
`
`19
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:25 1 heard Google was based on when it started. So
`
`09:25 2 that's an example of a graph algorithm.
`
`09:25 3 Q What was the next major area that you,
`
`09:25 4 sort of, picked up? And I recognize -- am I
`
`09:25 5 correct that different of these sort of overlap in
`
`09:25 6 time?
`
`09:25 7 A They all -- at some point it's just one
`
`09:25 8 big mix. Right? Let's see. I think that's the
`
`09:25 9 big ones.
`
`09:25 10 Q At some point did you become involved in
`
`09:25 11 technologies relating to either audio or video
`
`09:25 12 recognition?
`
`09:25 13 A Yes. That falls under the image
`
`09:25 14 processing.
`
`09:25 15 Q I see.
`
`09:25 16 When did you start working in the image
`
`09:25 17 processing area?
`
`09:25 18 A So, for example, if you talk about the
`
`09:26 19 image processing where I do face detection and
`
`09:26 20 face recognition, that was two years, two or three
`
`09:26 21 years into Google.
`
`09:26 22 Q And over time am I understanding you
`
`09:26 23 correctly that various other aspects of image
`
`09:26 24 processing arose --
`
`09:26 25 A Yes.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 21 of 219
`
`20
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:26 1 Q -- in your research?
`
`09:26 2 A Yes, of course.
`
`09:26 3 Q Now, I mentioned the general category a
`
`09:26 4 moment ago of audio or video recognition. How do
`
`09:26 5 you understand that concept? What is that, in
`
`09:26 6 your estimation?
`
`09:26 7 A Are you talking about -- ask that
`
`09:26 8 question again.
`
`09:26 9 Q Sure.
`
`09:26 10 You mentioned that you understood that
`
`09:26 11 to be part of your work in image processing.
`
`09:27 12 A Right.
`
`09:27 13 Q Let's start with, when did you start
`
`09:27 14 having any focus in that context of audio or video
`
`09:27 15 recognition?
`
`09:27 16 A So when you say "video recognition," I
`
`09:27 17 don't actually work with videos. I work with
`
`09:27 18 images. So most of the work I did was on images,
`
`09:27 19 and images are a component of videos, right? So
`
`09:27 20 that's where I come in. And -- yes.
`
`09:27 21 Q When did you start having involvement in
`
`09:27 22 connecting that to video, if you will?
`
`09:27 23 A Probably 2006, 2005, somewhere around
`
`09:27 24 there. Probably 2006.
`
`09:27 25 Q And were you also involved in audio
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 22 of 219
`
`21
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:27 1 recognition?
`
`09:27 2 A So what's interesting is the way we did
`
`09:27 3 image recognition, we applied the same thing to
`
`09:28 4 audio.
`
`09:28 5 Q So let's talk about, first, image
`
`09:28 6 recognition then.
`
`09:28 7 A Okay.
`
`09:28 8 Q What -- can you describe to me generally
`
`09:28 9 the approach you are mentioning as to how you
`
`09:28 10 approached image recognition.
`
`09:28 11 A It's all statistics and machine
`
`09:28 12 learning, is the high-level approach. We used a
`
`09:28 13 lot of machine learning classifiers to understand
`
`09:28 14 what is in an image and did matching based on
`
`09:28 15 that.
`
`09:28 16 Q And so when you say "we used machine
`
`09:28 17 learning classifiers," give me an example of what
`
`09:28 18 you mean by that.
`
`09:28 19 A A neural network or AdaBoost.
`
`09:28 20 Q How did you use AdaBoost, this neural
`
`09:28 21 network, for this purpose?
`
`09:28 22 A You have a set of training images that
`
`09:28 23 have some particular object or image type. You
`
`09:29 24 train a machine learning classifier to be able to
`
`09:29 25 distinguish those images from other images.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 23 of 219
`
`22
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:29 1 Q So is this a system where essentially
`
`09:29 2 you provide some constellation of two or more
`
`09:29 3 images to this system and then it looks for
`
`09:29 4 mechanisms to discriminate between them and, over
`
`09:29 5 time, using multiple sets, it develops some
`
`09:29 6 mechanism to systematically do so?
`
`09:29 7 A Usually you -- sorry, there's a
`
`09:29 8 distinguishing characteristic between train time
`
`09:29 9 and test time.
`
`09:29 10 Q Okay.
`
`09:29 11 A When you train, you present millions of
`
`09:29 12 images, right. So --
`
`09:29 13 Q And -- and do you present them all as a
`
`09:29 14 group or in some pattern -- groupings or
`
`09:29 15 subformations?
`
`09:29 16 A The way machine learning works is you
`
`09:29 17 usually separate it out between the class you wish
`
`09:29 18 to distinguish and the others.
`
`09:29 19 Q And so give me an example, kind of on an
`
`09:30 20 orders of magnitude --
`
`09:30 21 A Okay.
`
`09:30 22 Q -- when you're talking about
`
`09:30 23 presenting -- I assume we're talking about still
`
`09:30 24 images, to this AdaBoost network.
`
`09:30 25 A Yes.
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 24 of 219
`
`23
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:30 1 Q Give me an example of how that, in a
`
`09:30 2 practical sense, proceeds.
`
`09:30 3 A So, for example, let's say I wanted to
`
`09:30 4 detect images that have faces in them. I will
`
`09:30 5 give it millions of examples of images with faces
`
`09:30 6 and even more billions of examples of images
`
`09:30 7 without faces, and it statistically learns to
`
`09:30 8 recognize the difference.
`
`09:30 9 Q What's the sort of output, if you will,
`
`09:30 10 of this kind of a process? What do you get from
`
`09:30 11 having the neural network perform this training?
`
`09:30 12 A Ideally, you will get a yes or no, it
`
`09:30 13 has a face.
`
`09:30 14 Q And do you -- does that provide you with
`
`09:30 15 some tool or mechanism that can then be applied
`
`09:30 16 elsewhere to essentially perform these kinds of
`
`09:30 17 recognitions, some sort of algorithm for the
`
`09:31 18 discrimination?
`
`09:31 19 A Sorry, could you explain.
`
`09:31 20 Q Well, the output -- I understand, the
`
`09:31 21 neural network, ideally you -- it is trained to
`
`09:31 22 discriminate between an image with a face and
`
`09:31 23 without a face in your example, correct?
`
`09:31 24 A Correct.
`
`09:31 25 Q Does the system also then provide you
`
`Case 1:14-cv-02396-PGG-SN Document 241-4 Filed 11/12/20 Page 25 of 219
`
`24
`
`SHUMEET BALUJA, Ph.D. - CONFIDENTIAL
`
`
`
`09:31 1 with some kind of mechanism or algorithm to
`
`09:31 2 implement on another system, or do you then simply
`
`09:31 3 have the neural network that processes queries in
`
`09:31 4 the future for further use?
`
`09:31 5 A Sorry, the reason I don't -- I can't
`
`09:31 6 answer that, I don't know what you mean by "on
`
`09:31 7 another system."
`
`09:31 8 Q Okay. So let me go back a step then.
`
`09:31 9 A Okay.
`
`09:31 10 Q What was the purpose of training the
`
`09:31 11 neural network to discriminate, in your example,
`
`09:31 12 between faces and -- images with faces and without
`
`09:31 13 faces?
`
`09:31 14 A Right. The purpose was to find images
`
`09:31 15 of the faces.
`
`09:31 16 Q But to what end?
`
`09:31 17