[00:00:00] Speaker 03: Our next case for arguing is 24-1893, Network One Technologies versus Google. [00:00:07] Speaker 02: Good morning, Your Honor, and may it please the court, Brian Liddell, for the appellant Network One. [00:00:11] Speaker 02: The district court here [00:00:14] Speaker 02: aired in several ways. [00:00:17] Speaker 02: I'd like to start with the summary judgment of non-infringement, where the district clerk very clearly, rather than applying the Supreme Court in this court's many precedents about drawing all instances in favor of the non-moving party, [00:00:32] Speaker 02: here in the district court, very clearly looked at the evidence through a jaundiced eye of trying to find reasons to discount or disregard or just ignore evidence that was inconvenient for the position that was asserted on summary judgment. [00:00:50] Speaker 03: That's like a whole nother minute on characterization. [00:00:52] Speaker 03: How about you jump to where the evidence is that created a question of fact that should have precluded summary judgment on the each issue? [00:01:00] Speaker 02: Sure. [00:01:01] Speaker 02: So, as noted in the briefs, there are two distinct systems, two versions of Google's accused system. [00:01:07] Speaker 02: The first, which I'll address, is what's referred to as the LSH system. [00:01:12] Speaker 02: LSH stands for Locality Sensitive Hashing. [00:01:15] Speaker 02: It's a particular type of search system. [00:01:17] Speaker 02: And one of the clearest pieces of evidence on that was a series of academic papers and testimony by a Google engineer, Dr. Bellugia, who explained in an academic paper, and this is a quote, LSH and other hash functions are sub-linear in the number of elements examined compared to the size of the data set. [00:01:38] Speaker 02: That's at page 9716 of the appendix. [00:01:42] Speaker 02: That's a statement about all LSH systems. [00:01:45] Speaker 02: It's not limited to a particular version or a particular flavor. [00:01:49] Speaker 02: It's about all of them. [00:01:51] Speaker 02: He had more comments in another academic paper to the same effect about the fact that when the database is increased by 50%, we see that we can have a sub-layer computation increase. [00:02:03] Speaker 02: That's at page 9702 of the appendix. [00:02:06] Speaker 02: He gave testimony in deposition confirming that understanding at pages 7362 of the appendix 7404 to 7405 of his deposition. [00:02:17] Speaker 02: Now, the district court, rather than accepting the very reasonable and, indeed, we submit only reasonable evidence or inference from that evidence that this is a characterization of all LSH systems, the district court essentially said, well, that might not be about the implemented version that Google has. [00:02:37] Speaker 02: And so as a matter of law, I'm going to say that the system is not sublinear. [00:02:43] Speaker 02: And this wasn't just a failure to consider the evidence. [00:02:47] Speaker 02: It's fundamentally just a factually wrong conclusion. [00:02:51] Speaker 02: In preparing for today's argument, I actually did something that I couldn't have done six years ago when the summary judgment was being briefed. [00:02:58] Speaker 02: I asked Google's AI, is an LSH search sub-linear? [00:03:03] Speaker 02: And it told me, yes, this is sub-linear. [00:03:08] Speaker 02: It actually said, yes. [00:03:11] Speaker 03: So you think we're going to take into account brand new evidence that you did a search, not part of the record, it was a Google product, that, wow. [00:03:27] Speaker 02: It's indicative of, but Your Honor, the point is that the evidence was very consistent [00:03:35] Speaker 02: that this is a property of these searches, and the district court chose to disregard that or find a reason to say that it was something different, that it was somehow potentially not connected. [00:03:48] Speaker 02: That may be an influence that Google can argue to a jury should be drawn from the evidence. [00:03:53] Speaker 04: But it's not the only possibility. [00:03:54] Speaker 04: Let me ask you about non-exhausted. [00:03:56] Speaker 04: We sort of exhausted the first issue. [00:03:59] Speaker 04: So let me ask you about the non-exhausted. [00:04:01] Speaker 04: This was not an original claim, that term. [00:04:05] Speaker 04: That's correct, Your Honor. [00:04:06] Speaker 04: There wasn't even a patent at all. [00:04:09] Speaker 04: Right? [00:04:09] Speaker 02: That's correct. [00:04:11] Speaker 02: The term. [00:04:11] Speaker 04: This was added to very prior law to establish patentability. [00:04:18] Speaker 04: In this show, shouldn't something that serves that purpose be cleared? [00:04:26] Speaker 04: Haley, you are doing something with non-exhausted means. [00:04:30] Speaker 02: Well, I think it was added because it correctly reflected the discussion and the specification of the invention and using the kind of search that doesn't look at every possible thing in the dataset as compared to a search that does exhaustively look at everything in the dataset. [00:04:51] Speaker 03: Okay, so we've got to break this down. [00:04:54] Speaker 03: I don't think that's what you've ever said exhaustive means, that you have to look at every single thing in the data set. [00:04:59] Speaker 03: I could be wrong, but I'm pretty sure your version of exhaustive versus non-exhaustive turned on a linear approach that begins with the first entry of the data set and ends when you have a perfect match. [00:05:12] Speaker 03: That would not require looking at every single thing in the data set [00:05:16] Speaker 03: but I think you nonetheless say that's exhaustive. [00:05:19] Speaker 03: And my misunderstanding, it would help me to understand what you're saying, exhaustive means, because I understood your patent to say exhaustive meant, at column 9 and column 22, that you take a linear approach, like if we're talking about a dictionary, you start with the word aardvark, and you keep going until you hit the word ringer, and then you stop. [00:05:39] Speaker 03: That does not mean you looked at any of the words that came after the word ringer, [00:05:44] Speaker 03: but that's still exhaustive. [00:05:46] Speaker 03: That's what I thought your patent defined it as, which is not what you just said. [00:05:50] Speaker 02: Well, I think that's close, and our construction... Which part's close? [00:05:53] Speaker 03: The patent or you? [00:05:55] Speaker 02: Well, I think... [00:05:56] Speaker 02: I think they're more similar than perhaps you're suggesting, Your Honor. [00:06:02] Speaker 02: Our construction was that it's a search designed to locate, and the claims use either a neighbor or a near neighbor, without comparing to all possible matches, i.e., all references in the reference – all records in the reference dataset, even if the search doesn't locate a neighbor or result a match. [00:06:21] Speaker 02: And so, in your example, yes, a search that – a linear search where you start at one end and keep going would be an example of an exhaustive search. [00:06:31] Speaker 02: It's not necessarily the only type. [00:06:33] Speaker 02: But that is a possible type of exhaustive search. [00:06:38] Speaker 02: And it's distinguished from other types of searches in the specification – column 9, column 21, column 22. [00:06:44] Speaker 02: where a search technology or a search approach is used that eliminates or avoids comparing the request to certain portions of the data set. [00:06:54] Speaker 02: And, of course, as the specification makes clear, the dictionary is a less than optimal example for purposes of discussing this technology, because these are not the kinds of very ordered data sets that you're talking about with a dictionary where [00:07:10] Speaker 02: You have a listing of words in an alphabetical order that's very precise. [00:07:14] Speaker 02: The specification even says the search of a dictionary isn't a... Okay, this is a mostly verbal thing. [00:07:20] Speaker 03: It does a vector instead of just a straight binary thing. [00:07:22] Speaker 03: It's not that complicated. [00:07:23] Speaker 03: Well, maybe it is, but it doesn't seem that complicated. [00:07:27] Speaker 03: Anyway, but I guess I'm... [00:07:30] Speaker 03: what's not great for you is I still don't understand your precise argument, which by the way is really bad when the whole question is indefinite. [00:07:40] Speaker 03: So I need to understand your precise argument. [00:07:44] Speaker 03: Is it that column 9 and column 22 articulate for differentiation between exhaustive on the one hand and non-exhaustive on the other? [00:07:55] Speaker 03: Such, and by the way, if you can't tell by the tone of my voice, this is a friendly question. [00:08:00] Speaker 03: I'm just going to put it out there for you. [00:08:01] Speaker 03: So it defines exhaustive and non-exhaustive in terms of whether you have to do an iterative, linear, stepwise approach as starting at the beginning and moving along until you find the answer. [00:08:17] Speaker 03: That would be exhaustive. [00:08:19] Speaker 03: Examples of non-exhaustive searches, it seems to me, are KD tree, vantage point trees, clustering, [00:08:25] Speaker 03: hashing, you know, all those things will be non-exhaustive, because that would be a whole different way of trying to reduce the universe of things to look at and make your search more efficient. [00:08:39] Speaker 03: Is that an accurate characterization of how you think your patent appropriately defines exhaustive versus non-exhaustive? [00:08:46] Speaker 02: It is, Your Honor. [00:08:47] Speaker 02: And I think that the key there is, that's how one of skill in the art would understand it. [00:08:52] Speaker 02: And that's something that comes through both from the specification itself, but also from Professor Mitzenmacher's explanation of these different techniques and what differentiates them from one another. [00:09:03] Speaker 03: Mitsunaka was a Mitsun mess for me, so I don't know where I'd go there. [00:09:08] Speaker 03: But also, I mean, don't you want to maybe focus on the claims? [00:09:13] Speaker 03: Because don't the claims do a little work of further defining what it means to be non-exhaustive? [00:09:19] Speaker 03: I mean, you mentioned the description, but don't the dependent claims kind of help? [00:09:23] Speaker 02: Yes, indeed, Your Honor. [00:09:25] Speaker 02: The dependent claims, indeed, in some instances, reference things like a KB tree as the, sort of, as a dependent claim for that being the non-exhaustive search. [00:09:36] Speaker 03: And I think here, the problem with the – Not just a KD tree, it walks through absolutely everything that was in column 9 and column 22 that differentiates between exhaustive and non-exhaustive. [00:09:50] Speaker 03: Sublinear, KD trees, vantage point trees, middle vantage point first, like it kind of goes through all those pieces and it puts them in the bucket. [00:10:00] Speaker 03: These are the non-exhaustive. [00:10:01] Speaker 02: That's right. [00:10:02] Speaker 02: That's absolutely correct. [00:10:04] Speaker 02: And we do believe that that claim differentiation further illustrates and helps to lend to that clarity of what's being differentiated within the claims here. [00:10:19] Speaker 02: step back a little bit. [00:10:21] Speaker 02: The district court's essential finding was one that I think is inconsistent with several of this court's precedents, that essentially, because there were a couple of possible, according to the district court, constructions of the phrase, that rendered an indefinite. [00:10:38] Speaker 02: That's a reasoning that this court rejected in Nevereau, in Neonode, and it gets to a question of how is one a skill in the art understanding what's taught and what's disclosed in these claims and what's disclosed in specification, which is the kind of fact question that the court has held precludes a summary judgment-like finding of indefiniteness. [00:11:00] Speaker 03: Could you help me understand what the difference is between non-exhaustive and sub-linear? [00:11:05] Speaker 03: Like, give me an example of something that is sub-linear but is not non-exhaustive. [00:11:12] Speaker 02: So I don't think there could be a sub-linear search that's not non-exhaustive. [00:11:16] Speaker 02: There could be a non-exhaustive search that is not sub-linear. [00:11:20] Speaker 02: Sub-linear would be a subset of non-exhaustive searches because you could have a non-exhaustive search that, for example, looked at every other record in a database, in a data set. [00:11:32] Speaker 04: Please, patent 65. [00:11:34] Speaker 02: They are. [00:11:34] Speaker 02: Yes, Your Honor. [00:11:35] Speaker 04: They are exhausted. [00:11:37] Speaker 02: They are expired. [00:11:39] Speaker 02: But to finish my answer, if you looked at every other reference, if you looked at half the references, that would still scale in proportion to the size of the data set. [00:11:49] Speaker 02: It would be non-exhaustive, but it would not be sub-linear. [00:11:53] Speaker 02: Does that answer your question? [00:11:54] Speaker 02: Yes. [00:11:55] Speaker 02: I'll reserve the rest of my time for rebuttal. [00:11:57] Speaker 02: I would like to address the Siberia system briefly, ultimately, but I'm happy to do so on rebuttal. [00:12:03] Speaker 03: Mr. Terrasque, wow. [00:12:12] Speaker 01: Good morning and may it please the court, Andrew Trask for Apple Ease Google and YouTube. [00:12:17] Speaker 01: If the court doesn't prefer otherwise I'll start with the non-exhaustive terms since that was the focus of the argument of my colleague. [00:12:26] Speaker 01: So the question of indefiniteness is one of claim construction ultimately and depends on the intrinsic and if needed extrinsic evidence. [00:12:35] Speaker 01: This court, of course, evaluated the same intrinsic evidence in the prior IPR appeal involving the same specification. [00:12:43] Speaker 01: And the court did not address the Phillips construction. [00:12:45] Speaker 01: That's certainly the case. [00:12:47] Speaker 01: But the court did evaluate whether the specification drew a clear line between two reasonable constructions, whether an exhaustive search must examine all data within all records or must examine just all records. [00:13:01] Speaker 03: But it did so in the context of BRI, and it expressly said it was not making any decisions on indefiniteness. [00:13:10] Speaker 01: That is absolutely true. [00:13:12] Speaker 03: I guess I'm a little... I appreciate the language that you're pointing to in the prior opinion. [00:13:19] Speaker 03: I'm just not sure it prevents me from looking at the intrinsic evidence [00:13:25] Speaker 03: through the lens of a skilled artisan in this case. [00:13:28] Speaker 03: I mean, do you think it's, I know it's been your position in part, but do you think it's the law of the case? [00:13:34] Speaker 03: Because even the district court found it wasn't in that giant long footnote number 12. [00:13:37] Speaker 03: Even he found that he was not bound, that that prior decision of ours didn't actually reach any kind of binding decision on this. [00:13:47] Speaker 03: So are you saying it's the law of the case and I don't have a choice? [00:13:49] Speaker 03: I can't, I absolutely am bound by it and can't go back and look at that? [00:13:53] Speaker 01: I think if the court wants to revisit the question, it could do that. [00:13:57] Speaker 01: I think it would be appropriate for the only proper course would be for the court to reach the same conclusion, however. [00:14:02] Speaker 01: And that's because the question is the same, ultimately, with respect to whether the specification draws a clear line between [00:14:10] Speaker 01: a non-exhaustive and exhaustive search with respect to how much data must be examined within each record. [00:14:16] Speaker 01: The passages that you're on are focused on, columns 9 and 22, are the same passages that were the focus of this court's prior decision. [00:14:26] Speaker 01: And those passages do refer to specific types of searches, linear searches, KD trees, vantage point trees, but there's nothing in those [00:14:36] Speaker 01: passages describes any of those searches as exhaustive or non-exhaustive. [00:14:41] Speaker 01: The claims do. [00:14:42] Speaker 03: The claims do. [00:14:43] Speaker 03: The claims are part of the specification. [00:14:44] Speaker 03: The claims, these claims were... The claims expressly do. [00:14:49] Speaker 03: Claim 17, method of claim 15 wherein the non-exhausted search is sub-linear. [00:14:54] Speaker 03: 18, the method of claim 15 wherein the non-exhaustive search is based on a KD tree. [00:14:59] Speaker 03: Claim 19, non-exhaustive, I can keep going. [00:15:02] Speaker 03: I understand. [00:15:03] Speaker 03: Everyone, so you can't say this specification because specification as you know includes claims. [00:15:08] Speaker 03: This specification absolutely does, clearly and unequivocally tie, and by the way, our prior decision didn't look at any of the dependent claims. [00:15:17] Speaker 01: Two points in response to that, Chief Judge Moore. [00:15:19] Speaker 01: So one is that the – first of all, just as a matter of fact, these claims were not part of the original specification. [00:15:24] Speaker 01: They were added years later. [00:15:26] Speaker 01: But more fundamentally, those dependent claims do not tell you – do not tell the person with the ordinary skill how much data within each record needs to be examined in order for a search to be exhaustive or non-exhaustive. [00:15:39] Speaker 01: So I agree with you. [00:15:40] Speaker 01: Those dependent claims do – [00:15:43] Speaker 01: specified types of searches that depend from an independent claim that recites a non-exhausted search. [00:15:48] Speaker 01: But as to the key question here, which is the same question that this court grappled with in the prior appeal, how much data is enough? [00:15:55] Speaker 01: How much do you have to examine for a search to be exhaustive? [00:15:59] Speaker 03: Well, I mean, doesn't Column 9 tell us? [00:16:01] Speaker 03: Like, it explains that Column 9, Line 24, in previous work, it was not uncommon to form a linear search of N entries [00:16:13] Speaker 03: perhaps halted research when the first match is found. [00:16:17] Speaker 03: So that basically, and that says if n is large, this search would be computationally very expensive. [00:16:23] Speaker 03: Why isn't that on its face the definition of exhaustive? [00:16:28] Speaker 03: Because when it says other forms of matching include, and it goes through all the ones that are clearly defined as non-exhaustive by the claims, [00:16:38] Speaker 03: the defendant claims. [00:16:40] Speaker 03: And that's kind of how, to be honest, it's kind of how I understood this back when I read it, that they were defining exhaustive as the linear processing item by item of each and every item until you got to the match, not what he started to say when he stood up here before, which is every item in the whole N, because the whole N is the dictionary, then you wouldn't stop when you got to the word regular, you'd go all the way to whatever the last word is. [00:17:05] Speaker 03: you know, at the Z. But at the Z, where is this last? [00:17:07] Speaker 03: So, you know, I think, but I think the spec does. [00:17:12] Speaker 01: So that's... So there's, I appreciate the question, Your Honor. [00:17:15] Speaker 01: I think there's two layers to this. [00:17:16] Speaker 01: One is, what are the differences between, what searches are exhaustive versus non-exhaustive? [00:17:22] Speaker 01: And I understand your honor is saying that it could be read from the specification as to which is which. [00:17:26] Speaker 01: Specification doesn't say that, to be clear. [00:17:28] Speaker 01: It doesn't use the term. [00:17:29] Speaker 01: But even assuming that's right, there's a second layer to this. [00:17:32] Speaker 03: Specification does use the term because they use it in the claims, Mr. Triask, in the claims of private specification. [00:17:38] Speaker 01: Even accepting that, Your Honor, there's a second layer to this, which is even if one of the ordinary skill in the art could determine which searches qualify as exhaustive or non-exhaustive, it still doesn't tell the person of ordinary skill how much data within the particular search [00:17:54] Speaker 01: needs to be examined. [00:17:55] Speaker 01: So for example, is a linear search of all n entries exhaustive or non-exhaustive if it only examines the first letter of each character string? [00:18:04] Speaker 01: And that was the example that this court addressed in the prior appeal. [00:18:07] Speaker 01: If you only examine the first letter, of course you're going to miss some matches if you're looking for approximate matches where the data is towards the end of the string. [00:18:14] Speaker 01: Is that an exhaustive or non-exhaustive search? [00:18:16] Speaker 01: There's nothing in the [00:18:18] Speaker 01: claims or column 21 or 22 anywhere else in the patent that answer that question. [00:18:24] Speaker 01: And that's why this is fundamentally an indefinite term. [00:18:27] Speaker 01: There's no guidance in the intrinsic record. [00:18:29] Speaker 03: I certainly understand the point that you're making now, but I mean, I think it says a linear search of N entries. [00:18:39] Speaker 03: I could be mistaken, but to me that means you start at aardvark, whatever comes after aardvark, you go to next, even if you know your first letter is B. So an exhaustive search, it seems to me, is anything other than stepwise, entry by entry, all the way through N until you get to a match. [00:18:57] Speaker 01: But there's a dispute between the parties here, Your Honor, as to how much of that entry needs to be examined. [00:19:02] Speaker 01: Is it just the A or the B, or is it the entire entry, if it's a text string, for example? [00:19:08] Speaker 01: The example you're on again doesn't answer that question. [00:19:11] Speaker 03: Well, that's just because you're getting the difference between the simplicity of our example and the vector concept of the nearby, right? [00:19:17] Speaker 01: Well, this is the dispute between the parties in this case. [00:19:19] Speaker 01: It's a live dispute that needs to be resolved, and there's simply no evidence of record. [00:19:23] Speaker 01: And in fact, the expert of Network One [00:19:26] Speaker 03: Are you pivoting now to infringement, or are you still on? [00:19:29] Speaker 01: I just wanted to- No, I'm staying with exoticness, although I'm happy to pivot if Your Honor would prefer. [00:19:33] Speaker 01: But, you know, staying with the dictionary example that Your Honor raised, the expert from Network One- Oh, trust me. [00:19:41] Speaker 01: Trust me. [00:19:41] Speaker 01: Trust me. [00:19:42] Speaker 01: Trust me. [00:19:42] Speaker 01: Trust me. [00:19:43] Speaker 01: I read it. [00:19:44] Speaker 01: Right? [00:19:44] Speaker 03: And it's still- It's all over the place. [00:19:47] Speaker 03: I was exhausted reading his transcript. [00:19:50] Speaker 01: It was, yeah, it was an interesting deposition. [00:19:53] Speaker 01: But I think, Your Honor, the point is that there is... But that's right. [00:19:56] Speaker 03: The good news for me is I don't have to look at extrinsic evidence. [00:19:59] Speaker 03: If I think the spec is clear. [00:20:01] Speaker 03: That's right. [00:20:03] Speaker 01: That's right. [00:20:04] Speaker 01: But let's look at Network One's argument that, for example, there's a number of papers that are part of the intrinsic record, right? [00:20:10] Speaker 01: The Yiannilos papers and others. [00:20:13] Speaker 01: Those themselves are inconsistent about which is which, right? [00:20:16] Speaker 01: The Yiannilos 1 paper [00:20:18] Speaker 01: which is at A1400, says that a vantage point forest may be an exhaustive search, but it's Network 1's position that that's a non-exhaustive search. [00:20:26] Speaker 01: And the ANALOS 2 paper, which is at A1413, describes KD Trees as performing a nearly exhaustive search, but again, in Network 1's view, that's an exhaustive search. [00:20:35] Speaker 01: But that's a non-exhaustive search. [00:20:38] Speaker 01: So even intrinsic evidence is all over the place here, Your Honor. [00:20:41] Speaker 01: There's no way to draw a line between the two. [00:20:44] Speaker 01: And that's exactly why the Court found as it did in its prior decision, based on the same specification and the same underlying question with respect to what does the specification teach in terms of how to draw the line. [00:20:56] Speaker 01: It's simply a subjective inquiry. [00:20:58] Speaker 03: Network One argues... Mr. Trask, I'd love to send you to Siberia now. [00:21:01] Speaker 01: But we can talk about Siberia, Your Honor. [00:21:03] Speaker 03: That wasn't meant to be, you know, kind of being funny, but... Happy to discuss... Can we jump to Siberia? [00:21:11] Speaker 01: Of course. [00:21:11] Speaker 03: Yeah, I'll put my park on and go to Siberia. [00:21:14] Speaker 01: Of course. [00:21:15] Speaker 01: So Siberia is one of the two accused systems here, and... [00:21:20] Speaker 01: The District Court granted summary judgment of non-infringement, and it did so on several bases, one of which was, I think quite importantly here, agreement between the witnesses of both parties, right? [00:21:31] Speaker 01: So we had a Google witness, Ms. [00:21:34] Speaker 01: Pazula or Dr. Pazula, who testified that Siberia always searches a fixed portion of the index, and that means it scales proportionally and it's linear, not sub-linear. [00:21:44] Speaker 01: Dr. Misimaka at A6616, this is paragraph 229 of his declaration, he said, I generally agree with this notion. [00:21:52] Speaker 01: In other words, that the scanned portion of the search would scale linearly. [00:21:56] Speaker 01: So we have agreement between the witnesses that Siberia, as implemented, is not a sub-linear system and therefore does not meet the asserted claims. [00:22:05] Speaker 03: Okay, so what about 10265? [00:22:06] Speaker 01: 10265. [00:22:12] Speaker 03: This is a Google document. [00:22:13] Speaker 03: It's not confidential. [00:22:15] Speaker 03: So I'm going to be very careful about what I say. [00:22:17] Speaker 03: I think I know which part is confidential, and I don't think I'm going to reveal anything that is. [00:22:22] Speaker 03: But this document is talking about the Siberia system. [00:22:25] Speaker 03: 10265. [00:22:31] Speaker 03: It's a planning document, but planning for Siberia. [00:22:40] Speaker 01: Oh, yes, Your Honor. [00:22:41] Speaker 03: And it says, it is clear we will need a customized distributed searching service and a sublinear search with hash representation. [00:22:51] Speaker 03: It is clear we will need [00:22:54] Speaker 03: That doesn't mean Sublinear for sure definitely has. [00:22:58] Speaker 03: But this is a question of fact. [00:23:00] Speaker 03: And there's a Google document in the planning phases to develop this. [00:23:04] Speaker 03: And this Google document said, we have to have a sublinear search. [00:23:08] Speaker 03: Why isn't that enough to create a question of fact about whether the ultimate system does have a sublinear? [00:23:14] Speaker 01: Because it's not evidence that the system, as implemented, works in a sub-linear way under the party's construction. [00:23:21] Speaker 03: No, it says it has to have it, needs it, not just like, hey, maybe let's think about implementing this. [00:23:25] Speaker 03: It needs it, it must have it. [00:23:27] Speaker 01: Well, this is one of two documents cited by Network One, and together, I think they both indicate that there were various ways that this system could have been designed, right? [00:23:37] Speaker 03: Even if that's true. [00:23:38] Speaker 03: Question of fact, summary judgment. [00:23:40] Speaker 03: In fact, they've got a Google Doc that says, we're planning to develop a system. [00:23:45] Speaker 03: It has to have sublinear. [00:23:48] Speaker 03: Must have it. [00:23:49] Speaker 01: I think this court's cases support the district court's decision to... District court didn't address this document, and it was raised by them. [00:23:58] Speaker 03: I tracked it. [00:23:59] Speaker 03: It was raised. [00:23:59] Speaker 03: It was cited. [00:24:00] Speaker 03: District court never addressed this document. [00:24:03] Speaker 01: Well, Your Honor, this court's case is presumed that the district court analyzed the evidence presented by the parties. [00:24:10] Speaker 01: And in the EPAS versus 3Com case, which is cited in the brief, there was similar evidence. [00:24:15] Speaker 01: There were internal documents with plans and proposals for how to implement the accused system, and the court found that was not enough because it didn't implement, excuse me, it didn't demonstrate how the accused system actually functioned. [00:24:28] Speaker 01: So there was no genuine issue of material fact. [00:24:30] Speaker 01: This is no different from that. [00:24:31] Speaker 01: This is a planning document. [00:24:33] Speaker 01: It doesn't tell, it doesn't say anything about how the system was ultimately implemented. [00:24:37] Speaker 01: And as Network1's own evidence, the [00:24:39] Speaker 01: demonstrated, there were other ways that the system could have been implemented. [00:24:43] Speaker 01: There are various algorithms cited. [00:24:45] Speaker 01: None of the evidence actually demonstrates how the system was actually implemented other than the agreement between the witnesses, which states [00:24:54] Speaker 01: which by the way is focused specifically on the acute system, and the parties agree that as designed, it scales linearly. [00:25:00] Speaker 01: So the hypothetical possibility of... I don't even understand that. [00:25:03] Speaker 03: Like, I don't think the parties do agree that as designed, it scales linearly. [00:25:07] Speaker 03: I think it was in fact designed to scale in a sublinear search fashion. [00:25:13] Speaker 03: I don't follow you at all. [00:25:15] Speaker 03: What about Kumar's testimony? [00:25:17] Speaker 03: Kumar's testimony, I don't think, [00:25:18] Speaker 03: It says, the tree AH was used in the Siberia content ID. [00:25:24] Speaker 01: So the tree AH, Dr. Kumar explained that the tree AH algorithm is a category of algorithms. [00:25:29] Speaker 01: It's not one specific algorithm. [00:25:30] Speaker 03: And there's nothing in the specific algorithm. [00:25:34] Speaker 01: And that's it. [00:25:35] Speaker 03: This is semi-judgment. [00:25:36] Speaker 03: We know that the tree, we know, this tells us that the tree AH is Siberian. [00:25:43] Speaker 03: Then we have, and that's a Google Doc, then we have a Google Witness that says the tree AH is used in Siberia. [00:25:51] Speaker 03: How can that not be enough to survive summary judgment? [00:25:54] Speaker 01: Dr. Kumar was not testifying about the document with that graph shown in it. [00:25:58] Speaker 01: Dr. Kumar was testifying about it was deposition exhibit seven. [00:26:01] Speaker 01: That document's not even in the joint appendix because Network One never cites it on appeal. [00:26:05] Speaker 01: Network One, that document with the graph in it that refers to that algorithm was never shown to any witness in this case. [00:26:11] Speaker 01: It's simply pointed to by Network One and there's testimony saying that the algorithms, the first, the bottom algorithm there, their expert admitted at 6619 [00:26:21] Speaker 01: that Siberia does not use that algorithm. [00:26:23] Speaker 01: In the other algorithm, Dr. Kumar testified at A9289 that it's a category of algorithms and there's nothing time to actually... But the jury can make an inference. [00:26:32] Speaker 03: Like the jury can make an inference based on that and that's all they have to do to survive summary judgment. [00:26:36] Speaker 01: I don't think the jury can make a fair inference when... [00:26:40] Speaker 01: Google's witness testified unequivocally dr. Pazula a 7536 do you always search a fixed pressure of the index of Siberia and Network wins expert agreed with that point the only way that they can argue Sublinearity of Siberia is if it's the system is change if parameters are changed, but that's not Didn't you all do it once didn't you admit that you even did this once I [00:27:05] Speaker 01: that one parameter was changed, but there's no dispute that it nonetheless ran a linear algorithm. [00:27:10] Speaker 03: So here's my problem. [00:27:11] Speaker 03: Let's think of it like a radio station, right? [00:27:13] Speaker 03: You can turn the knob and you can go to different stations. [00:27:18] Speaker 03: And your system was designed to do that. [00:27:21] Speaker 03: It was designed to be able to turn the knob to go to different proportions. [00:27:26] Speaker 03: That's what it was designed to do. [00:27:27] Speaker 03: Your argument and your witness' testimony is [00:27:31] Speaker 03: When you turn the knob and you take your hand off the knob, from that point on, it's proportional. [00:27:36] Speaker 03: It's on only one station. [00:27:37] Speaker 03: There's only one radio station at that point in time, and so you're trying to step back and say, therefore, our system is not designed to allow you to do multiple stations, because when we're not touching the knob, you can only hear the one that's playing. [00:27:50] Speaker 03: That is how I understand this case. [00:27:54] Speaker 01: So what's wrong with it? [00:27:55] Speaker 01: I see my red light on. [00:27:56] Speaker 03: May I answer it? [00:27:57] Speaker 03: Mr. Trask, I'm asking you lots of questions. [00:27:59] Speaker 01: Keep going. [00:28:01] Speaker 01: So just to be clear, and I'm sure this is clear to your honor, that the system itself doesn't have knobs. [00:28:05] Speaker 01: It's kind of a metaphor. [00:28:06] Speaker 01: You know that. [00:28:07] Speaker 03: It also doesn't involve a dictionary or a book cannot be present. [00:28:12] Speaker 01: That's all correct. [00:28:12] Speaker 01: What's actually happening here when that parameter is changed is that a human engineer is going into the code and changing the code so that there's a different system. [00:28:21] Speaker 01: Now, that different system is also linear, because as Dr. Pazula testified, the system always scans a fixed portion of the index. [00:28:31] Speaker 03: But this is expressly designed to allow it to be modified. [00:28:36] Speaker 03: That was an important component of it. [00:28:39] Speaker 03: I don't know. [00:28:41] Speaker 03: I have no problem with you guys winning at trial, but really just... [00:28:46] Speaker 01: This case is similar, in fact, easier, I think, than the Telemac versus Top case, which is one of these cases about modifying the system. [00:28:53] Speaker 01: So that was a case where the source code included infringing functionality, but it wasn't enabled. [00:29:01] Speaker 01: And this quote said, well, because you'd have to enable that functionality, summary judgment of non-infringement was appropriate. [00:29:07] Speaker 01: Here, this case is easier because there is no code. [00:29:09] Speaker 01: Networking is pointed to no code. [00:29:11] Speaker 01: demonstrating why the system would grow, necessarily grow. [00:29:16] Speaker 01: In fact, at page 50 of their blue brief, Network One admitted that their expert, quote, was not generally replying that the number of elements examined would necessarily grow with any increase in reference size. [00:29:28] Speaker 01: So that's an admission that the system as designed does not grow together with the size of the reference index. [00:29:35] Speaker 01: This is a, if anything, a human implemented change, but it doesn't change the algorithm from the standpoint of being [00:29:40] Speaker 01: linear algorithm. [00:29:43] Speaker 01: I didn't get to the LSH system at all and I'm certainly out of time at this point. [00:29:47] Speaker 01: So if no other, we obviously stand on the briefing there with respect to why that system doesn't infringe as well and summary judgment is appropriate. [00:29:56] Speaker 01: So for all the reasons explained in the briefing and today, I'd urge that the court affirm the district court's judgment. [00:30:01] Speaker ?: Thank you. [00:30:16] Speaker 02: Well, I won't spend a lot of time on Siberia, because I think Your Honor made many of the points I wanted to raise in your questions. [00:30:23] Speaker 02: And I'm happy to address the reverse of those, if you prefer. [00:30:27] Speaker 02: But I do think that the court correctly described that there was a document the district court didn't address that clearly said it was a necessary. [00:30:35] Speaker 00: Why don't you address where the tunable knob makes this whole system sublinear? [00:30:44] Speaker 02: Well, Your Honor. [00:30:46] Speaker 02: Sublinearity refers to how the system functions as you change the size of the data set. [00:30:51] Speaker 02: So it's not really only looking at a single search, it's how it scales. [00:30:57] Speaker 02: The computation resources don't increase at a proportional rate to the size of the data set increasing. [00:31:04] Speaker 02: And that's exactly what these tunable knobs are addressed to, is scaling the system. [00:31:10] Speaker 02: So that the system is adjustable as it grows to maintain the sublinearity of the computational resources. [00:31:18] Speaker 02: And as we pointed out, as I think Your Honor's question raised, Google, in fact, did do exactly what was postulated. [00:31:26] Speaker 03: In order to save resources. [00:31:28] Speaker 02: Exactly. [00:31:28] Speaker 03: And that was the admission. [00:31:30] Speaker 03: Exactly. [00:31:30] Speaker 03: They did it in order to save resources. [00:31:33] Speaker 03: Your argument, I think, is why would it be unreasonable for a jury to infer the system was designed to allow someone to tune it to save resources? [00:31:41] Speaker 02: That's exactly correct, Your Honor. [00:31:42] Speaker 02: That's a perfectly reasonable inference. [00:31:45] Speaker 02: I understand you can argue a different inference, but we submit that the district court can't resolve that question on summary judgment. [00:31:53] Speaker 02: And I think that's the error here, is stepping across that line to weigh the [00:32:00] Speaker 02: the different possible inferences and to pick one as opposed to acknowledging that that's a reasonable inference that a jury could draw. [00:32:08] Speaker 02: And I think the same problem applies to the LSH system. [00:32:11] Speaker 02: I cited one of the pieces of evidence. [00:32:13] Speaker 02: There were other documents and other pieces of Google material that [00:32:18] Speaker 02: made clear that the LSH system was also sub-linear. [00:32:21] Speaker 03: I don't think you have such a good case on that. [00:32:23] Speaker 03: So you can either spend your time on that or you can touch on exhausted because you got like a minute. [00:32:26] Speaker 03: I mean I'll give you a minute or so extra because Mr. Trask had extra but you can choose to. [00:32:32] Speaker 03: Sure. [00:32:34] Speaker 02: So I mentioned Dr. Ballouge's testimony that LSH systems and LSH algorithms are, by their nature, sublinear. [00:32:42] Speaker 02: Dr. Mitsumaka also explained that an LSH system uses what's called an inverted index, and that a property of such an inverted index is that it is a sublinear data structure. [00:32:54] Speaker 02: Google also had produced a document about an addition to the LSH system called CoverCat. [00:33:00] Speaker 02: And that addition was to use functionally the same basic approach, the same basic matching system. [00:33:08] Speaker 02: And in fact, the document said. [00:33:09] Speaker 03: Is this that document you all never cited to the district court at all during the summary judgment proceedings? [00:33:14] Speaker 02: It was cited to the district court, Your Honor. [00:33:19] Speaker 03: Matthew? [00:33:20] Speaker 03: Was it cited to the district court during the summary judgment? [00:33:22] Speaker ?: It was cited. [00:33:28] Speaker 02: In that document, Your Honor, Google wrote that it's that system that was being added follows the same high-level architecture as the Content ID fingerprint system. [00:33:41] Speaker 02: That's at page 9380. [00:33:42] Speaker 02: And it also stated that where there are diversions from Content ID, they're going to be noted in the document. [00:33:49] Speaker 02: That's at page 9382 of the appendix. [00:33:53] Speaker 02: And then it goes on to explain that the LSH system scales in a sub-linear fashion at page 9387. [00:34:01] Speaker 02: The district court suggested there was a way to read that document that it maybe could be underscored not to apply to the system as implemented. [00:34:10] Speaker 02: We submit that, first of all, that's not the proper inference from the document, but it's certainly not the only inference that can be drawn from a document like that. [00:34:18] Speaker 03: And where that's... I think the big difference between this and the severely case for me is you don't have a document like you did at 10-265 where Google said this is necessary. [00:34:30] Speaker 03: We have no idea whether Google implemented in the LSH area what we have for Siberia that they say necessary. [00:34:40] Speaker 03: Gotta have this. [00:34:41] Speaker 03: Gotta have this in Siberia. [00:34:42] Speaker 03: Bring your parka. [00:34:43] Speaker 02: I think that the distinction is, Your Honor, that with LSH, [00:34:47] Speaker 02: Sub-linearity is a property of, LSH is a category of search, and categorically that kind of search is sub-linear in all instances. [00:34:57] Speaker 02: That's what Google's Witnesses testified to, like Dr. Bellugia. [00:35:01] Speaker 02: That's what the documents show. [00:35:05] Speaker 02: Lest there had been any confusion, Professor Mittenmacher also looked at the implementation and their source code to confirm that there was nothing about it that changed that in some way in the implemented system. [00:35:16] Speaker 02: But the testimony, at a minimum, can be understood to be that all LSH systems are sublinear. [00:35:24] Speaker 02: And the district court had to at least credit that as a reasonable inference to draw from the evidence. [00:35:31] Speaker 03: Okay, I thank both counsels. [00:35:34] Speaker 03: This is our communistic and under-support time.