1 00:00:00,012 --> 00:00:03,729 Perhaps [INAUDIBLE] perhaps in a, in a, in a couple years back, right? 2 00:00:03,729 --> 00:00:04,844 >> Yeah, one and half years back. 3 00:00:04,844 --> 00:00:09,073 >> One and half years and he's one of 4 00:00:09,073 --> 00:00:14,373 the I guess school mentors of, of Markov logic. 5 00:00:14,373 --> 00:00:14,802 [INAUDIBLE]. 6 00:00:14,802 --> 00:00:16,580 >> right, I learned to [INAUDIBLE] but maybe the 7 00:00:16,580 --> 00:00:19,490 first person who had influence in learning on that, yeah. 8 00:00:19,490 --> 00:00:23,690 He cares about this interesting area of integrating. 9 00:00:23,690 --> 00:00:24,320 [COUGH]. 10 00:00:24,320 --> 00:00:25,790 >> Logic and statistics. 11 00:00:27,080 --> 00:00:29,450 So welcome, thanks for visiting >> Thanks Gautham. 12 00:00:29,450 --> 00:00:30,370 Yeah. 13 00:00:30,370 --> 00:00:33,222 So, so this topic is going to be about Markov logic and I 14 00:00:33,222 --> 00:00:37,550 see some people, who probably would have at least parts of this talk. 15 00:00:37,550 --> 00:00:40,260 before, but hopefully there'll be something new, new for you also. 16 00:00:40,260 --> 00:00:43,264 And so here is the brief outline of the doc. 17 00:00:43,264 --> 00:00:45,830 so, I'll give some motivation for the language, right? 18 00:00:45,830 --> 00:00:47,580 That Markov logic that will propose. 19 00:00:47,580 --> 00:00:49,549 The background, of some of the languages which have 20 00:00:49,549 --> 00:00:52,460 been proposed earlier, what is the relationship within them. 21 00:00:52,460 --> 00:00:54,504 Then I'll formally define the language and I'll, 22 00:00:54,504 --> 00:00:57,150 describe some inference and learning [UNKNOWN] to that. 23 00:00:57,150 --> 00:01:00,240 Describe two applications that are well known, using Markov logic. 24 00:01:00,240 --> 00:01:02,466 And there are plenty of them, that people have, 25 00:01:02,466 --> 00:01:05,110 all over the world, in fact, using Markov logic. 26 00:01:05,110 --> 00:01:09,232 And, some future directions, and then conclude the talk. 27 00:01:09,232 --> 00:01:09,770 Okay. 28 00:01:09,770 --> 00:01:12,190 So, I'll try to give you two different kinds 29 00:01:12,190 --> 00:01:15,310 of motivations for, for the frame that has been proposed. 30 00:01:15,310 --> 00:01:15,620 Right? 31 00:01:15,620 --> 00:01:18,100 So this is the first, first motivation. 32 00:01:18,100 --> 00:01:19,860 So ideas of interface layer. 33 00:01:19,860 --> 00:01:23,770 Right, so if you think about any area [UNKNOWN] then you can think about 34 00:01:23,770 --> 00:01:26,150 applications being developed at one, at one 35 00:01:26,150 --> 00:01:27,965 friend and also you can think about infrastructure. 36 00:01:27,965 --> 00:01:31,040 And I'll use some examples and typically what people have 37 00:01:31,040 --> 00:01:33,809 seen is that, initially when an area sort of comes. 38 00:01:35,030 --> 00:01:38,410 In, fashion then, you have all these n-square combinations. 39 00:01:38,410 --> 00:01:40,514 There are n, maybe n different applications 40 00:01:40,514 --> 00:01:41,952 that you need to worry about and 41 00:01:41,952 --> 00:01:43,446 maybe there are n different so of 42 00:01:43,446 --> 00:01:46,180 infrastructure companies that you need to worry about. 43 00:01:46,180 --> 00:01:48,537 And for every little developing at the lowest level you need 44 00:01:48,537 --> 00:01:52,150 to worry about what application, might, how that application might be affected. 45 00:01:52,150 --> 00:01:54,610 So you have to worry about all this, cross connections. 46 00:01:54,610 --> 00:01:56,153 And all the time what people realizes 47 00:01:56,153 --> 00:01:58,460 that, maybe there's an interface layer that. 48 00:01:58,460 --> 00:02:01,442 Separates these two, then application layer can, really talk to 49 00:02:01,442 --> 00:02:05,020 interface layer and then interface layer can talk to infrastructure, right? 50 00:02:05,020 --> 00:02:07,627 So this separation really helps instead of having order n-square 51 00:02:07,627 --> 00:02:10,800 combinations, you will have now, linear number of combinations, right? 52 00:02:10,800 --> 00:02:14,170 So, just to give an example that will, make it clearer. 53 00:02:14,170 --> 00:02:16,166 If you think about networking, you have, 54 00:02:16,166 --> 00:02:18,570 applications like www, email and so, right? 55 00:02:18,570 --> 00:02:22,990 And infrastructure level, contains protocols, routers, and all such things. 56 00:02:22,990 --> 00:02:27,098 Now, after some time of, you know, thinking about networking, people thought 57 00:02:27,098 --> 00:02:28,944 that, maybe there is something like 58 00:02:28,944 --> 00:02:31,690 Internet, which really serves as interface layer. 59 00:02:31,690 --> 00:02:33,958 So what I mean by that, is that, once 60 00:02:33,958 --> 00:02:37,400 you've built to the, sort of the interface, how. 61 00:02:37,400 --> 00:02:39,072 What the protocols are or what the 62 00:02:39,072 --> 00:02:41,748 how you should, the application should be communicating. 63 00:02:41,748 --> 00:02:42,160 [COUGH]. 64 00:02:42,160 --> 00:02:45,150 >> With the internet then you don't need to worry about what exactly, 65 00:02:45,150 --> 00:02:47,142 how exactly the Routers, are implemented or 66 00:02:47,142 --> 00:02:49,300 what is happening at the infrastructure level. 67 00:02:49,300 --> 00:02:51,725 Let all the application be developed, almost independently 68 00:02:51,725 --> 00:02:53,600 of what is happening at the lowest level. 69 00:02:53,600 --> 00:02:56,110 Similarly, someone who works at the infrastructure level doesn't 70 00:02:56,110 --> 00:02:58,490 have to worry about exactly what applications are been. 71 00:02:58,490 --> 00:02:59,290 Implement it, right? 72 00:02:59,290 --> 00:03:00,876 As long as they are true, to the 73 00:03:00,876 --> 00:03:04,180 interface, to the middle layer, everything is good, right? 74 00:03:04,180 --> 00:03:05,610 So that is right, yah? 75 00:03:05,610 --> 00:03:07,672 So similarly, when you think about databases, 76 00:03:07,672 --> 00:03:10,086 you have various things like, ERP, online transaction 77 00:03:10,086 --> 00:03:12,752 processing systems, CRMs, and many such applications and, 78 00:03:12,752 --> 00:03:16,590 the infrastructure could be query optimization, transaction management,. 79 00:03:16,590 --> 00:03:20,240 And various things, you know, where you really work at the backend. 80 00:03:20,240 --> 00:03:21,700 And what is the interface layer? 81 00:03:21,700 --> 00:03:24,220 Well, people could argue various sort of things 82 00:03:24,220 --> 00:03:26,180 like maybe relational model or XML but for 83 00:03:26,180 --> 00:03:28,308 large part you could say maybe the relational 84 00:03:28,308 --> 00:03:31,300 model is something which serves as an interface layer. 85 00:03:31,300 --> 00:03:33,418 And as, long as you are true to that schemer, how 86 00:03:33,418 --> 00:03:34,970 we talk with the relational model, 87 00:03:34,970 --> 00:03:37,260 the applications can be developed independently. 88 00:03:37,260 --> 00:03:38,130 They can be optimized. 89 00:03:38,130 --> 00:03:39,995 They can come up with endless applications, 90 00:03:39,995 --> 00:03:42,070 without really worrying about the lowest level. 91 00:03:42,070 --> 00:03:45,129 Similarly people at the lowest level can't really optimize and 92 00:03:45,129 --> 00:03:48,370 come up with new algorithms, in [UNKNOWN] or whatever that is. 93 00:03:48,370 --> 00:03:50,691 Without really affecting the top level of, applications. 94 00:03:50,691 --> 00:03:51,191 Right? 95 00:03:52,210 --> 00:03:53,980 And similarly for A.I. right? 96 00:03:53,980 --> 00:03:54,120 . 97 00:03:54,120 --> 00:03:56,370 So, for A.I. there are a number of applications. 98 00:03:56,370 --> 00:03:59,650 Robotics, vision, NLP, planning, multi-agent systems. 99 00:03:59,650 --> 00:04:03,276 And infrastructure of course, corresponds to knowledge presentation, how do you 100 00:04:03,276 --> 00:04:06,350 do learning, to do inference, right, how can you make it efficient. 101 00:04:06,350 --> 00:04:09,373 And all those questions, are, are there, and, over 102 00:04:09,373 --> 00:04:13,360 last you know, 50, 60 years, people have heard about. 103 00:04:13,360 --> 00:04:14,665 Since that went of A.I. right? 104 00:04:14,665 --> 00:04:17,560 People have talk about what could be the interface layer. 105 00:04:17,560 --> 00:04:20,544 And then, like, one of the choices that, people 106 00:04:20,544 --> 00:04:24,750 talked was, maybe first-order logic was the right choice, right? 107 00:04:24,750 --> 00:04:25,520 Like [UNKNOWN]. 108 00:04:25,520 --> 00:04:26,090 Why so? 109 00:04:26,090 --> 00:04:29,740 Because it's very powerful and it can present entities, relationships, and. 110 00:04:29,740 --> 00:04:32,305 When we talk about high human level intelligence then 111 00:04:32,305 --> 00:04:34,585 it really is very compact, right, you can talk 112 00:04:34,585 --> 00:04:37,036 about, groups of people at one, time and then 113 00:04:37,036 --> 00:04:40,125 you can talk about the properties, infractions, or all that. 114 00:04:40,125 --> 00:04:40,470 Right. 115 00:04:40,470 --> 00:04:43,496 But then, all the time people realize that it is a problem with this, 116 00:04:43,496 --> 00:04:47,070 and the problem is that there is no explicit way to handle [UNKNOWN], right? 117 00:04:47,070 --> 00:04:49,690 There is no way to incorporate the fact that. 118 00:04:49,690 --> 00:04:51,820 The, rules that you have are noisy. 119 00:04:51,820 --> 00:04:52,082 Right? 120 00:04:52,082 --> 00:04:54,910 And that's, almost always true in the real world. 121 00:04:54,910 --> 00:04:56,420 So that did not really fly. 122 00:04:56,420 --> 00:04:59,987 Of course, first-order logic is still, very important languages, but 123 00:04:59,987 --> 00:05:02,169 there are issues in terms of if you have noisy data 124 00:05:02,169 --> 00:05:03,976 or, if you want to have ex, if you want to 125 00:05:03,976 --> 00:05:08,260 have explicit motion of probability, then, it doesn't really handle that. 126 00:05:08,260 --> 00:05:10,921 Last 15 or 20 years or maybe, at least, last 10, 15 127 00:05:10,921 --> 00:05:12,382 years have seen lot of, progress 128 00:05:12,382 --> 00:05:14,860 in something called statistical machine learning. 129 00:05:14,860 --> 00:05:15,040 Right. 130 00:05:15,040 --> 00:05:17,260 So the methods have become very, very well founded. 131 00:05:17,260 --> 00:05:21,792 And one set of models which has become really popular is called graphical models. 132 00:05:21,792 --> 00:05:24,010 Things like Bayesian networks or Markov networks. 133 00:05:24,010 --> 00:05:26,438 And, people thought that, maybe that gives you, 134 00:05:26,438 --> 00:05:29,210 an potentially an interface layer for A.I. right? 135 00:05:29,210 --> 00:05:31,320 And, that also. 136 00:05:31,320 --> 00:05:34,089 Was, quite a bit of success was there, but again, 137 00:05:34,089 --> 00:05:37,570 now the issue is sort of complimentary to, sort of logic. 138 00:05:37,570 --> 00:05:41,930 The graphical models in their, in the way they propose, are proportional. 139 00:05:41,930 --> 00:05:45,340 Right, so you really argue about each division separately. 140 00:05:45,340 --> 00:05:45,989 There's no. 141 00:05:45,989 --> 00:05:46,520 [COUGH]. 142 00:05:46,520 --> 00:05:48,290 >> Way or explicit notion of how to 143 00:05:48,290 --> 00:05:51,880 really handle relations in, into such a language, right. 144 00:05:51,880 --> 00:05:53,120 And the idea is that. 145 00:05:53,120 --> 00:05:55,200 Can we really combine, the power of these two 146 00:05:55,200 --> 00:05:59,240 together to potential provide a interface layer for artificial intelligence? 147 00:05:59,240 --> 00:06:01,316 And I guess next, right, so this is area, which 148 00:06:01,316 --> 00:06:03,700 has really been for the last 10 years or so. 149 00:06:03,700 --> 00:06:07,300 So people from statistically and logically have combined. 150 00:06:07,300 --> 00:06:11,556 Their strengths, and this area called statistical-relational learning, has come 151 00:06:11,556 --> 00:06:14,970 about, which really combines the power of logic and probability. 152 00:06:14,970 --> 00:06:17,320 On one side, logic, and probability on the other side. 153 00:06:17,320 --> 00:06:19,880 And which potentially can provide an interface layer. 154 00:06:19,880 --> 00:06:22,470 >> So, in the recording [UNKNOWN], that 155 00:06:22,470 --> 00:06:26,000 doesn't mean that you don't have exactly [UNKNOWN]. 156 00:06:26,000 --> 00:06:29,430 I can edit out anything that we don't want to. 157 00:06:29,430 --> 00:06:32,870 >> Surely, yeah, yeah, so, so feel free to ask, ask questions. 158 00:06:32,870 --> 00:06:34,950 I guess that is appropriate, yeah. 159 00:06:34,950 --> 00:06:37,044 So I should mention that, at, at least this part 160 00:06:37,044 --> 00:06:39,620 of the slide has been, borrowed from my advisor Pedro Domingos. 161 00:06:39,620 --> 00:06:42,750 Right, so, I should, I should mention that I did, right. 162 00:06:42,750 --> 00:06:47,863 And I guess this, this, is a still, in the sense this is, I wouldn't really claim 163 00:06:47,863 --> 00:06:52,320 that this is entirely true and, but I will in this talk I will try to maybe. 164 00:06:52,320 --> 00:06:54,819 Try to, give you some argument that how Markov Logic which 165 00:06:54,819 --> 00:06:56,655 combines sort of, strength of Markov 166 00:06:56,655 --> 00:06:59,358 networks and infrastructure logic, infrastructure logic, 167 00:06:59,358 --> 00:07:02,265 could potentially serve as an interface layer and hopefully when we 168 00:07:02,265 --> 00:07:06,200 look at the inference learning and various applications that have been developed. 169 00:07:06,200 --> 00:07:08,567 Hopefully we'll get some idea and maybe, some of you 170 00:07:08,567 --> 00:07:11,630 [INAUDIBLE] is that this can potentially serve as an interface later. 171 00:07:11,630 --> 00:07:13,752 Of course a long way to go, but it, it's a 172 00:07:13,752 --> 00:07:16,810 good motivation to look at it in this, in this way. 173 00:07:16,810 --> 00:07:20,007 Right, so, Markov Logic as an interface layer, potential interface layer, 174 00:07:20,007 --> 00:07:23,443 [INAUDIBLE] you can develop applications, and then you can develop infrastructure, 175 00:07:23,443 --> 00:07:25,971 almost independently, and as long as you are familiar with the 176 00:07:25,971 --> 00:07:28,958 semantics of this, of this interface layer, in this case Markov Logic. 177 00:07:28,958 --> 00:07:29,270 Right. 178 00:07:29,270 --> 00:07:33,940 So this is, this is the first, sort of, motivation for the talk. 179 00:07:33,940 --> 00:07:36,050 And the second one is, little different. 180 00:07:36,050 --> 00:07:37,380 Coming from a different perspective. 181 00:07:37,380 --> 00:07:39,719 Let's say if you are a pragmatist, you won't really worry about all 182 00:07:39,719 --> 00:07:42,310 this big bowls of, you know, having an interface layer for a year, right? 183 00:07:42,310 --> 00:07:46,170 Then, maybe, what you really care about are, you know, small application. 184 00:07:46,170 --> 00:07:48,690 Little applications and if you can do good on that. 185 00:07:48,690 --> 00:07:50,290 And so this example is from that. 186 00:07:50,290 --> 00:07:50,642 Right? 187 00:07:50,642 --> 00:07:53,553 So, think about let's say if you have a social network 188 00:07:53,553 --> 00:07:57,820 analysis, and, you're looking to model your world, or your domain. 189 00:07:57,820 --> 00:08:01,490 Let's say it is a university, or your community, where people are smoking. 190 00:08:01,490 --> 00:08:02,310 People have cancer. 191 00:08:02,310 --> 00:08:03,450 People have friendships. 192 00:08:03,450 --> 00:08:06,180 And, you know that smoking leads to, cancer. 193 00:08:06,180 --> 00:08:07,213 Right, so this is a very good. 194 00:08:07,213 --> 00:08:07,523 [COUGH]. 195 00:08:07,523 --> 00:08:09,630 >> Rule of thumb, but we know that this is not always true. 196 00:08:09,630 --> 00:08:09,915 Right? 197 00:08:09,915 --> 00:08:13,209 People who smoke, they, may or may not have cancer, but at the same time 198 00:08:13,209 --> 00:08:15,315 we know that, if someone smokes then they're 199 00:08:15,315 --> 00:08:18,060 more likely to have cancer, and, than not. 200 00:08:18,060 --> 00:08:20,080 Right, so this is a very rough rule of thumb. 201 00:08:20,080 --> 00:08:20,300 Right? 202 00:08:20,300 --> 00:08:23,902 And again think about, that you don't want to really go for every little person 203 00:08:23,902 --> 00:08:28,110 your community and your domain saying that John has, if John smokes, John has cancer. 204 00:08:28,110 --> 00:08:29,500 If Anna smokes, Anna has cancer. 205 00:08:29,500 --> 00:08:29,725 And so on. 206 00:08:29,725 --> 00:08:29,960 Right? 207 00:08:29,960 --> 00:08:33,030 So, you want to, really specify this, very compactly. 208 00:08:33,030 --> 00:08:34,971 May be some kind of, using some kind of 209 00:08:34,971 --> 00:08:37,520 logic, logical language at the same time you want 210 00:08:37,520 --> 00:08:40,190 to have this, explicit notion of uncertainty and we 211 00:08:40,190 --> 00:08:43,900 will, really formally see what that notion of uncertainty is. 212 00:08:43,900 --> 00:08:44,184 Right? 213 00:08:44,184 --> 00:08:45,661 So similarly you could say that, 214 00:08:45,661 --> 00:08:47,760 friendship leads to similar smoking habits. 215 00:08:47,760 --> 00:08:49,006 Right, so if I am, friend with lot of 216 00:08:49,006 --> 00:08:50,510 people who smoke, then I am more likely to. 217 00:08:50,510 --> 00:08:51,730 To smoke that not. 218 00:08:51,730 --> 00:08:52,057 Right? 219 00:08:52,057 --> 00:08:54,937 >> Similarly if I'm friends with a lot of people who don't 220 00:08:54,937 --> 00:09:00,100 smoke, than I'm more likely to, not be smoking than, having smoking habits. 221 00:09:00,100 --> 00:09:02,757 Right, so this is again, a very good rule of thumb, and 222 00:09:02,757 --> 00:09:06,680 this has been verified by social science that, this in fact is true. 223 00:09:06,680 --> 00:09:06,950 Right? 224 00:09:06,950 --> 00:09:09,040 So how do capture both of these facts? 225 00:09:09,040 --> 00:09:09,710 Right. 226 00:09:09,710 --> 00:09:11,989 That is where bio automatic logic will, comes in. 227 00:09:13,480 --> 00:09:14,960 Right, so this is, the idea, right? 228 00:09:14,960 --> 00:09:18,571 This is, a goal that we, start with, so combining logic and probability, 229 00:09:18,571 --> 00:09:20,277 right, so the real world, problem that 230 00:09:20,277 --> 00:09:22,838 are characterised by entities and relationships, right, 231 00:09:22,838 --> 00:09:26,200 and there is explicit motion of certain behavior that you want to capture, 232 00:09:26,200 --> 00:09:29,970 so you want to capture both these things, the relational aspect of the problem. 233 00:09:29,970 --> 00:09:32,360 And the probabilistic aspect of the problem, right? 234 00:09:32,360 --> 00:09:34,075 Logical languages can handle 235 00:09:34,075 --> 00:09:36,980 relationships, horn clauses, first-order clauses. 236 00:09:36,980 --> 00:09:39,604 Different logical languages, right, you could, really choose depending 237 00:09:39,604 --> 00:09:41,760 on your application what might be the right choice. 238 00:09:41,760 --> 00:09:43,930 Probability can represent uncertainty, right? 239 00:09:43,930 --> 00:09:46,410 So there is, things like Markov networks, Bayesian networks. 240 00:09:46,410 --> 00:09:49,227 There are many different statistical, learning techniques, 241 00:09:49,227 --> 00:09:51,710 but here we'll focus mostly on graphical models. 242 00:09:51,710 --> 00:09:53,610 Like and then how do we really combine the two. 243 00:09:53,610 --> 00:09:53,862 Right? 244 00:09:53,862 --> 00:09:56,177 So when I say combine, you don't want, something that is 245 00:09:56,177 --> 00:09:59,230 heuristic based but you want something which is really well founded. 246 00:09:59,230 --> 00:10:01,069 Both in, terms of the theory which has been 247 00:10:01,069 --> 00:10:04,780 developed for logical languages and for the probabilistic languages. 248 00:10:04,780 --> 00:10:04,930 Right. 249 00:10:04,930 --> 00:10:06,240 So that is the goal. 250 00:10:06,240 --> 00:10:11,160 And if you look at the history of this, this goes way past in 1980's, right. 251 00:10:11,160 --> 00:10:15,270 There is, something called probabilistic logic by Nilsson. 252 00:10:15,270 --> 00:10:17,931 And as you see there has been, consistent progress in 253 00:10:17,931 --> 00:10:20,473 this direction of course I go, in all the details of 254 00:10:20,473 --> 00:10:23,313 this but just to give you summary, [UNKNOWN] each language 255 00:10:23,313 --> 00:10:26,350 is different in terms of, the two components I just listed. 256 00:10:26,350 --> 00:10:28,503 What is the logical language and what is sort 257 00:10:28,503 --> 00:10:31,330 of the, inference or the statistical part of it. 258 00:10:31,330 --> 00:10:33,361 Right, for example, if you think about 259 00:10:33,361 --> 00:10:37,110 basic logic programs proposition casting and loop rate. 260 00:10:37,110 --> 00:10:39,933 Right, in 2001 they combine horn clauses with Bayesian networks. 261 00:10:39,933 --> 00:10:40,282 Right. 262 00:10:40,282 --> 00:10:43,723 So similar rela, relation Markov networks [UNKNOWN] the combined 263 00:10:43,723 --> 00:10:47,840 language which is essentially sequel queries, with Markov networks. 264 00:10:47,840 --> 00:10:51,424 There is a series called BLOG which has, slightly complicated language 265 00:10:51,424 --> 00:10:55,930 and it combines with the, Bayesian networks on the, probabilistic side. 266 00:10:55,930 --> 00:10:59,040 And again, there's a lot of literature on this you can read. 267 00:10:59,040 --> 00:11:01,370 And Markov logic is in fact, is sort of a latecomer. 268 00:11:01,370 --> 00:11:02,440 It was proposed back in 2006. 269 00:11:02,440 --> 00:11:06,220 There was one short page which came earlier but the general paper 270 00:11:06,220 --> 00:11:10,658 appeared in 2006 and that combines about a full first order logic. 271 00:11:10,658 --> 00:11:13,550 For this talk, we'll focus on finite first order logic. 272 00:11:13,550 --> 00:11:15,058 But there are extensions where you can 273 00:11:15,058 --> 00:11:17,300 really go, to the powerful first order logic. 274 00:11:17,300 --> 00:11:18,707 And Markov networks, right? 275 00:11:18,707 --> 00:11:22,028 And and I believe there have been, couple more later, but I won't 276 00:11:22,028 --> 00:11:25,707 really say that Markov logic is probably the most popular ones among these 277 00:11:25,707 --> 00:11:29,381 and one of the reasons is that it has very well developed software 278 00:11:29,381 --> 00:11:33,090 system which you can freely download and use it for your own application. 279 00:11:33,090 --> 00:11:35,495 So that, really has helped to make it popular and lot 280 00:11:35,495 --> 00:11:37,500 of people as I said in the beginning, not just at 281 00:11:37,500 --> 00:11:40,358 University of Washington, where it was proposed but all across the 282 00:11:40,358 --> 00:11:43,296 world people are using it for their own, application and programs. 283 00:11:43,296 --> 00:11:44,430 Right? 284 00:11:44,430 --> 00:11:45,357 [BLANK_AUDIO] 285 00:11:45,357 --> 00:11:47,662 Okay, so I guess that, ends the motivation. 286 00:11:47,662 --> 00:11:49,183 I guess it would, maybe a good point 287 00:11:49,183 --> 00:11:51,610 to ask any questions or interrupt at this point. 288 00:11:51,610 --> 00:11:53,689 And then, if not then I guess I'll 289 00:11:53,689 --> 00:11:56,160 continue with the, with the rest of the talk.