1 00:03:06,485 --> 00:03:12,323 Yeah, that's out of [INAUDIBLE] course of jaguars [INAUDIBLE] 2 00:03:12,323 --> 00:03:14,922 >> Well, I was trying to figure out 3 00:03:14,922 --> 00:03:18,153 like what is the [INAUDIBLE] >> Yeah, like what is the objection? 4 00:03:18,153 --> 00:03:19,035 >> I didn't follow that. 5 00:03:19,035 --> 00:03:24,196 >> We are talking in the season 12. 6 00:04:33,392 --> 00:04:36,786 Portlow works doesn't tell you anything about the Jaguar. 7 00:04:38,230 --> 00:04:39,980 That's it, yeah. 8 00:04:39,980 --> 00:04:45,711 Like there's too many layers of probabilistic 9 00:04:45,711 --> 00:04:49,727 fields between granularities. 10 00:04:49,727 --> 00:04:55,119 So, yeah, the idea is like, mental Theory of Everything isn't a theory, 11 00:04:55,119 --> 00:04:59,160 because it doesn't explain the fundamentals. 12 00:04:59,160 --> 00:05:04,398 We ultimately explained everything but like an explanation for 13 00:05:04,398 --> 00:05:10,800 biology given in terms of fundamental particles would be literally useless. 14 00:05:10,800 --> 00:05:12,587 >> Yeah [CROSSTALK] >> Yeah, sure. 15 00:05:12,587 --> 00:05:16,081 Okay, then we're all set up. 16 00:05:16,081 --> 00:05:20,777 So [INAUDIBLE] >> Assuming that wee we're talking about 17 00:05:20,777 --> 00:05:21,390 that in a minute. 18 00:05:21,390 --> 00:05:22,459 The objection raised, 19 00:05:22,459 --> 00:05:25,884 we shouldn't be comparing those things cuz they;re at different scales. 20 00:05:25,884 --> 00:05:30,364 >> That's what I'm trying to figure out [INAUDIBLE] 21 00:05:30,364 --> 00:05:31,479 >> The point I think, 22 00:05:31,479 --> 00:05:35,944 the paragraph talks about how [INAUDIBLE] everything that you have [INAUDIBLE] 23 00:05:35,944 --> 00:05:40,351 the Jaguar but also the opposite is also just is true, that it has nothing. 24 00:05:40,351 --> 00:05:45,774 But if you learned something new about sports, you're not gonna be like, 25 00:05:45,774 --> 00:05:51,714 man, that means the Jaguars are like, doesn't give you any knowledge about that. 26 00:05:55,037 --> 00:05:58,924 [LAUGH] >> The microphone [INAUDIBLE] 27 00:05:58,924 --> 00:06:00,815 >> [INAUDIBLE] the road mic is here. 28 00:06:00,815 --> 00:06:01,891 Get yourself. 29 00:06:01,891 --> 00:06:06,495 >> [INAUDIBLE] >> Yeah. 30 00:07:37,584 --> 00:07:43,635 Is this connected to the something called Latent Semantic Analysis? 31 00:07:43,635 --> 00:07:46,099 >> I'm not super familiar with like Latent Semantic Analysis, but 32 00:07:46,099 --> 00:07:49,745 I think there's sort of a connection >> That latent space is that connection. 33 00:07:49,745 --> 00:07:52,087 >> Yeah. >> Latent variable models, so 34 00:07:52,087 --> 00:07:53,654 that's the only key. 35 00:08:07,440 --> 00:08:13,660 Everyone to our persona 2018 our. 36 00:08:16,380 --> 00:08:25,000 Yeah, he did my vision Mickey goes back in 2015 and 37 00:08:25,000 --> 00:08:29,880 now this work is like part of What are we doing 38 00:08:29,880 --> 00:08:34,610 since my networks course I feel like as an instructor, 39 00:08:34,610 --> 00:08:36,570 I feel like I didn't do that back, so- >> [LAUGH] 40 00:08:36,570 --> 00:08:38,270 >> Really interesting and 41 00:08:38,270 --> 00:08:41,340 we just submitted the work to the conference, so, 42 00:08:41,340 --> 00:08:46,840 I'm happy about it, and I, it like also I do advertisement for 43 00:08:46,840 --> 00:08:48,350 networks, cuz that's- >> [LAUGH] 44 00:08:48,350 --> 00:08:52,780 >> That you can write papers But anyway, 45 00:08:52,780 --> 00:08:56,890 so that's like Jasper is like a system scientist at heart but 46 00:08:56,890 --> 00:08:59,090 he's heading in computer science. 47 00:08:59,090 --> 00:09:07,090 Yeah, yeah, I look forward to a lot more work works with it. 48 00:09:07,090 --> 00:09:08,060 There's just a teaser. 49 00:09:09,200 --> 00:09:09,700 Okay. 50 00:09:11,250 --> 00:09:13,470 Okay, damn Jasper. 51 00:09:13,470 --> 00:09:18,720 I'm in computer science which is an obscure 52 00:09:18,720 --> 00:09:23,720 branch of cybernetics with few practical applications. 53 00:09:23,720 --> 00:09:30,110 So yeah, so we've been working on this since spring, 54 00:09:30,110 --> 00:09:33,830 and we just submitted a abstract to the networks on conference. 55 00:09:34,910 --> 00:09:38,630 And we will send on manuscript, and 56 00:09:38,630 --> 00:09:40,909 there's a couple sources that will come up during this talk. 57 00:09:45,180 --> 00:09:51,680 We're going to cover so come up plays a phenomenon 58 00:09:51,680 --> 00:09:56,180 in social networks and we're not talking about catching too late in space. 59 00:09:58,490 --> 00:10:00,870 Random geometric graphs in Brownian motion are sorta the building 60 00:10:00,870 --> 00:10:01,750 blocks of our model. 61 00:10:01,750 --> 00:10:06,390 But we'll be metric grass is our model it was named by someone else. 62 00:10:09,090 --> 00:10:13,156 And implementation here I'm talking about modeling choices, not the code. 63 00:10:13,156 --> 00:10:17,162 Some observations. 64 00:10:17,162 --> 00:10:26,849 [INAUDIBLE] And what we're working, on the social networks? 65 00:10:26,849 --> 00:10:33,750 Go kind of know what they are the master relationships between actors. 66 00:10:33,750 --> 00:10:38,113 And so two big questions are what the actor sells networks and vice versa. 67 00:10:42,578 --> 00:10:43,530 This is social networking. 68 00:10:45,910 --> 00:10:49,420 It's one that they've disguised lead algorithms karate club. 69 00:10:53,640 --> 00:10:54,610 This concept of the mob. 70 00:10:56,300 --> 00:11:01,320 So this is the idea that actors in a social 71 00:11:01,320 --> 00:11:06,430 network who have attributes in common are more likely to be related in the network. 72 00:11:10,590 --> 00:11:13,100 Both ways act as a common attributes. 73 00:11:13,100 --> 00:11:16,520 Form ties and type to hackers indicate common attributes and 74 00:11:16,520 --> 00:11:19,311 answers to the question so talked about before. 75 00:11:19,311 --> 00:11:19,818 Sort off. 76 00:11:22,511 --> 00:11:24,555 And so it's a concept from sociology. 77 00:11:24,555 --> 00:11:27,575 So it's typically expressed informally. 78 00:11:27,575 --> 00:11:31,865 Obviously people who do studies measuring those have to come up with definitions and 79 00:11:33,395 --> 00:11:38,961 we're gonna talk about one so welcome to this 80 00:11:40,760 --> 00:11:46,010 That's the idea that the attributes the actors have given them coordinates in 81 00:11:46,010 --> 00:11:48,860 some kind of abstract space or possibly physical space. 82 00:11:50,300 --> 00:11:54,510 So for example, we're all in this room, physically and 83 00:11:54,510 --> 00:11:58,889 that allows us to interact because we're located close together in some space. 84 00:11:59,950 --> 00:12:03,840 We're all in this room because we are interested in similar topics. 85 00:12:03,840 --> 00:12:09,480 That's a sort of more abstract kind of space that we're positioned in. 86 00:12:11,070 --> 00:12:15,680 So this becomes sort of version of what geographers called Tobler's Law 87 00:12:15,680 --> 00:12:18,370 with everything related to everything else. 88 00:12:18,370 --> 00:12:20,640 But your things are more related than distant things. 89 00:12:23,110 --> 00:12:27,410 And this is just something we talked about a while ago. 90 00:12:27,410 --> 00:12:29,950 It's really hard to compute examples of networks that 91 00:12:29,950 --> 00:12:31,730 you can't crop up latent space for. 92 00:12:35,130 --> 00:12:41,700 Started with like, weird abstract examples But this is like a latent space. 93 00:12:41,700 --> 00:12:44,810 People might be familiar with Twitter social relations 94 00:12:47,410 --> 00:12:53,028 Is the least political football complex meme I could find. 95 00:12:53,028 --> 00:12:59,760 So here's what we're talking about basically. 96 00:12:59,760 --> 00:13:05,750 So this is a space looks like the compass. 97 00:13:05,750 --> 00:13:08,643 And here's the graph derived from positions notes. 98 00:13:10,383 --> 00:13:18,884 So a few, a few things go into the space. 99 00:13:18,884 --> 00:13:21,060 So we've got a distance functions here. 100 00:13:21,060 --> 00:13:25,360 This is actually just a function of the distance and 101 00:13:25,360 --> 00:13:28,340 distances, Euclidean distance as the crow flies. 102 00:13:28,340 --> 00:13:32,679 We could change that to Manhattan distance. 103 00:13:34,530 --> 00:13:39,380 Get some minor changes, he did something, more different like harming distance. 104 00:13:41,440 --> 00:13:44,400 So, in harming distance you're only 105 00:13:44,400 --> 00:13:47,330 in the same coordinate if you have exactly the same coordinate. 106 00:13:47,330 --> 00:13:52,173 So, who's connected here is this point, a point? 107 00:13:52,173 --> 00:13:55,954 >> A point [INAUDIBLE] >> You get the point. 108 00:13:55,954 --> 00:14:01,779 Until we might think of a space like this, 109 00:14:01,779 --> 00:14:09,118 giving us probabilities that nodes are connected. 110 00:14:09,118 --> 00:14:16,430 Zero and around. 111 00:14:16,430 --> 00:14:21,126 Change my ties, I just change the weights and also, the probabilities, 112 00:14:21,126 --> 00:14:25,311 which on this cut off backup, working as the most likely networks 113 00:14:28,037 --> 00:14:31,810 So, I made this realization in the network's course in spring. 114 00:14:33,840 --> 00:14:36,750 Is the adjacency matrix. 115 00:14:36,750 --> 00:14:40,540 What we're looking at that's not attainable, that's what's the table. 116 00:14:43,360 --> 00:14:49,030 And the network that is represented by the agency matrix is not this one. 117 00:14:49,030 --> 00:14:50,710 So what are these ties? 118 00:14:50,710 --> 00:14:56,180 >> Well, so this is the Network is represented by a matrix with a caveat. 119 00:14:56,180 --> 00:15:02,210 So if we turn this all the way down, then no, that's here. 120 00:15:02,210 --> 00:15:03,260 That's literally true. 121 00:15:03,260 --> 00:15:05,436 >> That's the matrix right there. 122 00:15:05,436 --> 00:15:10,629 At zero, that's the one represented by the adjacency matrix right there. 123 00:15:11,630 --> 00:15:14,920 So this is the network, the matrix is not pictured. 124 00:15:14,920 --> 00:15:16,880 So the matrix is just a table. 125 00:15:16,880 --> 00:15:20,870 And so if the chief entry guy throw is nonzero, 126 00:15:20,870 --> 00:15:24,590 agree this is the one represented by that with no cut off. 127 00:15:24,590 --> 00:15:26,629 Yes and one has no rows or columns. 128 00:15:33,480 --> 00:15:33,980 Yeah. 129 00:15:35,617 --> 00:15:39,240 So in real life, we know that attributes change and so 130 00:15:39,240 --> 00:15:40,190 when we >> [COUGH] 131 00:15:40,190 --> 00:15:41,423 >> Made the simulation 132 00:15:41,423 --> 00:15:45,346 string we got talking about possible dynamics in play in space. 133 00:15:48,177 --> 00:15:50,756 There's a lot of ways you could do that 134 00:15:53,530 --> 00:15:57,010 Social behavior is really complicated and interdependent. 135 00:15:59,320 --> 00:16:01,010 People influence each other. 136 00:16:01,010 --> 00:16:04,730 They're influenced by social forces, not necessarily people. 137 00:16:04,730 --> 00:16:07,640 You know, you might have boys communicated to you, you might. 138 00:16:07,640 --> 00:16:10,300 Decide it's time to change your beliefs because you see everyone changing their 139 00:16:10,300 --> 00:16:13,410 beliefs, but you're not particularly influenced by the beliefs they choose. 140 00:16:13,410 --> 00:16:14,250 All sorts of things happen. 141 00:16:14,250 --> 00:16:19,060 So, for example, in 2016, a big chunk of America changed their beliefs. 142 00:16:20,160 --> 00:16:24,395 Not talking about the election, talking about this, 143 00:16:24,395 --> 00:16:27,532 Duke Chapman survey of American fears. 144 00:16:27,532 --> 00:16:31,491 And in the year 2016, 15% of Americans 145 00:16:31,491 --> 00:16:36,374 decided that ancient advanced civilizations existed. 146 00:16:36,374 --> 00:16:37,422 I think that's crazy. 147 00:16:37,422 --> 00:16:40,390 >> [LAUGH] >> That's just an example. 148 00:16:40,390 --> 00:16:41,550 >> It's quite a jump. 149 00:16:41,550 --> 00:16:45,426 >> Yeah, that's weird, and it didn't grow at all basically in this year. 150 00:16:45,426 --> 00:16:49,693 Yeah, all these are going up really fast, 5% a year. 151 00:16:49,693 --> 00:16:51,845 >> Do you have older data than that? 152 00:16:51,845 --> 00:16:53,225 >> No, that's from their website. 153 00:16:53,225 --> 00:16:53,873 >> Yeah. 154 00:16:59,123 --> 00:17:04,013 >> So null models, did a lot of these in network science. 155 00:17:06,156 --> 00:17:10,873 We use them to distinguish things that can be explained by randomness 156 00:17:10,873 --> 00:17:12,673 from things that can't. 157 00:17:12,673 --> 00:17:15,414 So three examples. 158 00:17:15,414 --> 00:17:17,307 This is an Erdős–Rényi graph, 159 00:17:17,307 --> 00:17:20,750 where nodes are basically connected globally at random. 160 00:17:23,404 --> 00:17:28,552 This is a configuration model where you decide how many connections you 161 00:17:28,552 --> 00:17:34,330 want each node to have, and then you tie those loose ends together a random way. 162 00:17:34,330 --> 00:17:38,065 And you get a network with the exact degree sequence you want. 163 00:17:41,333 --> 00:17:45,759 And this is the stochastic block model where you specify some community structure 164 00:17:45,759 --> 00:17:47,500 for each of these communities. 165 00:17:49,520 --> 00:17:52,395 Have a much higher likelihood of connecting within their own community 166 00:17:52,395 --> 00:17:52,920 then between. 167 00:17:54,120 --> 00:17:58,472 So you specify some probabilities for those, and 168 00:17:58,472 --> 00:18:02,934 then you pick randomly the exact size that happen. 169 00:18:08,141 --> 00:18:11,560 The null models show what can be explained by chance. 170 00:18:13,120 --> 00:18:15,720 The random dynamics are useful for testing non-random alternatives. 171 00:18:16,800 --> 00:18:22,238 Again, null model for space network dynamics you have random dynamics and 172 00:18:22,238 --> 00:18:23,800 networks that don't. 173 00:18:23,800 --> 00:18:25,850 So all these cases, there's something that's random and 174 00:18:25,850 --> 00:18:27,827 then there's some desired property that we've fixed. 175 00:18:31,731 --> 00:18:38,415 And so the temporal networks are networks that change over time, 176 00:18:38,415 --> 00:18:43,286 which we tend to think that most of them do that. 177 00:18:47,041 --> 00:18:49,216 And so it's just about that. 178 00:18:49,216 --> 00:18:52,520 So static networks could be instantaneous snapshots for 179 00:18:52,520 --> 00:18:57,290 aggregates over a window of time in a natural temporal network. 180 00:18:57,290 --> 00:19:01,919 So a good null model for temporal networks would be random but 181 00:19:01,919 --> 00:19:07,746 it would be time dependent, and that would separate it from a static model. 182 00:19:11,749 --> 00:19:15,947 The building blocks for this random geometric graphs, they're graphs, 183 00:19:15,947 --> 00:19:18,418 they're geometric, and they are random. 184 00:19:18,418 --> 00:19:20,083 And they look like that. 185 00:19:22,793 --> 00:19:27,600 And you can be forgiven for thinking that there is some non-random structure here, 186 00:19:27,600 --> 00:19:31,392 sort of looks like there's clusters that are really connected and 187 00:19:31,392 --> 00:19:34,460 clusters that are, but that's by chance. 188 00:19:34,460 --> 00:19:39,582 Random data tends to be clumpier than people think it would be, and 189 00:19:39,582 --> 00:19:45,906 that effect is magnified when the points become these dense connected clusters. 190 00:19:48,070 --> 00:19:49,876 >> It's like the highways in Oregon. 191 00:19:49,876 --> 00:19:53,383 >> [LAUGH] >> Yeah, and 192 00:19:53,383 --> 00:19:54,772 so- >> Yeah, 193 00:19:54,772 --> 00:19:58,620 it's possible because it's any spatial point process. 194 00:19:58,620 --> 00:20:04,380 So any geographers use spatial process to model highway connections, 195 00:20:04,380 --> 00:20:07,400 any flow, can we test it. 196 00:20:08,570 --> 00:20:10,260 >> So this is what they are in detail. 197 00:20:10,260 --> 00:20:12,235 You have some point process, 198 00:20:12,235 --> 00:20:15,952 is a statistical process that places points in a space. 199 00:20:15,952 --> 00:20:18,990 Uniformally at random is the one we're gonna talk about most. 200 00:20:20,090 --> 00:20:24,755 And then you have nodes within some radius of each other connected, and 201 00:20:24,755 --> 00:20:27,173 this comes up in wireless networks. 202 00:20:27,173 --> 00:20:29,285 So it's not exactly the [INAUDIBLE], 203 00:20:29,285 --> 00:20:33,450 because that had a function of distance and this is just always a cut off. 204 00:20:34,970 --> 00:20:38,420 >> Is that what you mean by these networks are geometric, 205 00:20:38,420 --> 00:20:41,651 that you're making these lengths based on space? 206 00:20:41,651 --> 00:20:42,584 >> Yeah, exactly. 207 00:20:42,584 --> 00:20:46,163 >> Okay, okay. 208 00:20:46,163 --> 00:20:48,881 The other are part of our model is the Brownian motion. 209 00:20:50,734 --> 00:20:54,231 So Brownian motion, you move a normally distributed distance in either 210 00:20:54,231 --> 00:20:56,252 direction for each of your coordinates. 211 00:20:58,952 --> 00:21:05,392 Brown discovered this in particles in fluid, shows up in lots of contexts. 212 00:21:08,173 --> 00:21:12,026 And it preserves the intensity of the underlying point process, 213 00:21:12,026 --> 00:21:13,924 which will be important later. 214 00:21:16,255 --> 00:21:20,870 The mobile geometric graphs combine those two. 215 00:21:20,870 --> 00:21:27,652 So we came up with this and then we found two papers on this from 2010, 2014. 216 00:21:30,426 --> 00:21:33,873 And so the Brownian motion adds time dependence while preserving the randomness 217 00:21:33,873 --> 00:21:35,251 in the random geometric graph. 218 00:21:37,551 --> 00:21:42,262 So the existing work on this focuses on mathematical statistics of communication 219 00:21:42,262 --> 00:21:46,577 properties, which we'll talk a little bit more about later, but it's, 220 00:21:48,400 --> 00:21:53,022 A lot of properties are only shown for internet networks, 221 00:21:53,022 --> 00:21:58,033 which we're not gonna deal with internet networks very much. 222 00:21:59,775 --> 00:22:03,220 And it's all in this sorta statistical way of writing. 223 00:22:03,220 --> 00:22:07,497 So it's all talking about pre-written things rather than simulating them. 224 00:22:09,902 --> 00:22:13,579 So next slide. 225 00:22:13,579 --> 00:22:16,572 So, let me zoom out a little. 226 00:22:19,004 --> 00:22:22,362 So this looks flat as well. 227 00:22:22,362 --> 00:22:23,930 These should be perfect circles. 228 00:22:23,930 --> 00:22:30,360 But so this is our model, or representation of it. 229 00:22:30,360 --> 00:22:32,040 And it's sort of similar to the one, 230 00:22:32,040 --> 00:22:35,780 we have this way of defining an adjacency matrix. 231 00:22:35,780 --> 00:22:40,435 In practice, we really don't do the whole matrix cuz that would be really 232 00:22:40,435 --> 00:22:42,545 computationally inefficient. 233 00:22:42,545 --> 00:22:45,606 And then we have this positions or this vector x, 234 00:22:45,606 --> 00:22:49,797 and that evolves according to Brownian motion for each particle. 235 00:22:57,105 --> 00:23:00,292 And just look at, let's see, 236 00:23:00,292 --> 00:23:05,080 this is what it looks like when it's running. 237 00:23:05,080 --> 00:23:06,760 This is a pretty sparse one. 238 00:23:06,760 --> 00:23:09,103 Let's count the number of nodes. 239 00:23:10,948 --> 00:23:12,301 We can turn up the radius. 240 00:23:12,301 --> 00:23:17,528 I got this transitioned to a big connected component. 241 00:23:17,528 --> 00:23:21,304 We turn it all the way down we have little tiny components. 242 00:23:23,005 --> 00:23:26,050 We turned up the diffusion constant, 243 00:23:26,050 --> 00:23:30,629 then less information is preserved between time steps. 244 00:23:31,840 --> 00:23:35,040 So you can see the network changes really dramatically. 245 00:23:36,732 --> 00:23:39,720 This is the one major artifact in our simulation 246 00:23:39,720 --> 00:23:44,030 is that our simulation's discrete, so with a high diffusion constant 247 00:23:44,030 --> 00:23:47,100 nodes can just skip over each other without ever connecting, 248 00:23:47,100 --> 00:23:50,770 which would not happen if they were actually moving in space. 249 00:23:50,770 --> 00:23:55,467 So that's just one thing that we're being conscious of. 250 00:23:56,901 --> 00:24:02,911 So we set the diffusion constant pretty well usually. 251 00:24:08,987 --> 00:24:13,439 So, We're looking at a few different topologies for this. 252 00:24:13,439 --> 00:24:16,157 In this example, particles just get its shadow. 253 00:24:16,157 --> 00:24:22,080 Particles just get stuck at the boundary. 254 00:24:22,080 --> 00:24:23,920 I can't cross it. 255 00:24:23,920 --> 00:24:27,015 And this is not like any of the simulations we're actually talking about 256 00:24:27,015 --> 00:24:27,592 in my paper. 257 00:24:27,592 --> 00:24:32,214 So there's three different boundary conditions, so 258 00:24:32,214 --> 00:24:34,729 we started with no boundary. 259 00:24:34,729 --> 00:24:38,762 The particles are initializing some cube in space and 260 00:24:38,762 --> 00:24:42,800 they just get to diffuse out into open space forever. 261 00:24:44,170 --> 00:24:48,357 We tried moving those across the boundary, and we tried simulating 262 00:24:48,357 --> 00:24:53,340 the Taurus on that part to exit one side and come back on the other side. 263 00:24:53,340 --> 00:24:59,800 So each of these has pretty strong effects on network statistics. 264 00:24:59,800 --> 00:25:04,940 This is the average clustering coefficient over time for 265 00:25:04,940 --> 00:25:06,422 Taurus and it stays about same. 266 00:25:06,422 --> 00:25:15,300 This is the average clustering coefficient with no boundaries, with just a piece out. 267 00:25:15,300 --> 00:25:19,930 This transient will go on for a really long time, sort of forever. 268 00:25:19,930 --> 00:25:27,346 But it'll get increasingly few connections to become rare particles that for 269 00:25:27,346 --> 00:25:31,852 me [INAUDIBLE] it looks like if you had them exit. 270 00:25:31,852 --> 00:25:33,416 >> [COUGH] >> It will go down here pretty quickly. 271 00:25:35,134 --> 00:25:37,771 Although that has gonna do with this being a really tiny space. 272 00:25:39,133 --> 00:25:45,724 And then we looked at some different initial conditions. 273 00:25:45,724 --> 00:25:51,070 It may one uniform random starting points within some bounding box. 274 00:25:52,220 --> 00:25:55,824 But we also looked at having all the nodes start from one point for 275 00:25:55,824 --> 00:25:58,986 a small area that's not the same as the outer boundary. 276 00:25:58,986 --> 00:26:04,029 And so those two cases could be a model for 277 00:26:04,029 --> 00:26:07,140 Community Solutions. 278 00:26:07,140 --> 00:26:09,190 It's a starting point, obviously, 279 00:26:09,190 --> 00:26:14,590 it's fully connected to begin with and it just falls apart. 280 00:26:14,590 --> 00:26:21,575 To start a small area, it might have like a more realistic topology to begin with. 281 00:26:21,575 --> 00:26:25,703 Essentially, these initial conditions can be combined with boundary conditions, and 282 00:26:25,703 --> 00:26:27,638 so we have a lot of possible way to do this. 283 00:26:27,638 --> 00:26:32,667 The one we've looked at most probably is on the Taurus, 284 00:26:32,667 --> 00:26:38,017 where it behaves more like the statistician say it should, 285 00:26:38,017 --> 00:26:40,158 for Internet network. 286 00:26:40,158 --> 00:26:43,540 >> The intensity. 287 00:26:43,540 --> 00:26:49,525 >> I'm gonna have to put that one [INAUDIBLE] 288 00:26:49,525 --> 00:26:51,576 >> In thing, I think here, 289 00:26:51,576 --> 00:26:55,690 intensity is the chances of connections. 290 00:26:55,690 --> 00:26:57,720 So when you have like the Brownian motion, 291 00:26:57,720 --> 00:27:01,580 you could do abstract problem of like gamma two particles. 292 00:27:01,580 --> 00:27:06,210 What are the odds that they will meet in space, wanted space? 293 00:27:06,210 --> 00:27:10,220 And that it's in one dimension, it's like really dense. 294 00:27:10,220 --> 00:27:12,380 Mathematically, it's like a dense set. 295 00:27:12,380 --> 00:27:14,810 And for two dimensions, it's to get sparse. 296 00:27:14,810 --> 00:27:18,190 Three, it gets very sparse, but still, they keep reading. 297 00:27:18,190 --> 00:27:19,671 >> We're gonna try that part. 298 00:27:19,671 --> 00:27:26,438 >> So that's like the intensity like so, if you have more higher dimensions, 299 00:27:26,438 --> 00:27:31,743 the chances of Brownian motion meeting can drop up at four. 300 00:27:31,743 --> 00:27:33,750 Like after four, they don't need it. 301 00:27:33,750 --> 00:27:36,037 >> In this context, it's basically density, but 302 00:27:36,037 --> 00:27:38,450 have some technical meaning and processes. 303 00:27:38,450 --> 00:27:40,420 So I guess it's the correct word to use. 304 00:27:43,510 --> 00:27:49,910 This is behavior on course, so again, this is average clustering coefficient. 305 00:27:49,910 --> 00:27:53,889 The black line is the average for a static route the metric route, 306 00:27:53,889 --> 00:27:57,770 so I computed 300 of those and just took the average characteristic. 307 00:27:57,770 --> 00:28:01,579 And you can see that for the dynamic ones, they sort of wander around, but 308 00:28:01,579 --> 00:28:02,693 they stay near this. 309 00:28:07,807 --> 00:28:09,910 This is average shortest path length. 310 00:28:09,910 --> 00:28:15,980 So this one's interesting because shortest path length is actually defined for 311 00:28:15,980 --> 00:28:16,790 single components. 312 00:28:16,790 --> 00:28:20,850 You can't have shortest paths between those that are connected. 313 00:28:20,850 --> 00:28:24,320 Since there's actually the average over components and 314 00:28:24,320 --> 00:28:27,802 this is a small dense network. 315 00:28:27,802 --> 00:28:30,970 So it's 100 particles in a six by six square. 316 00:28:32,300 --> 00:28:36,723 And let me go back up to the model [COUGH] and 317 00:28:36,723 --> 00:28:44,020 it looks like something like this. 318 00:28:45,020 --> 00:28:50,290 We have one giant blob, and occasionally, a single particle splits off from that. 319 00:28:51,960 --> 00:28:57,614 We might see that any second now, or we might scroll back down. 320 00:28:57,614 --> 00:28:58,124 >> Wait, 321 00:28:58,124 --> 00:29:04,340 why did changing the radius be equivalent to that the radius of their connection? 322 00:29:04,340 --> 00:29:09,878 >> So I mean, increasing the radius is like having a smaller, denser system. 323 00:29:09,878 --> 00:29:10,637 >> Yes. 324 00:29:10,637 --> 00:29:14,061 >> We're actually gonna see the a digital 325 00:29:14,061 --> 00:29:18,843 taking up the exact system that generated that graph. 326 00:29:18,843 --> 00:29:22,286 Yeah, so every time one particle splits off, 327 00:29:22,286 --> 00:29:27,600 that's a component that has one path and the path is like zero. 328 00:29:27,600 --> 00:29:30,986 So there's now two components, one of them is zero, and so 329 00:29:30,986 --> 00:29:34,390 this drops by half every time a single particle splits off. 330 00:29:37,300 --> 00:29:41,789 And this is the same thing for a Western system is not one giant component. 331 00:29:42,830 --> 00:29:46,095 And it just looks more normal. 332 00:29:46,095 --> 00:29:50,440 Cuz that's both the effect of the Taurus and 333 00:29:50,440 --> 00:29:55,570 the effect of our phase transition to the coordinated graph. 334 00:29:58,240 --> 00:30:02,021 So dimensionality and sparsity is all in that film screen, yep. 335 00:30:05,011 --> 00:30:08,613 So if you fix every variable in the model except dimension and you increase 336 00:30:08,613 --> 00:30:12,519 dimension like how this can create drastic changes in the shape of the network. 337 00:30:15,897 --> 00:30:18,968 As we crewed up the same about dimension four, 338 00:30:18,968 --> 00:30:23,750 exact intersections and branching paths are extremely rare. 339 00:30:23,750 --> 00:30:27,017 Which is like actual timestamp temporal data. 340 00:30:27,017 --> 00:30:31,517 Like if you had emails, those be dated down to the second and 341 00:30:31,517 --> 00:30:35,663 at any given second, there would be probably one now. 342 00:30:35,663 --> 00:30:41,530 So I'm not saying this is necessarily makes it a great model for 343 00:30:41,530 --> 00:30:45,000 this but it's just a connection. 344 00:30:45,000 --> 00:30:51,030 And when I'm dealing with exactly a second year intersections, 345 00:30:51,030 --> 00:30:54,470 which means it's not there are still some connection for 346 00:30:54,470 --> 00:30:58,210 both dimensions but they're very very still. 347 00:30:58,210 --> 00:31:02,450 And other behaviors like component lifetime change sharply as well. 348 00:31:02,450 --> 00:31:05,980 So, this is all gonna be on Taurus. 349 00:31:05,980 --> 00:31:08,830 This is a one dimensional Taurus, so it's a circle. 350 00:31:08,830 --> 00:31:09,680 It's very dense. 351 00:31:10,680 --> 00:31:13,737 It looks like alien language from arrival. 352 00:31:13,737 --> 00:31:16,529 [SOUND] So we got the two dimensions. 353 00:31:16,529 --> 00:31:21,520 So this is the graph that I was talking about before. 354 00:31:21,520 --> 00:31:25,170 And you can see it's one by one, 355 00:31:25,170 --> 00:31:30,830 I was trying to make enough simulation notes, okay? 356 00:31:32,260 --> 00:31:37,470 This is three dimensions, so it tends to form these stringy little components. 357 00:31:37,470 --> 00:31:38,700 I think it's interesting. 358 00:31:41,030 --> 00:31:44,827 But there's still a few triads, this five together is pretty rare. 359 00:31:48,504 --> 00:31:53,860 But, yeah, it's a pretty sparse, four dimensions, more than same. 360 00:31:53,860 --> 00:32:00,210 Triads are very rare to get a lot of lines, five dimensions. 361 00:32:01,780 --> 00:32:05,190 You have two connections out of 100. 362 00:32:05,190 --> 00:32:08,300 >> Meaning that they were assigned a five coordinate 363 00:32:08,300 --> 00:32:10,230 random location at the beginning and 364 00:32:10,230 --> 00:32:14,930 then your radius calculation was still same as if it was a lower dimension. 365 00:32:14,930 --> 00:32:15,430 >> Yep. 366 00:32:19,121 --> 00:32:23,924 But it's still a fairly dense thing cuz the radius connection is also it's going 367 00:32:23,924 --> 00:32:25,351 to higher dimensions. 368 00:32:25,351 --> 00:32:31,531 It's now a sphere instead of a, well it's a five sphere instead of a circle. 369 00:32:33,962 --> 00:32:37,854 So even for fairly dense ones, it gets very sparse in high dimensions. 370 00:32:37,854 --> 00:32:41,496 Because obviously lots of other ways you can control sparsity, 371 00:32:41,496 --> 00:32:46,082 we can make the radius smaller, we can have fewer particles over a larger space, 372 00:32:46,082 --> 00:32:48,795 but dimension makes these really big changes. 373 00:32:54,550 --> 00:32:58,067 So, What we're working on, 374 00:32:58,067 --> 00:33:04,158 we've talked about, The null model application. 375 00:33:04,158 --> 00:33:08,587 There's maybe some suggesting this random geometric 376 00:33:08,587 --> 00:33:13,217 graphs could have applications for wireless networks. 377 00:33:13,217 --> 00:33:15,278 And we've also talked about using it for 378 00:33:15,278 --> 00:33:17,652 testing the latent space inference methods. 379 00:33:17,652 --> 00:33:21,390 So if we can generate this data where there's a latent space in the network and 380 00:33:21,390 --> 00:33:25,011 then we can see people who claim they have methods for taking the network and 381 00:33:25,011 --> 00:33:26,413 deriving the latent space. 382 00:33:26,413 --> 00:33:27,662 We can see if that works on our model. 383 00:33:30,534 --> 00:33:32,211 What would Newman do? 384 00:33:32,211 --> 00:33:37,257 If you don't know, Mark Newman is this network theorist who writes these high 385 00:33:37,257 --> 00:33:42,140 level elaborations of reference models, basically that's all he does. 386 00:33:42,140 --> 00:33:45,964 So he would document and explain standard network characteristics and 387 00:33:45,964 --> 00:33:47,692 examine the phase transition. 388 00:33:47,692 --> 00:33:50,244 So we're working on that. 389 00:33:50,244 --> 00:33:56,110 And then there's some stuff we can do from temporal network analysis. 390 00:33:56,110 --> 00:33:59,280 There's a lot of different ways people analyze temporal networks. 391 00:33:59,280 --> 00:34:03,231 It's very active, people don't seem to have settled all that stuff yet. 392 00:34:03,231 --> 00:34:07,693 But one thing they all agree on is percolation properties or contagion or 393 00:34:07,693 --> 00:34:11,131 communication, those are all basically the same, and 394 00:34:11,131 --> 00:34:14,370 they're all something everyone likes basically. 395 00:34:14,370 --> 00:34:19,120 And so Peres et al, that's the statisticians have worked out statistics 396 00:34:19,120 --> 00:34:23,940 for certain types of spreading, and they're focused on communications. 397 00:34:24,980 --> 00:34:29,666 So they basically say, how long would it take for a message to get from one end of 398 00:34:29,666 --> 00:34:35,782 the network to the other or between two nodes, If whenever it arrives at 399 00:34:35,782 --> 00:34:39,379 a node it's instantaneously communicated to every node that's with that code. 400 00:34:39,379 --> 00:34:43,668 So it's not quite the standard way of doing contagion, so 401 00:34:43,668 --> 00:34:48,572 something like the susceptible infected recovered model would be 402 00:34:48,572 --> 00:34:52,110 more standard and we might do that. 403 00:34:52,110 --> 00:34:55,431 Although these guys want it, it's pretty easy to program it. 404 00:34:57,541 --> 00:35:00,490 And there's just tons and 405 00:35:00,490 --> 00:35:06,110 tons of possible variants that seem easy to do. 406 00:35:06,110 --> 00:35:08,246 So, nodes entering and exiting, 407 00:35:08,246 --> 00:35:13,110 having non-binary relations between nodes, swapping the Brownian node for 408 00:35:13,110 --> 00:35:17,910 a different kind of random walk is something that statisticians suggested. 409 00:35:20,540 --> 00:35:23,500 We have tons of different initial conditions, blah, blah, blah, and 410 00:35:23,500 --> 00:35:25,060 there's just lots of ways to do this. 411 00:35:28,610 --> 00:35:30,592 And that's all I've got in there. 412 00:35:30,592 --> 00:35:34,537 So we're at 25 minutes. 413 00:35:34,537 --> 00:35:36,501 I guess I can talk a little bit about the simulations. 414 00:35:39,595 --> 00:35:46,634 So I have done the simulations in Python, 415 00:35:46,634 --> 00:35:50,067 Extracting the graphs from the point positions is the hard part. 416 00:35:50,067 --> 00:35:54,483 All methods for doing that have worst case running time of 417 00:35:54,483 --> 00:35:58,257 n squared if all the particles are at some point. 418 00:35:58,257 --> 00:36:01,360 We've used cell lists which are, 419 00:36:01,360 --> 00:36:06,612 basically it's a very simple thing, you just divide up, 420 00:36:09,349 --> 00:36:13,781 Space into a grid and you get 421 00:36:13,781 --> 00:36:18,810 a grid with length and radius. 422 00:36:18,810 --> 00:36:22,491 And so you just check each square and the squares around it, 423 00:36:22,491 --> 00:36:24,486 all points in those [INAUDIBLE]. 424 00:36:24,486 --> 00:36:28,170 And so that's pretty fast, that's used in molecular dynamics. 425 00:36:29,970 --> 00:36:33,580 >> This is you trying to get out of using the whole adjacency matrix? 426 00:36:33,580 --> 00:36:34,660 >> Yeah, exactly. 427 00:36:34,660 --> 00:36:39,781 So in the cases where you're forced to compute the whole adjacency matrix, 428 00:36:39,781 --> 00:36:41,311 then it's n squared. 429 00:36:41,311 --> 00:36:46,207 But if it's on torus, for example, and the density is basically constant, 430 00:36:46,207 --> 00:36:49,385 minutes, then that's the number of particles. 431 00:36:51,882 --> 00:36:56,172 There's one catch there, which is that when I said you have to check your cell 432 00:36:56,172 --> 00:36:57,756 and cell on all sides of you, 433 00:36:57,756 --> 00:37:01,480 that neighborhood increases dramatically with dimension. 434 00:37:01,480 --> 00:37:05,910 So this actually is not great in high dimensions. 435 00:37:05,910 --> 00:37:08,100 So the other method we've used is a spatial tree, 436 00:37:09,600 --> 00:37:11,940 there's few different kinds of those. 437 00:37:11,940 --> 00:37:16,090 KD tree is probably more standard in machine learning. 438 00:37:16,090 --> 00:37:18,639 We used an R tree which is more standard in geography. 439 00:37:18,639 --> 00:37:23,264 The R tree at the edge, that's self balancing. 440 00:37:23,264 --> 00:37:27,702 So with a KD tree you have to store all your points to begin with, and 441 00:37:27,702 --> 00:37:29,460 with an R tree you don't. 442 00:37:31,380 --> 00:37:35,316 And since we're rebuilding the network a ton of times, 443 00:37:35,316 --> 00:37:38,566 that take's us about half the time. 444 00:37:38,566 --> 00:37:39,791 That's that's how we get these networks. 445 00:37:42,491 --> 00:37:47,365 And that- >> Can you go down to your dimensionality 446 00:37:47,365 --> 00:37:48,290 images? 447 00:37:52,733 --> 00:37:56,350 If you had a dimension of six- >> Yeah. 448 00:37:56,350 --> 00:38:02,770 >> And you have 100 nodes, in the first one you're gonna get one every 0.6. 449 00:38:02,770 --> 00:38:08,579 And in two dimensions when you square it you're gonna get 1 every 3.6 square units. 450 00:38:08,579 --> 00:38:14,086 And in three dimensions you're gonna get 1 every 2.16 cube units. 451 00:38:14,086 --> 00:38:22,160 So it feels like your radius dimension didn't recalculate with the volume change. 452 00:38:22,160 --> 00:38:25,580 >> Yeah, that's somewhat true. 453 00:38:25,580 --> 00:38:29,330 So the results we're talking about here, that's a difference between 454 00:38:29,330 --> 00:38:33,380 what we're doing and the mathematical results on Brownian motion. 455 00:38:33,380 --> 00:38:36,290 So like I said, the Brownian motion thing, 456 00:38:36,290 --> 00:38:39,720 that sorta guarantees that these shouldn't happen in high dimensions. 457 00:38:39,720 --> 00:38:41,510 Doesn't depend on radius at all. 458 00:38:41,510 --> 00:38:42,210 >> Why not? 459 00:38:42,210 --> 00:38:45,030 If you're- >> Cuz they're not talking about radiuses, 460 00:38:45,030 --> 00:38:47,810 they're talking about exact collisions between particles. 461 00:38:48,970 --> 00:38:54,196 >> So we're saying those should never or almost never happen in high dimensions, 462 00:38:54,196 --> 00:38:57,786 but we do get a little of them because we have this radius 463 00:38:57,786 --> 00:38:59,672 >> Scroll down to the next two, 464 00:38:59,672 --> 00:39:04,704 did your radius grow dimensionally enough so it should, two dimensions and 465 00:39:04,704 --> 00:39:10,127 three dimensions shouldn't be so sparse compared to each other if you allowed for 466 00:39:10,127 --> 00:39:12,660 a larger radius because your volume. 467 00:39:12,660 --> 00:39:15,548 >> Yeah, yeah, I mean, so yeah, the radius is growing as a sphere and 468 00:39:15,548 --> 00:39:16,930 the volume is growing as cubed. 469 00:39:19,150 --> 00:39:20,790 So there is a big difference there. 470 00:39:22,880 --> 00:39:27,090 But there's, underlying that, so that's an artifact of what we're using this for. 471 00:39:27,090 --> 00:39:31,441 But underlying that there is a stronger result that basically these 472 00:39:31,441 --> 00:39:35,955 connections shouldn't be there [INAUDIBLE] cuz we have radius at all. 473 00:39:35,955 --> 00:39:37,388 So yeah, I definitely agree with what you're saying. 474 00:39:41,010 --> 00:39:42,417 If we turn the radius way down, 475 00:39:42,417 --> 00:39:45,080 we'd get something closer to the mathematical results. 476 00:39:46,780 --> 00:39:49,355 >> Yeah, I'm interested in turning radius up so 477 00:39:49,355 --> 00:39:53,855 that your five dimensional one is more congruent with your two dimensional one. 478 00:39:53,855 --> 00:39:56,090 >> Yeah, and I'd be curious to see what happened with it too. 479 00:39:56,090 --> 00:40:00,833 >> [INAUDIBLE] I think the appropriate thing to do would be to make 480 00:40:00,833 --> 00:40:02,417 the space smaller. 481 00:40:02,417 --> 00:40:05,745 You're not changing what kind of interaction is happening, 482 00:40:05,745 --> 00:40:09,137 you're saying they'd have to be events in order to interact. 483 00:40:09,137 --> 00:40:10,640 >> Yeah. >> Yeah. 484 00:40:10,640 --> 00:40:12,540 Great, I mean, you're right. 485 00:40:12,540 --> 00:40:15,735 You could just find the ratio difference from square to cube. 486 00:40:15,735 --> 00:40:16,320 >> Yeah. 487 00:40:16,320 --> 00:40:20,178 >> On your new dimensionality and multiply that by your radius and voila, 488 00:40:20,178 --> 00:40:24,115 you'd have something equivalent in terms of density of distribution. 489 00:40:24,115 --> 00:40:26,570 >> Yeah, [INAUDIBLE]. 490 00:40:26,570 --> 00:40:31,544 >> So, I know you talk about using different kinds of distance in your 2d 491 00:40:31,544 --> 00:40:36,720 model, but, it seems like you use, you're using Hamming distance. 492 00:40:36,720 --> 00:40:41,100 As you go up in dimensions you should get any more connections even or 493 00:40:41,100 --> 00:40:42,430 like dense connections. 494 00:40:42,430 --> 00:40:43,555 >> Yeah, I don't know. 495 00:40:43,555 --> 00:40:49,120 >> If we use Hamming distance then it makes the non model properties. 496 00:40:49,120 --> 00:40:52,942 We are crucially dependent on that being a Brownian motion. 497 00:40:52,942 --> 00:40:57,300 So, that's what makes it statistically 498 00:40:57,300 --> 00:41:00,203 useful eventually when you try to build a test for randomness. 499 00:41:00,203 --> 00:41:00,709 >> Yeah. 500 00:41:00,709 --> 00:41:05,029 >> They claiming that the metric distance based on a metric space, 501 00:41:05,029 --> 00:41:09,282 as in real space, Hamming distance and other kind of metrics. 502 00:41:09,282 --> 00:41:13,501 They don't ,they mess with the Brownian properties and therefore, 503 00:41:13,501 --> 00:41:17,160 you will not make, you can create all the networks you want. 504 00:41:17,160 --> 00:41:21,070 But the point is that we are kind of trying to figure out whether it 505 00:41:21,070 --> 00:41:22,210 makes sense. 506 00:41:22,210 --> 00:41:26,510 If it's the latest phase dynamics, it makes sense as a Brownian motion. 507 00:41:26,510 --> 00:41:28,180 >> Right? >> So, the Brownian motion distance and 508 00:41:28,180 --> 00:41:33,820 everything delays on the distances between points being 509 00:41:33,820 --> 00:41:38,950 classical Cartesian matrix B so we can add all in, that's always like a. 510 00:41:38,950 --> 00:41:40,080 >> So, yeah that, 511 00:41:40,080 --> 00:41:44,170 I guess like the the points in space like if you're still talking about. 512 00:41:44,170 --> 00:41:47,660 You're using Brownian motion to represent, and 513 00:41:47,660 --> 00:41:52,140 the whole point of this was talking about social network actors and monopoly. 514 00:41:52,140 --> 00:41:55,010 You're saying they're moving around some latent cultural space or 515 00:41:55,010 --> 00:42:01,280 some behavioral space, attribute space they're still randomly 516 00:42:01,280 --> 00:42:04,760 drifting around you're still following Brownian machine as they move around. 517 00:42:04,760 --> 00:42:09,710 And, we're just doing, if you're using the Hamming distance to establish links. 518 00:42:09,710 --> 00:42:10,500 I don't, I don't know. 519 00:42:10,500 --> 00:42:12,120 I guess I'm not sure I follow why. 520 00:42:12,120 --> 00:42:16,130 >> Yeah, it did mess up [INAUDIBLE], is 521 00:42:16,130 --> 00:42:20,070 like the model is as it is designed to be used with. 522 00:42:21,200 --> 00:42:21,700 >> Okay? >> So, and 523 00:42:21,700 --> 00:42:27,160 the reason it's designed to be used with us we are saying that everybody is 524 00:42:27,160 --> 00:42:31,760 thinking about all kinds of fancy proposed models. 525 00:42:31,760 --> 00:42:32,640 We are saying that, okay, 526 00:42:32,640 --> 00:42:36,740 the ridge structure, the generative to randomness is enough. 527 00:42:36,740 --> 00:42:44,720 So we're trying to say that what kind of noises can create, what? 528 00:42:44,720 --> 00:42:45,670 Okay, let's do it this way. 529 00:42:45,670 --> 00:42:50,100 We are trying to figure out with this, sound structure over and 530 00:42:50,100 --> 00:42:52,920 about a systematic randomness versus a systematic randomness. 531 00:42:54,590 --> 00:42:55,630 We really are. 532 00:42:55,630 --> 00:42:59,730 Is a Brownian motion and so when you're trying to build statistical tests, so 533 00:42:59,730 --> 00:43:04,670 we are going to have to conform to the natural metric space of 534 00:43:04,670 --> 00:43:08,360 the Brownian motion and the natural metric space is cognition. 535 00:43:08,360 --> 00:43:11,980 Although yes, we could kind of build all kinds of metrics. 536 00:43:11,980 --> 00:43:16,670 But if you want to say something more about, this is nice and 537 00:43:16,670 --> 00:43:22,300 statistically we want to do a type one type two error with a font of gold, 538 00:43:22,300 --> 00:43:25,090 then we have to rely on only one condition metrics. 539 00:43:25,090 --> 00:43:28,760 So, but if you're, if you don't really care about statistics, 540 00:43:28,760 --> 00:43:32,280 you just want to play around with any mistakes, 541 00:43:32,280 --> 00:43:37,700 which have no consequence, then you can use all kinds of other methods. 542 00:43:37,700 --> 00:43:42,250 But the point is that we have designed, if we say it's an unmodel, 543 00:43:42,250 --> 00:43:44,890 it's like it's the most boring model possible. 544 00:43:44,890 --> 00:43:47,540 There's no knowledge of that it's just random. 545 00:43:47,540 --> 00:43:54,363 So, then we have to respect the mathematical structure of the dynamics and 546 00:43:54,363 --> 00:43:58,806 the mathematical structure of dynamics demands 547 00:43:58,806 --> 00:44:03,263 that we use only the Cartesian metric distance. 548 00:44:03,263 --> 00:44:07,836 So other metric distance will mess up things and you could do it but 549 00:44:07,836 --> 00:44:11,240 we will not be, it will be more useless. 550 00:44:11,240 --> 00:44:13,870 >> Well, no, it could be more realistic. 551 00:44:13,870 --> 00:44:17,980 >> I'm hearing you say that because Brownian motion is in space, 552 00:44:17,980 --> 00:44:22,210 you have to use Cartesian distance to measure because they go together. 553 00:44:22,210 --> 00:44:23,290 >> Yeah. >> Right. 554 00:44:23,290 --> 00:44:24,720 >> And that's the whole point. 555 00:44:24,720 --> 00:44:25,800 We want to change the. 556 00:44:26,950 --> 00:44:27,520 >> Definition. 557 00:44:27,520 --> 00:44:34,435 >> Definition of metric, then the, why does the deep, Brownian motion is, 558 00:44:34,435 --> 00:44:39,860 a deep connections to the underlying topology and differentiable properties. 559 00:44:39,860 --> 00:44:42,170 So, you're saying that if you're changing the metric, 560 00:44:42,170 --> 00:44:44,060 you're changing the calculus and everything. 561 00:44:44,060 --> 00:44:46,578 >> Sure. >> So therefore, you got to do everything 562 00:44:46,578 --> 00:44:51,480 you got to redefine your calculus for a different metric space and so on. 563 00:44:51,480 --> 00:44:56,400 And so, then you cannot get back to saying concrete things about, 564 00:44:56,400 --> 00:44:58,220 okay, is it really a random structure? 565 00:44:58,220 --> 00:45:00,860 So that's, really like you're saying that if it's, 566 00:45:00,860 --> 00:45:05,080 we are assuming Cartesian Leighton space that's like a depot and 567 00:45:05,080 --> 00:45:09,180 we are assuming kind of a Cartesian geometry, Cartesian topology. 568 00:45:09,180 --> 00:45:12,800 So, the natural metric spaces, square root of x y xa. 569 00:45:12,800 --> 00:45:15,130 >> Yeah, that being 6x squared. 570 00:45:15,130 --> 00:45:16,130 So, that's the. 571 00:45:16,130 --> 00:45:18,370 >> But, you put spaces. 572 00:45:18,370 --> 00:45:23,170 But, the point is that you're trying to keep, 573 00:45:23,170 --> 00:45:28,018 guide our play to the role that this is nice. 574 00:45:28,018 --> 00:45:30,030 And that it's got to do with, 575 00:45:30,030 --> 00:45:33,050 we're going to build some statistical tests eventually. 576 00:45:33,050 --> 00:45:37,350 So that's the, they need to have good statistical tests for, 577 00:45:37,350 --> 00:45:39,000 to calculate type one and type two error. 578 00:45:39,000 --> 00:45:42,800 You need good metric space and good topology. 579 00:45:42,800 --> 00:45:46,322 So, if you're gonna to go to Hamming, you lose everything, 580 00:45:46,322 --> 00:45:50,210 then becomes,as is the model is useless and become boring as well. 581 00:45:50,210 --> 00:45:54,760 >> Well, can you summarize what he's saying about why you're doing it? 582 00:45:54,760 --> 00:45:56,790 What is it for? 583 00:45:56,790 --> 00:46:00,300 >> Thought this unfamiliar, maybe this is, maybe you answered this or 584 00:46:00,300 --> 00:46:03,780 maybe this was a question from Alice. 585 00:46:03,780 --> 00:46:06,940 Have you tested any non random dynamic network 586 00:46:06,940 --> 00:46:09,069 models against this [INAUDIBLE] model? 587 00:46:09,069 --> 00:46:11,530 >> Not yet. >> That's the goal, right is to. 588 00:46:11,530 --> 00:46:13,010 >> Yeah, >> This is your base you 589 00:46:13,010 --> 00:46:14,375 can compare things to, 590 00:46:14,375 --> 00:46:18,965 what, do you have an expectation of how to go get to work out metrics first, or. 591 00:46:18,965 --> 00:46:21,480 >> Yeah, so we do have to work out some stuff. 592 00:46:23,270 --> 00:46:27,960 >> We show the, the testing coefficient preamble again ad hoc testing. 593 00:46:27,960 --> 00:46:32,930 >> Yeah. So, we kind of suspect that this behavior 594 00:46:32,930 --> 00:46:35,972 will be most useful there. 595 00:46:39,746 --> 00:46:42,000 You can predict the average clustering coefficient, 596 00:46:42,000 --> 00:46:43,820 just from the properties of static graphs. 597 00:46:43,820 --> 00:46:48,160 It doesn't actually save you much time because you still have to compute a bunch 598 00:46:48,160 --> 00:46:48,948 of networks. 599 00:46:48,948 --> 00:46:54,903 But, we expect that deviation from this will be pretty easy to measure. 600 00:46:57,104 --> 00:47:01,500 But, that depends on various assumptions about the weight in space. 601 00:47:01,500 --> 00:47:05,010 So, every ecology will bring its own assumptions and 602 00:47:05,010 --> 00:47:07,690 of course brings the assumption that the space is closed and 603 00:47:07,690 --> 00:47:15,140 you can move far enough in one direction to wind up clear to the end. 604 00:47:17,940 --> 00:47:22,114 >> So, in and I was thinking this when you made the comparison of this to molecular 605 00:47:22,114 --> 00:47:23,330 dynamics models. 606 00:47:23,330 --> 00:47:28,110 So, if you have a model where you 607 00:47:28,110 --> 00:47:32,250 have levers on the particles of an attractive motion and 608 00:47:32,250 --> 00:47:36,870 the repulsive motion and the random motion, you can vary independently. 609 00:47:36,870 --> 00:47:39,872 You got a null model for a lot of different, 610 00:47:39,872 --> 00:47:43,200 situation that are possible things. 611 00:47:43,200 --> 00:47:44,142 >> Yeah. >> Or you could connect this 612 00:47:44,142 --> 00:47:45,300 to various other models. 613 00:47:45,300 --> 00:47:47,460 >> Yeah and that would be like a more detailed version. 614 00:47:47,460 --> 00:47:52,595 There's lots of things it could be, you could have people's opinion wizard 615 00:47:52,595 --> 00:47:57,576 influencing, you have a gravity thing and you know, yeah, definitely. 616 00:47:57,576 --> 00:48:03,759 >> [INAUDIBLE] Hypothetically let's say you have say E for the five dimensions, 617 00:48:03,759 --> 00:48:10,620 let's say we know from domain knowledge that social psychology function. 618 00:48:10,620 --> 00:48:13,890 That's like five dimensional latent space. 619 00:48:13,890 --> 00:48:18,380 And the altitudes do more around in this five dimensional space, right? 620 00:48:18,380 --> 00:48:23,500 So then our null model, say that practical unit you say 621 00:48:23,500 --> 00:48:28,290 near intersections, that's hardly gonna be any connection between any anybody else. 622 00:48:28,290 --> 00:48:33,160 So again, the the null model random model gives you a possible, okay? 623 00:48:33,160 --> 00:48:37,560 But suppose in data we actually see a dense network 624 00:48:37,560 --> 00:48:42,660 in a latent space with high dimensional homophily then we can rule it via kind of 625 00:48:42,660 --> 00:48:46,280 distinctly seeing that there's actually some underlying attractive, 626 00:48:46,280 --> 00:48:50,960 it's not about Brownian in nature, that's something concrete going on there. 627 00:48:50,960 --> 00:48:51,850 So therefore, 628 00:48:51,850 --> 00:48:56,500 we could say that the null model is safe like zero clustering coefficient. 629 00:48:56,500 --> 00:49:00,830 And you're saying that this particular empirically measured by 630 00:49:02,180 --> 00:49:07,590 temporal network is signal nonzero statistically, therefore, we would claim 631 00:49:07,590 --> 00:49:12,532 that a that's something apart from an over and above random a different thing there. 632 00:49:12,532 --> 00:49:16,450 So that's the, and therefore, you're making something substantive. 633 00:49:16,450 --> 00:49:20,048 You're kind of increasing the knowledge from random to something significant. 634 00:49:20,048 --> 00:49:22,020 >> So that's, >> Okay, 635 00:49:22,020 --> 00:49:26,180 I guess my question is, I still like to just keep going back to this but 636 00:49:26,180 --> 00:49:28,360 it seems like there's two pieces to this right? 637 00:49:28,360 --> 00:49:34,090 There's a piece where we're putting our points in this latent space and 638 00:49:34,090 --> 00:49:37,564 we want them to drift around, we want there to be random variation, so 639 00:49:37,564 --> 00:49:40,490 people's altitudes are changing. 640 00:49:40,490 --> 00:49:45,290 And that makes sense, and I understand why you need to use geometric distance for 641 00:49:45,290 --> 00:49:47,370 them to be moving in that space. 642 00:49:47,370 --> 00:49:51,010 But then what I would say is like the network piece, 643 00:49:51,010 --> 00:49:55,470 which points are linked to each other point seems like a separate thing. 644 00:49:55,470 --> 00:49:58,350 >> Yeah, >> It doesn't, you don't necessarily need 645 00:49:58,350 --> 00:50:03,660 to follow geometric distance for those points to still be behaving according 646 00:50:03,660 --> 00:50:08,400 to like a Brownian motion, so you have the points moving in a Brownian motion and 647 00:50:08,400 --> 00:50:11,970 then you using another distance calculation to form the network. 648 00:50:15,290 --> 00:50:18,450 You're still getting this random that attribute you're still being able to say 649 00:50:18,450 --> 00:50:23,060 you'd still no model for, there's no reason for the attributes changing. 650 00:50:23,060 --> 00:50:28,880 They're just changing randomly, but the network is, maybe closer to what? 651 00:50:28,880 --> 00:50:30,170 >> Yeah, that is correct. 652 00:50:30,170 --> 00:50:35,450 But the point is that it's the same reason say for example, 653 00:50:35,450 --> 00:50:41,200 why do people for example, people when you measure similarity between graphs right? 654 00:50:41,200 --> 00:50:45,857 So people use adjacency matrix equal to say hamming distance 655 00:50:45,857 --> 00:50:49,601 on the adjacency matrix differences in zeros and 656 00:50:49,601 --> 00:50:54,090 ones, between your compare graph 1 to graph 2, right? 657 00:50:54,090 --> 00:50:59,514 We want adjacency AIJ 1 and you got AIJ 2, it would say, 658 00:50:59,514 --> 00:51:05,720 let's create a similarity metric between AIJ 1 minus AIJ 2. 659 00:51:05,720 --> 00:51:10,660 It could be isometric, but the point is that this particular metric 660 00:51:10,660 --> 00:51:15,430 of difference between two graphs is kind of totally useless, and 661 00:51:15,430 --> 00:51:21,010 that's the reason mathematically it's not a well-defined similarity measure. 662 00:51:21,010 --> 00:51:24,080 So that's the reason people use this Laplacian. 663 00:51:24,080 --> 00:51:28,490 The Laplacian of a graph, it needs more real value. 664 00:51:28,490 --> 00:51:32,240 So the Laplacian of a graph actually may make sense and 665 00:51:32,240 --> 00:51:38,380 you can compare two graph Laplacian says, okay, Laplacian one minus Laplacian two. 666 00:51:38,380 --> 00:51:43,190 So, that's the reason whenever you're looking at, for example, FMRI brain 667 00:51:43,190 --> 00:51:47,035 connectivity networks, they actually use Laplacian, they will likely use adjacency, 668 00:51:47,035 --> 00:51:54,136 and the Laplacian also relies on this underlying smooth, space. 669 00:51:54,136 --> 00:51:57,620 So it's possible but then when you can, 670 00:51:57,620 --> 00:52:03,240 it's okay maybe you could kind of go this is like saying agent based modelling and 671 00:52:03,240 --> 00:52:07,400 you can put all kind of net lower you can look up anything you want. 672 00:52:07,400 --> 00:52:12,290 You can have fun, but if you wanna do statistics, he got a metric display. 673 00:52:12,290 --> 00:52:15,393 I'm not stopping anybody to do any other metric. 674 00:52:15,393 --> 00:52:19,690 You could kind of create any kind of temporal networks with any kind of metrics 675 00:52:19,690 --> 00:52:24,790 with if you want, but if you want to say that how are you going to distinguish 676 00:52:24,790 --> 00:52:30,150 randomness from you have to rely on the admission matrix. 677 00:52:30,150 --> 00:52:33,330 That's all, from simulation purposes you could go 678 00:52:34,520 --> 00:52:37,458 wild with any kind of metric you want, no problem. 679 00:52:37,458 --> 00:52:39,920 >> I don't think that's where he was pointing at you.- 680 00:52:39,920 --> 00:52:41,129 >> But you could change 681 00:52:41,129 --> 00:52:44,000 the hamming [INAUDIBLE] Distance, right? 682 00:52:44,000 --> 00:52:48,800 That your kind of thing is meaningful and that hamming distance is going to be 683 00:52:48,800 --> 00:52:55,135 the basis for your type one, type two test,or any other disconnected test. 684 00:52:55,135 --> 00:52:57,525 >> That's about how you're defining modeling. 685 00:52:57,525 --> 00:52:58,265 Like what is the modeling? 686 00:52:58,265 --> 00:52:58,903 >> Yeah. >> So if 687 00:52:58,903 --> 00:53:02,149 you're talking about the geometric distances, 688 00:53:02,149 --> 00:53:06,718 the only distance [CROSSTALK] some sum over all the cultural features. 689 00:53:06,718 --> 00:53:12,503 >> So that means you could [INAUDIBLE] distance through the Hamming distances and 690 00:53:12,503 --> 00:53:17,487 everything, you could come up with dynamic networks all claiming 691 00:53:17,487 --> 00:53:22,471 is that the ulterior motive we have here will not be sensitive if you 692 00:53:22,471 --> 00:53:27,020 want to get generate like for example, you're nobody. 693 00:53:27,020 --> 00:53:29,741 You're not going to consult a sociologist before 694 00:53:29,741 --> 00:53:32,470 you do a sociology simulation in net lower right? 695 00:53:32,470 --> 00:53:34,720 You just do whatever you want. 696 00:53:34,720 --> 00:53:40,420 So it says same way you do whatever you want in the same way that you want to say, 697 00:53:40,420 --> 00:53:45,000 we got to say that is a meaningful statisticians who take it seriously, 698 00:53:45,000 --> 00:53:49,500 then it's like the limitation of the model instead of the model is that 699 00:53:49,500 --> 00:53:51,220 they stick to one metrics. 700 00:53:51,220 --> 00:53:55,470 That's otherwise you could do any kind of metrics. 701 00:53:55,470 --> 00:53:58,580 >> Let's put point pin in that for a sec, because I think we agree. 702 00:53:58,580 --> 00:54:03,150 What I think he's asking and I wanna know is you have a set of 703 00:54:03,150 --> 00:54:08,460 links in your adjacency matrix, but we didn't care we didn't use them. 704 00:54:08,460 --> 00:54:15,380 We pointed out where they were similar on say a three axis attribute, 705 00:54:15,380 --> 00:54:19,560 this node is red and this other node is red, and that puts them near each other. 706 00:54:19,560 --> 00:54:22,770 And this node is blue and that other nodes new and that's the latent space, and 707 00:54:22,770 --> 00:54:29,660 we're making a relationship and network out of their location in the latent space. 708 00:54:29,660 --> 00:54:33,780 Is the connections in the adjacency matrix useful at all? 709 00:54:33,780 --> 00:54:35,000 Does it come into play? 710 00:54:35,000 --> 00:54:39,790 >> So that, yeah, so the when we determined that nodes are near each 711 00:54:39,790 --> 00:54:42,920 other that creates these two matrix sets. 712 00:54:42,920 --> 00:54:44,450 >> Wait, tell me that again though. 713 00:54:44,450 --> 00:54:47,581 You started with your own- [CROSSTALK] >> Matrix okay, you're not 714 00:54:47,581 --> 00:54:48,403 >> Right.- [CROSSTALK] 715 00:54:48,403 --> 00:54:51,427 >> So we only made by similarities and 716 00:54:51,427 --> 00:54:52,582 attribute. 717 00:54:52,582 --> 00:54:56,115 [CROSSTALK] >> The distances I think. 718 00:54:56,115 --> 00:54:56,766 >> Yes. 719 00:54:56,766 --> 00:54:58,050 >> So this equation here. 720 00:54:58,050 --> 00:54:59,390 >> Wonderful, I get it. 721 00:54:59,390 --> 00:55:02,969 I just thought you had a network to start but you don't, 722 00:55:02,969 --> 00:55:07,680 you have some nodes whose attributes may form behind a fuzzy network. 723 00:55:07,680 --> 00:55:11,032 >> Sorry, I mentioned earlier there are other people who think they have methods 724 00:55:11,032 --> 00:55:13,686 for starting with the network and again when space from that. 725 00:55:13,686 --> 00:55:14,320 >> Yeah. 726 00:55:14,320 --> 00:55:18,970 >> I think, before you're asking for 727 00:55:18,970 --> 00:55:24,241 a sort of summary or example, >> Yeah, the what support. 728 00:55:24,241 --> 00:55:29,251 >> So what I've been thinking of lately is this data 729 00:55:29,251 --> 00:55:34,512 set that was a some sort of conference in 2002. 730 00:55:34,512 --> 00:55:38,202 And a bunch of participants wore sensors so they could tell when they were near 731 00:55:38,202 --> 00:55:41,830 each other and you created a network data set from that. 732 00:55:41,830 --> 00:55:47,900 And so you'd expect, let's say it's building space, there is physical space. 733 00:55:47,900 --> 00:55:51,100 So, people are moving around in two or three dimensions. 734 00:55:51,100 --> 00:55:53,320 And they're near each other or not. 735 00:55:53,320 --> 00:55:55,358 But also [CROSSTALK] from walls. 736 00:55:55,358 --> 00:55:58,580 But was that >> But also they have topics special? 737 00:55:58,580 --> 00:55:59,120 >> Yeah, exactly. 738 00:55:59,120 --> 00:56:01,773 So they'll be sorta non random, 739 00:56:01,773 --> 00:56:06,870 very non random aggregations of people >> You do hang out with your specialty, 740 00:56:06,870 --> 00:56:09,948 maybe get away from your specialty and hang out with everybody else. 741 00:56:09,948 --> 00:56:14,764 Yeah, and so there's all those ways in which people moving around 742 00:56:14,764 --> 00:56:19,666 the conference are not random we can measure by considering the case 743 00:56:19,666 --> 00:56:22,685 where they were moving around randomly. 744 00:56:22,685 --> 00:56:24,880 >> Yeah, as a thing to measure against. 745 00:56:24,880 --> 00:56:31,022 >> Yeah. 746 00:56:31,022 --> 00:56:33,080 >> The poster graph that we said that maybe looks like a highway map of Oregon. 747 00:56:33,080 --> 00:56:37,880 When that came up, it made me think of [COUGH] you may have seen someone compare 748 00:56:37,880 --> 00:56:41,930 like, here's an image of the super clusters in the universe and 749 00:56:41,930 --> 00:56:44,558 here's an image of neurons in the brain. 750 00:56:44,558 --> 00:56:47,403 >> [LAUGH] >> And ooh, 751 00:56:47,403 --> 00:56:48,800 isn't it neat how they look so similar. 752 00:56:48,800 --> 00:56:49,369 >> Right. 753 00:56:49,369 --> 00:56:51,762 >> And I think, well, that's just how things look. 754 00:56:51,762 --> 00:56:52,861 >> That are random. 755 00:56:52,861 --> 00:56:56,184 >> [LAUGH] >> There's this kind of physics roughly 756 00:56:56,184 --> 00:56:59,201 between them and astronauts and they move around. 757 00:56:59,201 --> 00:57:01,449 They could just be random so you need to compare it. 758 00:57:01,449 --> 00:57:07,152 >> I think this network analysis should not be done with images. 759 00:57:07,152 --> 00:57:08,839 It should be just done with numbers like that. 760 00:57:08,839 --> 00:57:14,276 It's really easy to see them, although it cannot be, we're kind of 761 00:57:14,276 --> 00:57:19,297 using this to exploit the image, I think- >> I think with the numbers, you would 762 00:57:19,297 --> 00:57:23,007 still detect some sort of cluster here and might still think it was non-random. 763 00:57:23,007 --> 00:57:25,512 >> Yeah, I mean, the image is useful, I think, 764 00:57:25,512 --> 00:57:28,295 in the sense of these things look similar to me why? 765 00:57:28,295 --> 00:57:29,650 How do I quantify it? 766 00:57:29,650 --> 00:57:34,896 And if you can do that and then say, but whatever quantities I come up with 767 00:57:34,896 --> 00:57:40,560 are still apparent in null models then maybe these are really special. 768 00:57:40,560 --> 00:57:46,123 >> Yeah, I think at least my goal has 769 00:57:46,123 --> 00:57:52,695 been always to kind of create. 770 00:57:52,695 --> 00:57:55,877 How much can you say in the null model, that's like you have no knowledge and 771 00:57:55,877 --> 00:57:56,983 it captures everything. 772 00:57:56,983 --> 00:58:00,274 Then [INAUDIBLE]. 773 00:58:00,274 --> 00:58:06,517 So you gotta have strong philosophically, you got to go to alternate hypothesis. 774 00:58:06,517 --> 00:58:07,541 It should be very strong. 775 00:58:07,541 --> 00:58:10,754 That's why I like that we got growth of scientific knowledge or 776 00:58:10,754 --> 00:58:12,216 even go to the alternative. 777 00:58:12,216 --> 00:58:18,630 But until that time, don't think that you cannot acknowledge like so 778 00:58:18,630 --> 00:58:24,662 to speak to [INAUDIBLE] >> I'm gonna [INAUDIBLE] thank Jeff for 779 00:58:24,662 --> 00:58:26,022 speaking. 780 00:58:26,022 --> 00:58:27,377 >> Yeah, great. 781 00:58:27,377 --> 00:58:33,614 >> [APPLAUSE] >> [INAUDIBLE] questions 782 00:58:33,614 --> 00:58:39,347 [INAUDIBLE] that people have [INAUDIBLE] so 783 00:58:39,347 --> 00:58:43,555 [INAUDIBLE] >> Yeah, you have to get up, please. 784 00:58:43,555 --> 00:58:44,772 >> These are all graphs. 785 00:58:44,772 --> 00:58:49,393 Are there any value over when we're looking at the hyper graphs? 786 00:58:49,393 --> 00:58:54,768 >> Yeah, so that's one of the [CROSSTALK] >> Okay. 787 00:58:54,768 --> 00:58:55,987 >> So that's like the hyper, 788 00:58:55,987 --> 00:58:58,749 that's like the more cooler way of saying hyper aggressive. 789 00:58:58,749 --> 00:58:59,334 >> Yeah. 790 00:58:59,334 --> 00:59:04,251 >> But we did talk about the video of it and then instead of v equal to 4 be 791 00:59:04,251 --> 00:59:08,069 the cut off, I could have v equal to 3 as the cut off. 792 00:59:08,069 --> 00:59:13,552 So after 3 [INAUDIBLE] you will not have a problem with [INAUDIBLE]. 793 00:59:13,552 --> 00:59:19,350 So that's like [INAUDIBLE]. 794 00:59:19,350 --> 00:59:23,061 >> Yeah, so this is the equivalent of a random integer graph or 795 00:59:23,061 --> 00:59:24,479 sequential complex. 796 00:59:24,479 --> 00:59:27,873 So, all three of these nodes are within each other so 797 00:59:27,873 --> 00:59:30,659 they got a triangle between them. 798 00:59:30,659 --> 00:59:35,510 >> Yeah, the interesting part is the math 799 00:59:35,510 --> 00:59:40,850 is really well settled on these things. 800 00:59:40,850 --> 00:59:43,158 So whenever we do simulation, 801 00:59:43,158 --> 00:59:46,897 we can actually say whether it makes sense or not. 802 00:59:46,897 --> 00:59:51,694 So that's one of the things that they're pretty sure that what we're 803 00:59:51,694 --> 00:59:52,843 doing is right. 804 00:59:52,843 --> 00:59:54,993 And it also kind of gives us room for 805 00:59:54,993 --> 01:00:00,420 publish more papers because there are places where the math is not set to data. 806 01:00:00,420 --> 01:00:05,070 Then we can just simulate and see what intuitions we get for this regard. 807 01:00:05,070 --> 01:00:07,960 Interesting thing for be less than four. 808 01:00:07,960 --> 01:00:11,950 >> So where are you out on the March 4th publication? 809 01:00:11,950 --> 01:00:16,472 >> So we just basically crammed all this week to get the conference 810 01:00:16,472 --> 01:00:18,020 fast-track ready. 811 01:00:18,020 --> 01:00:20,040 >> And then a paper has to be submitted. 812 01:00:20,040 --> 01:00:23,919 When is that conference, where is that conference? 813 01:00:23,919 --> 01:00:25,159 >> May. 814 01:00:25,159 --> 01:00:26,491 >> It's in Vermont. 815 01:00:26,491 --> 01:00:27,083 >> Vermont. 816 01:00:27,083 --> 01:00:29,728 >> So they have this international network, so for 817 01:00:29,728 --> 01:00:34,440 the conference they alternate between Europe and United States every year. 818 01:00:34,440 --> 01:00:36,970 So luckily for us, it's here. 819 01:00:36,970 --> 01:00:38,770 >> Are you going? 820 01:00:38,770 --> 01:00:42,357 >> No, it's not accepted [CROSSTALK] >> I mean I they accept it, 821 01:00:42,357 --> 01:00:43,766 you'll physically go to Vermont? 822 01:00:43,766 --> 01:00:44,832 >> I think so. 823 01:00:44,832 --> 01:00:47,713 >> [LAUGH] >> Yesterday. 824 01:00:47,713 --> 01:00:53,137 >> [LAUGH] >> Basically we use the abstract to push 825 01:00:53,137 --> 01:00:58,300 out as to create an outline for a paper so that we have. 826 01:00:58,300 --> 01:01:04,430 So we'll probably do it after I have more pressing things. 827 01:01:05,750 --> 01:01:09,869 Maybe in May, by the time you go to the conference the paper might be ready. 828 01:01:09,869 --> 01:01:11,934 >> [LAUGH] >> What's the conference again? 829 01:01:11,934 --> 01:01:14,402 >> [INAUDIBLE] >> Uh-huh, 830 01:01:14,402 --> 01:01:16,973 and it is on- >> Right, I actually might be giving 831 01:01:16,973 --> 01:01:18,965 lecture that week, so- >> [CROSSTALK] 832 01:01:18,965 --> 01:01:21,941 >> think 27 to June 1st, 833 01:01:21,941 --> 01:01:24,644 like last week of May. 834 01:01:24,644 --> 01:01:28,565 They got to two days of tutorials and three days of conference. 835 01:01:28,565 --> 01:01:32,758 >> He went to a conferences last week at astrophysics and 836 01:01:32,758 --> 01:01:38,450 physics conference in Seattle at ANS and in it's a five day conference. 837 01:01:38,450 --> 01:01:42,604 It happens in Seattle once every four years, most around the country. 838 01:01:42,604 --> 01:01:47,800 But it's pretty cool, world-class physicists and 839 01:01:47,800 --> 01:01:54,388 all kinds of amazing connections that you make and great lectures. 840 01:01:54,388 --> 01:01:59,020 >> Yeah, I'm pretty excited about it. 841 01:02:02,391 --> 01:02:02,995 >> Cool. 842 01:02:05,534 --> 01:02:10,250 >> [LAUGH] 843 01:02:10,250 --> 01:02:15,386 >> Any use is purely accidental. 844 01:02:15,386 --> 01:02:19,639 >> I also enjoy the dots and lines moving around the screen. 845 01:02:19,639 --> 01:02:22,895 [CROSSTALK] >> Is that D3 you were using this under? 846 01:02:22,895 --> 01:02:23,804 >> Yeah. 847 01:02:23,804 --> 01:02:27,500 >> One of these days you probably use a deployment log, 848 01:02:27,500 --> 01:02:29,605 everybody excited about it. 849 01:02:29,605 --> 01:02:32,023 >> Yeah, maybe. 850 01:02:32,023 --> 01:02:34,142 >> Library's last >> What's that? 851 01:02:34,142 --> 01:02:35,917 >> Lifespan, a natural lifespan and 852 01:02:35,917 --> 01:02:38,862 by the time you're doing that it won't be as [INAUDIBLE]. 853 01:02:38,862 --> 01:02:40,210 >> That's true. >> Yeah, we'll all be on [INAUDIBLE] 854 01:02:40,210 --> 01:02:40,753 >> Yeah, 855 01:02:40,753 --> 01:02:45,378 D3 is now considered like sort of low level visualisation. 856 01:02:45,378 --> 01:02:48,257 >> What's next, processing JS? 857 01:02:48,257 --> 01:02:49,122 >> I don't think it's processing. 858 01:02:49,122 --> 01:02:52,018 There's like a lot of libraries that like have premade me graph visualizations 859 01:02:52,018 --> 01:02:52,587 to integrate. 860 01:02:52,587 --> 01:02:55,227 >> What did you call the dark tourist? 861 01:02:55,227 --> 01:02:56,669 It was a couple slides above. 862 01:02:56,669 --> 01:03:02,354 [CROSSTALK] >> Is your talk 863 01:03:02,354 --> 01:03:09,261 link in the seminar, 864 01:03:09,261 --> 01:03:14,640 can we see that? 865 01:03:14,640 --> 01:03:16,805 >> Well it's on mine. 866 01:03:16,805 --> 01:03:21,333 [CROSSTALK] >> If you wants to share the slides, 867 01:03:21,333 --> 01:03:22,132 he can. 868 01:03:22,132 --> 01:03:22,909 >> Yeah. 869 01:03:22,909 --> 01:03:24,294 >> Absolutely, that's fine. 870 01:03:24,294 --> 01:03:25,130 >> That'd be great. 871 01:03:25,130 --> 01:03:26,448 >> Bit of a let down. 872 01:03:26,448 --> 01:03:27,295 >> Yeah. 873 01:03:27,295 --> 01:03:32,226 >> [COUGH] >> Here I'll start it up then. 874 01:03:32,226 --> 01:03:35,106 >> [LAUGH] >> I didn't like it from the semi-page. 875 01:03:35,106 --> 01:03:40,510 >> Well, I did a similar thing in the NetLogo 876 01:03:40,510 --> 01:03:44,534 last semester for ABS Look Up. 877 01:03:44,534 --> 01:03:46,546 Ran a better riot model and 878 01:03:46,546 --> 01:03:52,233 just randomly distributed all the nodes in the time slot in the 2D space. 879 01:03:52,233 --> 01:03:56,783 And then based on a radius that you can change, that would build a random graph 880 01:03:56,783 --> 01:04:00,219 and then do a perturbation and see if it creates a cascade. 881 01:04:00,219 --> 01:04:02,562 >> What kind of cascade? 882 01:04:02,562 --> 01:04:06,496 >> Like a threshold cascade model sort of thing. 883 01:04:07,996 --> 01:04:10,225 >> Yeah, it's like the riot model. 884 01:04:10,225 --> 01:04:10,894 Once one person starts rioting, everyone else has like 885 01:04:10,894 --> 01:04:14,314 Everyone's got a threshold of how many people are they observing or 886 01:04:14,314 --> 01:04:18,038 they're connected to that are like what proportion of that are rioting. 887 01:04:18,038 --> 01:04:21,382 If that's higher than their threshold then they start rioting or 888 01:04:21,382 --> 01:04:22,974 believing the level changes. 889 01:04:22,974 --> 01:04:23,824 >> Upholding the leveling. >> Yeah, yeah, exactly. 890 01:04:23,824 --> 01:04:28,535 It's, he uses riot, a riot because it's very just realize it. 891 01:04:28,535 --> 01:04:30,038 >> It was comical at the time. 892 01:04:30,038 --> 01:04:33,902 >> Right, and then there's but they could use for 893 01:04:33,902 --> 01:04:39,905 any number of things like if you see enough people with different phones. 894 01:04:39,905 --> 01:04:44,138 >> Yeah I think even there like it was like done by Duncan worth, 895 01:04:44,138 --> 01:04:47,790 like the one that we discussed in Denver last night. 896 01:04:47,790 --> 01:04:49,887 So they're even there. 897 01:04:49,887 --> 01:04:52,773 There was this book by Malcolm Gladwell. 898 01:04:52,773 --> 01:04:56,830 He had this pop psychology theories of how things read. 899 01:04:56,830 --> 01:05:00,843 And Duncan was actually told pure random, just by chance. 900 01:05:00,843 --> 01:05:07,032 That kind of network with no centrality can produce the same kind of spread so 901 01:05:07,032 --> 01:05:13,623 they can disprove the few of the exam that a non model can produce yield behavior. 902 01:05:13,623 --> 01:05:18,357 So the argument is that the whole point of that book was kind of flawed so. 903 01:05:18,357 --> 01:05:20,089 >> Which one? 904 01:05:20,089 --> 01:05:22,830 Blinker? Tipping point? >> Tipping point. 905 01:05:22,830 --> 01:05:25,146 >> Yes, he wrote a great article about it, too. 906 01:05:25,146 --> 01:05:28,805 >> I mean, tipping point is just a pop psychology of catastrophe right? 907 01:05:31,490 --> 01:05:36,355 >> Yes, it's really just that systems have 908 01:05:36,355 --> 01:05:40,534 thresholds and then their hooray. 909 01:05:40,534 --> 01:05:50,534 >> [CROSSTALK]