0 00:00:00,000 --> 00:00:30,000 Dear viewer, these subtitles were generated by a machine via the service Trint and therefore are (very) buggy. If you are capable, please help us to create good quality subtitles: https://c3subtitles.de/talk/176 Thanks! 1 00:00:09,860 --> 00:00:11,959 Hi. We are Anna and Dominic, 2 00:00:11,960 --> 00:00:13,639 and we're here to tell you something 3 00:00:13,640 --> 00:00:16,369 about brain computer interfaces. 4 00:00:16,370 --> 00:00:18,649 I'm a mathematician and he's a cognitive 5 00:00:18,650 --> 00:00:21,919 scientist, so we are in this topic 6 00:00:21,920 --> 00:00:24,349 first. We'll tell you why we are excited 7 00:00:24,350 --> 00:00:25,729 about this topic. 8 00:00:25,730 --> 00:00:27,229 Then we'll dove into the technical 9 00:00:27,230 --> 00:00:29,779 details, how data is represented 10 00:00:29,780 --> 00:00:31,849 inside of the brain, how to 11 00:00:31,850 --> 00:00:34,039 read it out with the computer and how to 12 00:00:34,040 --> 00:00:35,689 build your own BCI. 13 00:00:35,690 --> 00:00:37,219 And finally, we'll have a quick 14 00:00:37,220 --> 00:00:39,080 discussion and answer some questions. 15 00:00:43,200 --> 00:00:45,839 So let's talk about communication. 16 00:00:45,840 --> 00:00:48,149 Language is the most natural 17 00:00:48,150 --> 00:00:49,679 way to communicate with your fellow 18 00:00:49,680 --> 00:00:50,789 humans. 19 00:00:50,790 --> 00:00:53,159 It's the most intuitive way of expressing 20 00:00:53,160 --> 00:00:54,329 your thoughts. 21 00:00:54,330 --> 00:00:56,399 And also the most common method 22 00:00:56,400 --> 00:00:59,069 to communicate with your computer. 23 00:00:59,070 --> 00:01:01,259 Obviously, the word is very powerful, 24 00:01:02,910 --> 00:01:04,979 but which devices can we use to put 25 00:01:04,980 --> 00:01:07,619 diverse from our minds into a computer? 26 00:01:07,620 --> 00:01:10,049 Of course, there's the keyboard 27 00:01:10,050 --> 00:01:11,609 compares to talking. 28 00:01:11,610 --> 00:01:13,829 It's quite slow and 29 00:01:13,830 --> 00:01:16,349 quite hard to learn, but extremely 30 00:01:16,350 --> 00:01:17,670 accurate by design. 31 00:01:19,620 --> 00:01:21,359 There are all kinds of motion tracking 32 00:01:21,360 --> 00:01:22,469 tools. 33 00:01:22,470 --> 00:01:23,470 The mouse. 34 00:01:25,210 --> 00:01:27,459 The touchpad touch screen graphic 35 00:01:27,460 --> 00:01:29,739 tablet, because we want text 36 00:01:29,740 --> 00:01:32,439 input. We knew to use handwriting 37 00:01:32,440 --> 00:01:34,629 higher information density 38 00:01:34,630 --> 00:01:36,909 easier to use than a keyboard, but very 39 00:01:36,910 --> 00:01:39,310 slow at actually transferring a message. 40 00:01:42,390 --> 00:01:44,189 Then we have voice input. 41 00:01:44,190 --> 00:01:46,469 Everybody knows that dictating as much 42 00:01:46,470 --> 00:01:47,550 profit and writing 43 00:01:49,320 --> 00:01:51,029 since a couple of years, even our 44 00:01:51,030 --> 00:01:53,579 computers understand what we say. 45 00:01:53,580 --> 00:01:55,289 This could become very interesting in the 46 00:01:55,290 --> 00:01:57,509 long run because 47 00:01:57,510 --> 00:01:59,819 talking raises the transfer speed limit 48 00:01:59,820 --> 00:02:00,820 considerably. 49 00:02:02,250 --> 00:02:04,200 Have you thought about the information? 50 00:02:06,510 --> 00:02:07,829 Have you thought about the information 51 00:02:07,830 --> 00:02:09,990 transfer that's really happening here 52 00:02:11,130 --> 00:02:13,499 because the actual chain of information 53 00:02:13,500 --> 00:02:15,299 looks more like this? 54 00:02:15,300 --> 00:02:17,999 It's the brain that sends the information 55 00:02:18,000 --> 00:02:20,159 to a body part, which interacts with 56 00:02:20,160 --> 00:02:21,869 the device that sends information to the 57 00:02:21,870 --> 00:02:24,419 computer and at both 58 00:02:24,420 --> 00:02:26,489 and both ends of the communication can 59 00:02:26,490 --> 00:02:28,769 deal with much higher data rates 60 00:02:28,770 --> 00:02:31,159 than usually five to 15 bytes per second. 61 00:02:31,160 --> 00:02:34,049 That's the measurement. 62 00:02:34,050 --> 00:02:36,149 So what would happen if you cut out 63 00:02:36,150 --> 00:02:37,919 the middleman and connect the device 64 00:02:37,920 --> 00:02:39,030 directly to the brain? 65 00:02:42,650 --> 00:02:45,019 So as for language spoilers, 66 00:02:45,020 --> 00:02:46,849 it would be surprisingly underwhelming 67 00:02:48,110 --> 00:02:49,939 thinking birds isn't that much faster 68 00:02:49,940 --> 00:02:52,039 than speaking them when you read out the 69 00:02:52,040 --> 00:02:53,479 words from your brain. 70 00:02:53,480 --> 00:02:55,399 The data quality is horrible. 71 00:02:55,400 --> 00:02:57,679 There would be a lot of misunderstanding 72 00:02:57,680 --> 00:02:58,680 and you would have 73 00:03:00,230 --> 00:03:02,389 and you would have to sorry, 74 00:03:04,070 --> 00:03:05,839 and you would have to teach the software 75 00:03:05,840 --> 00:03:07,759 where the words are in your brain. 76 00:03:07,760 --> 00:03:09,649 So you need a couple of hours before you 77 00:03:09,650 --> 00:03:11,059 could get even started. 78 00:03:12,110 --> 00:03:14,479 But of course, that's such an interface 79 00:03:14,480 --> 00:03:16,009 you could also use something else than 80 00:03:16,010 --> 00:03:17,010 words as input. 81 00:03:20,450 --> 00:03:22,430 For example, let's look at art. 82 00:03:23,480 --> 00:03:25,549 When I imagine a picture and we 83 00:03:25,550 --> 00:03:27,619 want to show it to other people, 84 00:03:27,620 --> 00:03:30,409 I can just pick up my canvas and 85 00:03:30,410 --> 00:03:33,499 take a paintbrush and start painting, 86 00:03:33,500 --> 00:03:35,599 or I could also use a graphics tablet to 87 00:03:35,600 --> 00:03:38,029 paint with a word or a brush. 88 00:03:38,030 --> 00:03:40,099 Or I could use a specialized mouse 89 00:03:40,100 --> 00:03:42,199 to put together 3D objects and 90 00:03:42,200 --> 00:03:44,059 render the scene. 91 00:03:44,060 --> 00:03:46,069 But each of these strategies is going to 92 00:03:46,070 --> 00:03:48,019 cost me several hours. 93 00:03:48,020 --> 00:03:49,609 And even if I actually had some knowledge 94 00:03:49,610 --> 00:03:51,889 in painting, the end result would never 95 00:03:51,890 --> 00:03:53,599 look exactly like my imagination. 96 00:03:55,100 --> 00:03:56,869 Here's why a brain computer interface 97 00:03:56,870 --> 00:03:58,939 interface is starting to shine because 98 00:03:58,940 --> 00:04:00,650 it can access the image in your head 99 00:04:01,970 --> 00:04:03,349 entirely and immediately. 100 00:04:06,010 --> 00:04:08,919 There's not just arts and 101 00:04:08,920 --> 00:04:10,989 language, there's also gaming, 102 00:04:10,990 --> 00:04:12,969 there's music, there's interfaces 103 00:04:12,970 --> 00:04:15,039 everywhere, interfaces 104 00:04:15,040 --> 00:04:16,869 that are hard to learn and even harder to 105 00:04:16,870 --> 00:04:18,969 master interfaces that 106 00:04:18,970 --> 00:04:21,040 put a limit on our credit creativity. 107 00:04:22,180 --> 00:04:24,099 Having a really good brain computer 108 00:04:24,100 --> 00:04:26,289 interface would make lots of 109 00:04:26,290 --> 00:04:28,449 these devices redundant. 110 00:04:28,450 --> 00:04:30,429 And as soon as a computer can guess what 111 00:04:30,430 --> 00:04:32,859 we want to do, a lot of frustration 112 00:04:32,860 --> 00:04:34,389 starts to disappear from learning. 113 00:04:35,920 --> 00:04:37,779 And since there's more to the brain than 114 00:04:37,780 --> 00:04:40,059 just hearing and seeing, maybe 115 00:04:40,060 --> 00:04:41,919 it makes sense to transfer data from 116 00:04:41,920 --> 00:04:43,089 other senses as well. 117 00:04:44,170 --> 00:04:46,329 Unfortunately, this is still very much 118 00:04:46,330 --> 00:04:48,459 a vision into the future, 119 00:04:48,460 --> 00:04:50,769 but we have already today 120 00:04:50,770 --> 00:04:52,779 some serious possibilities of controlling 121 00:04:52,780 --> 00:04:54,100 a computer with our minds. 122 00:04:55,270 --> 00:04:57,789 It's not even that expensive, but I'll 123 00:04:57,790 --> 00:04:59,619 come back to that in the last part of our 124 00:04:59,620 --> 00:05:00,620 talk. 125 00:05:04,010 --> 00:05:05,749 Before we can build a system that can 126 00:05:05,750 --> 00:05:07,819 read from our brain, we need to know 127 00:05:07,820 --> 00:05:09,740 how the data is represented in there. 128 00:05:13,640 --> 00:05:15,739 For that, we need to take a look at the 129 00:05:15,740 --> 00:05:17,329 very basic map of the brain 130 00:05:18,350 --> 00:05:20,419 we see and imagine with the back 131 00:05:20,420 --> 00:05:21,559 of the brain. 132 00:05:21,560 --> 00:05:23,240 It's called the occipital cortex. 133 00:05:24,470 --> 00:05:26,269 Hearing them talking happens in the lower 134 00:05:26,270 --> 00:05:28,669 side areas directly next to the ears, 135 00:05:28,670 --> 00:05:29,670 which makes sense. 136 00:05:31,100 --> 00:05:32,719 The top of the brain is responsible for 137 00:05:32,720 --> 00:05:35,959 skin and muscles and movement in general. 138 00:05:35,960 --> 00:05:38,119 And what the front part does is most 139 00:05:38,120 --> 00:05:39,799 difficult to understand and even for 140 00:05:39,800 --> 00:05:41,719 scientists. So let's just say it does the 141 00:05:41,720 --> 00:05:42,720 actual thinking. 142 00:05:47,780 --> 00:05:49,759 So let's start with a simple example, 143 00:05:49,760 --> 00:05:51,230 let's imagine a Red Bull. 144 00:05:52,790 --> 00:05:54,889 First, we look up the meaning of a 145 00:05:54,890 --> 00:05:57,229 red ball in the meaning library 146 00:05:57,230 --> 00:05:58,230 down here. 147 00:05:58,910 --> 00:06:01,099 Then we send this object to 148 00:06:01,100 --> 00:06:02,989 the appearance library in the front. 149 00:06:02,990 --> 00:06:05,389 This area is a library of visual objects. 150 00:06:06,720 --> 00:06:08,989 Resolve the object code into its color 151 00:06:08,990 --> 00:06:11,119 and shape codes, which 152 00:06:11,120 --> 00:06:12,740 are immediately sent downwards. 153 00:06:14,060 --> 00:06:16,129 Color and Shape Center then convert these 154 00:06:16,130 --> 00:06:18,529 property codes into the visual meaning 155 00:06:19,700 --> 00:06:21,949 and build up a link to the correct 156 00:06:21,950 --> 00:06:23,299 space in the visual buffer. 157 00:06:25,040 --> 00:06:26,239 As soon as this connection is 158 00:06:26,240 --> 00:06:28,399 established, we see the image of 159 00:06:28,400 --> 00:06:29,540 a red ball in our minds. 160 00:06:33,930 --> 00:06:35,969 This red ball is, of course, storage 161 00:06:35,970 --> 00:06:38,009 somewhere in the brain to along with 162 00:06:38,010 --> 00:06:39,479 millions of other datasets. 163 00:06:40,770 --> 00:06:42,239 All this information is stored and the 164 00:06:42,240 --> 00:06:44,369 connections between neurons. 165 00:06:44,370 --> 00:06:46,199 Neurons never work alone because they're 166 00:06:46,200 --> 00:06:48,360 too weak. I'll explain that in a minute 167 00:06:49,380 --> 00:06:50,759 to create a meaningful signal. 168 00:06:50,760 --> 00:06:52,919 They have to work together, and 169 00:06:52,920 --> 00:06:54,689 they do that in groups of hundreds of 170 00:06:54,690 --> 00:06:57,779 neurons, which is called a mini column. 171 00:06:57,780 --> 00:06:59,849 Each mini column stores one 172 00:06:59,850 --> 00:07:02,159 unit of information, together 173 00:07:02,160 --> 00:07:03,989 with 100 other million homes. 174 00:07:03,990 --> 00:07:06,269 They form one memory unit called a hyper 175 00:07:06,270 --> 00:07:08,909 column organized in 176 00:07:08,910 --> 00:07:11,309 like. This column forms. 177 00:07:11,310 --> 00:07:13,559 Each one of these hexagonal memory units 178 00:07:13,560 --> 00:07:15,629 is incredibly flexible. 179 00:07:15,630 --> 00:07:17,219 They can remember everything about Red 180 00:07:17,220 --> 00:07:19,679 Bulls, how it looks, how it feels, 181 00:07:19,680 --> 00:07:21,449 how it sounds large, and even your 182 00:07:21,450 --> 00:07:22,819 emotions toward Red Bulls. 183 00:07:25,680 --> 00:07:27,299 But because we only have two million 184 00:07:27,300 --> 00:07:29,459 hyper columns, it would be a ways to 185 00:07:29,460 --> 00:07:32,039 use a separate one for every ball. 186 00:07:32,040 --> 00:07:34,649 That's why your memory unit for Red Balls 187 00:07:34,650 --> 00:07:36,749 is the same human unit as for green, 188 00:07:36,750 --> 00:07:38,879 black, blue and green and 189 00:07:38,880 --> 00:07:39,880 yellow balls as well. 190 00:07:41,730 --> 00:07:43,919 So what happens when the hyper 191 00:07:43,920 --> 00:07:46,229 column of neurons recognizes the 192 00:07:46,230 --> 00:07:49,139 special ball from your childhood? 193 00:07:49,140 --> 00:07:50,579 It activates the mini column. 194 00:07:50,580 --> 00:07:52,169 It stores the emotion of your childhood 195 00:07:53,340 --> 00:07:55,289 before all the neurons in this minute 196 00:07:55,290 --> 00:07:57,359 column, but just idling 197 00:07:57,360 --> 00:08:00,179 their activity was low, random, 198 00:08:00,180 --> 00:08:01,949 slow and completely synchronized. 199 00:08:03,660 --> 00:08:05,189 But as soon as the activation signal 200 00:08:05,190 --> 00:08:07,409 arrives it, they generate waves 201 00:08:07,410 --> 00:08:09,089 of activity. 202 00:08:09,090 --> 00:08:10,679 Now, each one of these yellow dots on the 203 00:08:10,680 --> 00:08:12,869 bottom is the activity of a single 204 00:08:12,870 --> 00:08:15,119 neuron because single 205 00:08:15,120 --> 00:08:17,249 neurons can only generate a short 206 00:08:17,250 --> 00:08:18,779 flash of activity. 207 00:08:18,780 --> 00:08:20,130 They have to work together 208 00:08:21,780 --> 00:08:23,789 and produce these waves. 209 00:08:23,790 --> 00:08:26,039 These waves are the typical data format 210 00:08:26,040 --> 00:08:27,749 inside the brain. 211 00:08:27,750 --> 00:08:29,729 Waves can encode more information by 212 00:08:29,730 --> 00:08:31,919 changing frequency and phase. 213 00:08:31,920 --> 00:08:33,989 The amplitude, however, is mostly the 214 00:08:33,990 --> 00:08:35,339 same. 215 00:08:35,340 --> 00:08:37,499 And in this case, this wave is part of 216 00:08:37,500 --> 00:08:39,418 the memory of the red ball of your 217 00:08:39,419 --> 00:08:40,419 childhood. 218 00:08:42,360 --> 00:08:43,440 In case you were wondering, 219 00:08:44,850 --> 00:08:46,949 this is how hyper 220 00:08:46,950 --> 00:08:49,289 columns communicate with each other in 221 00:08:49,290 --> 00:08:50,290 slow motion. 222 00:08:52,380 --> 00:08:54,539 This is a slow motion part 223 00:08:54,540 --> 00:08:56,639 of the actual cortex cortical surface 224 00:08:56,640 --> 00:08:58,859 stimulated from a rat cortex. 225 00:09:00,240 --> 00:09:02,819 You'll see that there's some structure 226 00:09:02,820 --> 00:09:05,849 and some clear loops, but it's 227 00:09:05,850 --> 00:09:08,009 all flowing, shifting and 228 00:09:08,010 --> 00:09:10,169 never really repeating. 229 00:09:10,170 --> 00:09:11,759 There are some nodes, there are some 230 00:09:11,760 --> 00:09:14,339 loops, and all of these loops up 231 00:09:14,340 --> 00:09:17,009 forming in the later stages of the brain. 232 00:09:17,010 --> 00:09:19,889 More complex and more 233 00:09:19,890 --> 00:09:21,419 and larger loops. 234 00:09:21,420 --> 00:09:23,759 And these are what we perceive 235 00:09:23,760 --> 00:09:24,760 as information. 236 00:09:27,150 --> 00:09:28,949 This is how all these electrical waves 237 00:09:28,950 --> 00:09:30,929 are traveling through the brain. 238 00:09:41,910 --> 00:09:44,059 So before further, I'd give 239 00:09:44,060 --> 00:09:46,229 throughout my warning, I will use some 240 00:09:46,230 --> 00:09:47,369 formulas. 241 00:09:47,370 --> 00:09:48,419 Please don't be. 242 00:09:48,420 --> 00:09:49,420 Yeah. 243 00:09:50,880 --> 00:09:51,689 Don't be alarmed. 244 00:09:51,690 --> 00:09:53,759 Yeah, it's 245 00:09:53,760 --> 00:09:54,749 OK. 246 00:09:54,750 --> 00:09:55,750 They won't do anything. 247 00:09:58,000 --> 00:10:00,509 And what we now have to do is to find the 248 00:10:00,510 --> 00:10:02,489 origin of this electrical waves. 249 00:10:04,260 --> 00:10:06,509 So how do we do this 250 00:10:06,510 --> 00:10:08,429 without physical access to the brain? 251 00:10:09,600 --> 00:10:11,849 Well, the easiest and already established 252 00:10:11,850 --> 00:10:13,229 way is to use an EEG. 253 00:10:14,490 --> 00:10:16,319 This is a device that measures electric 254 00:10:16,320 --> 00:10:18,179 fields with the help of some electrodes 255 00:10:19,320 --> 00:10:21,479 there are used for research or 256 00:10:21,480 --> 00:10:23,369 medical purposes. 257 00:10:23,370 --> 00:10:24,959 That's some of this 258 00:10:26,280 --> 00:10:28,349 and also smaller ones for consumers 259 00:10:28,350 --> 00:10:30,299 like this one on the other side. 260 00:10:31,860 --> 00:10:33,569 They differ mostly in the price and the 261 00:10:33,570 --> 00:10:34,890 number of sensors. 262 00:10:36,300 --> 00:10:38,339 You have probably seen one in the media 263 00:10:38,340 --> 00:10:40,499 already publicly to 264 00:10:40,500 --> 00:10:42,659 explain how this work. 265 00:10:42,660 --> 00:10:45,029 For now, it's so efficient to know that 266 00:10:45,030 --> 00:10:46,529 you measure electric fields. 267 00:10:47,580 --> 00:10:49,679 The big question is now how does this 268 00:10:49,680 --> 00:10:50,680 all work? 269 00:10:52,500 --> 00:10:54,599 The electric field from each neural 270 00:10:54,600 --> 00:10:56,939 cluster is strong enough to be measured 271 00:10:56,940 --> 00:10:58,619 outside of the head. 272 00:10:58,620 --> 00:11:01,259 It is usually between 200 and 500 273 00:11:01,260 --> 00:11:02,260 millibars. 274 00:11:03,950 --> 00:11:06,139 The electrodes of an EEG are 275 00:11:06,140 --> 00:11:07,609 placed all over the head. 276 00:11:09,500 --> 00:11:11,509 We have to place the electrodes directly 277 00:11:11,510 --> 00:11:13,669 onto the head face because air 278 00:11:13,670 --> 00:11:15,889 connectivity is very low. 279 00:11:15,890 --> 00:11:18,289 Then we combine all the electrodes 280 00:11:18,290 --> 00:11:19,879 electrode voltage so that we know the 281 00:11:19,880 --> 00:11:21,949 voltage for each point on 282 00:11:21,950 --> 00:11:22,950 a head. Surface 283 00:11:24,950 --> 00:11:27,319 neuron clusters also produce a magnetic 284 00:11:27,320 --> 00:11:29,089 field at the same time. 285 00:11:29,090 --> 00:11:30,979 This could also be measured from a small 286 00:11:30,980 --> 00:11:32,209 distance. 287 00:11:32,210 --> 00:11:34,549 But the device for measuring 288 00:11:34,550 --> 00:11:36,799 very weak magnetic fields are far 289 00:11:36,800 --> 00:11:39,559 too big to carry around and also very 290 00:11:39,560 --> 00:11:40,560 expensive. 291 00:11:41,330 --> 00:11:43,399 They also have to be isolated 292 00:11:43,400 --> 00:11:46,129 from other strong fields. 293 00:11:46,130 --> 00:11:47,869 Even regular watches 294 00:11:49,110 --> 00:11:51,649 create and magnetic fields that magnitude 295 00:11:51,650 --> 00:11:53,690 stronger than the whole brain. 296 00:11:55,160 --> 00:11:57,529 The research for portable devices 297 00:11:57,530 --> 00:12:00,079 are already starting, but currently 298 00:12:00,080 --> 00:12:01,789 are the only handy devices for custom 299 00:12:01,790 --> 00:12:02,790 usage. 300 00:12:04,110 --> 00:12:05,859 Since the electric fields is cost 301 00:12:05,860 --> 00:12:08,009 somewhere inside of the brain, 302 00:12:08,010 --> 00:12:09,479 we have to look at what happens on the 303 00:12:09,480 --> 00:12:11,699 way outside our head 304 00:12:11,700 --> 00:12:14,039 consists of different tissues, 305 00:12:14,040 --> 00:12:16,199 which all put up some amount of 306 00:12:16,200 --> 00:12:18,899 resistance to electric fields. 307 00:12:18,900 --> 00:12:20,999 So instead of traveling straight 308 00:12:21,000 --> 00:12:23,009 through to head, the electric field gets 309 00:12:23,010 --> 00:12:24,359 distorted. 310 00:12:24,360 --> 00:12:27,389 It could be imagined like this 311 00:12:27,390 --> 00:12:28,679 when jogging. 312 00:12:28,680 --> 00:12:30,419 You have different speeds on different 313 00:12:30,420 --> 00:12:32,549 kind of surfaces 314 00:12:32,550 --> 00:12:34,439 when you are dragging on asphalt. 315 00:12:34,440 --> 00:12:36,599 You are quicker than on sand. 316 00:12:36,600 --> 00:12:39,179 It's the same with the electric fields 317 00:12:39,180 --> 00:12:41,579 in our head. For example, our bones 318 00:12:41,580 --> 00:12:43,679 have a very high resistance, and the 319 00:12:43,680 --> 00:12:45,479 liquid around our brain is very 320 00:12:45,480 --> 00:12:46,859 conductive. 321 00:12:46,860 --> 00:12:48,899 And one brain tissue even has different 322 00:12:48,900 --> 00:12:50,999 conductivity for different directions. 323 00:12:51,000 --> 00:12:52,799 So it's got a little bit complicated 324 00:12:52,800 --> 00:12:53,800 here. 325 00:12:56,100 --> 00:12:57,959 We are finished with the electoral 326 00:12:57,960 --> 00:13:00,479 physical set up, active neurons create 327 00:13:00,480 --> 00:13:02,609 an electric field inside of the head. 328 00:13:02,610 --> 00:13:05,429 We know about the conductivity of 329 00:13:05,430 --> 00:13:07,499 different tissue layers and the 330 00:13:07,500 --> 00:13:09,929 voltage on the heads of face without 331 00:13:09,930 --> 00:13:11,580 any other knowledge. You're stuck here, 332 00:13:13,080 --> 00:13:14,939 like in many occasions, for solving 333 00:13:14,940 --> 00:13:16,019 problems. 334 00:13:16,020 --> 00:13:18,269 Now we are required a mighty toolbox 335 00:13:18,270 --> 00:13:19,829 of mathematical science to find a 336 00:13:19,830 --> 00:13:21,329 solution. 337 00:13:21,330 --> 00:13:23,459 What we want to do now, we want 338 00:13:23,460 --> 00:13:24,869 to translate our setting 339 00:13:25,890 --> 00:13:28,019 and our problem too 340 00:13:28,020 --> 00:13:30,029 into the language of mathematics. 341 00:13:30,030 --> 00:13:31,769 Solve the problem. Then translate the 342 00:13:31,770 --> 00:13:34,439 mathematical solution back into reality 343 00:13:34,440 --> 00:13:36,569 and to warn you beforehand within the 344 00:13:36,570 --> 00:13:38,699 mathematical world now, and 345 00:13:38,700 --> 00:13:40,589 as I said, we will need another 346 00:13:40,590 --> 00:13:42,659 translation into something a computer 347 00:13:42,660 --> 00:13:43,660 can solve. 348 00:13:45,740 --> 00:13:46,740 The first step 349 00:13:48,650 --> 00:13:50,239 in translating into a mathematical 350 00:13:50,240 --> 00:13:51,919 language is also calls to model the 351 00:13:51,920 --> 00:13:52,920 problem. 352 00:13:53,690 --> 00:13:55,819 Building a mathematical model uses 353 00:13:55,820 --> 00:13:58,189 principles, laws and concepts 354 00:13:58,190 --> 00:14:00,319 from other fields of science. 355 00:14:00,320 --> 00:14:02,599 In our case, we use a modified version 356 00:14:02,600 --> 00:14:04,849 of the famous Maxwell equations, which 357 00:14:04,850 --> 00:14:06,979 describe the connection between magnetic 358 00:14:06,980 --> 00:14:09,379 and electric fields and forces. 359 00:14:09,380 --> 00:14:11,119 They were first described by a physical 360 00:14:11,120 --> 00:14:13,489 scientist named Jesse Maxwell 361 00:14:13,490 --> 00:14:14,809 in 1865. 362 00:14:16,190 --> 00:14:18,289 Those equations can be simplified and 363 00:14:18,290 --> 00:14:19,970 adapted to our application. 364 00:14:22,510 --> 00:14:24,399 The information we get as a result is 365 00:14:24,400 --> 00:14:25,400 written here. 366 00:14:26,910 --> 00:14:28,440 First, we look at the left side. 367 00:14:29,460 --> 00:14:32,159 The ex describes any point in the head 368 00:14:32,160 --> 00:14:34,229 you have X is the electric 369 00:14:34,230 --> 00:14:36,629 future at the position X and 370 00:14:36,630 --> 00:14:38,829 when you derive that you get 371 00:14:38,830 --> 00:14:40,169 the current at this point. 372 00:14:41,490 --> 00:14:43,379 Now your fault is with the electric 373 00:14:43,380 --> 00:14:45,509 connectivity and compute the 374 00:14:45,510 --> 00:14:46,510 divergence. 375 00:14:47,670 --> 00:14:49,829 That's what's happened mathematically. 376 00:14:49,830 --> 00:14:52,889 Well, what you can imagine happening 377 00:14:52,890 --> 00:14:54,959 in the left side is the difference 378 00:14:54,960 --> 00:14:57,299 between the incoming flow 379 00:14:57,300 --> 00:14:59,669 and the outgoing flow of the electric 380 00:14:59,670 --> 00:15:01,200 field at every point x. 381 00:15:03,840 --> 00:15:05,579 The right side is all about communal 382 00:15:05,580 --> 00:15:08,439 activity, was this Britain, there 383 00:15:08,440 --> 00:15:10,589 is the so-called so-called current 384 00:15:10,590 --> 00:15:11,590 dipole. 385 00:15:12,300 --> 00:15:14,459 This means that the 386 00:15:14,460 --> 00:15:16,799 whole equation is zero everywhere, except 387 00:15:16,800 --> 00:15:18,840 at the position of the activity. 388 00:15:20,190 --> 00:15:22,229 Together, this means for every point in 389 00:15:22,230 --> 00:15:24,659 the head, the same amount of electricity 390 00:15:24,660 --> 00:15:27,089 flows in it flows out, with the exception 391 00:15:27,090 --> 00:15:28,140 of the active neurons. 392 00:15:31,030 --> 00:15:32,199 So that was the equation. 393 00:15:32,200 --> 00:15:34,150 Let's talk about solving this equation, 394 00:15:35,560 --> 00:15:37,899 the usual way to solve this 395 00:15:37,900 --> 00:15:40,119 is to assume that it had connectivity 396 00:15:40,120 --> 00:15:42,309 and neuronal activity. 397 00:15:42,310 --> 00:15:45,099 And now 398 00:15:45,100 --> 00:15:47,619 the head connectivity and the location 399 00:15:47,620 --> 00:15:50,019 of the activity are already known. 400 00:15:50,020 --> 00:15:51,969 And you would start calculating you had 401 00:15:51,970 --> 00:15:52,970 currents. 402 00:15:53,590 --> 00:15:56,109 This is called the forward problem. 403 00:15:56,110 --> 00:15:58,209 But of course, we don't know the 404 00:15:58,210 --> 00:15:59,949 current location that because that's 405 00:15:59,950 --> 00:16:01,090 something we want to compute. 406 00:16:03,670 --> 00:16:04,670 So. 407 00:16:06,920 --> 00:16:08,479 We want to calculate the activity 408 00:16:08,480 --> 00:16:10,549 location from the conductivity and the 409 00:16:10,550 --> 00:16:12,769 measured surface wellhead for 410 00:16:12,770 --> 00:16:14,959 just the equation needs to be 411 00:16:14,960 --> 00:16:16,249 of turned around. 412 00:16:16,250 --> 00:16:17,809 This is called the inverse problem and 413 00:16:17,810 --> 00:16:19,309 it's surprisingly difficult. 414 00:16:22,680 --> 00:16:24,749 We encounter the next problem when 415 00:16:24,750 --> 00:16:27,179 we look at these things we measure, 416 00:16:27,180 --> 00:16:29,579 we want to know where the activity inside 417 00:16:29,580 --> 00:16:30,690 the brain is located, 418 00:16:31,920 --> 00:16:34,229 so we measure the electric field 419 00:16:34,230 --> 00:16:35,609 on the head surface. 420 00:16:35,610 --> 00:16:37,769 But we cannot measure inside to 421 00:16:37,770 --> 00:16:39,840 have that something different over there. 422 00:16:42,150 --> 00:16:44,309 Medical, you have probably around 423 00:16:44,310 --> 00:16:46,649 220 sensors, 424 00:16:46,650 --> 00:16:48,809 so you measure electric field 425 00:16:48,810 --> 00:16:50,819 only has a resolution of one hundred and 426 00:16:50,820 --> 00:16:52,529 twenty points. 427 00:16:52,530 --> 00:16:54,899 This gives us much less information 428 00:16:54,900 --> 00:16:55,950 than we actually need. 429 00:16:57,330 --> 00:16:59,519 We can deal with such kind of less 430 00:16:59,520 --> 00:17:02,099 information if you make some presumptions 431 00:17:02,100 --> 00:17:04,529 about the location of activity. 432 00:17:04,530 --> 00:17:07,229 Usually, we limit ourselves 433 00:17:07,230 --> 00:17:09,358 to places in the brain with actual 434 00:17:09,359 --> 00:17:11,879 neurons, namely the cortex. 435 00:17:11,880 --> 00:17:13,739 But of course, the resolution of our 436 00:17:13,740 --> 00:17:16,169 estimation gets better, the more sensors 437 00:17:16,170 --> 00:17:17,170 we have. 438 00:17:20,089 --> 00:17:22,639 Once we have faced these problems, 439 00:17:22,640 --> 00:17:24,529 we now can convert the equation into a 440 00:17:24,530 --> 00:17:26,689 computer friendly format, 441 00:17:26,690 --> 00:17:29,659 a computer cannot deal with equation 442 00:17:29,660 --> 00:17:31,759 that describes something for everyone 443 00:17:31,760 --> 00:17:33,049 in the head. 444 00:17:33,050 --> 00:17:35,149 That would be infinitely many points, 445 00:17:35,150 --> 00:17:37,309 and infinite memory is still quite 446 00:17:37,310 --> 00:17:38,310 expensive. 447 00:17:39,020 --> 00:17:41,119 So fortunately, some 448 00:17:41,120 --> 00:17:43,309 clever mathematicians develop 449 00:17:43,310 --> 00:17:45,439 algorithms for solving the problem, 450 00:17:45,440 --> 00:17:46,670 which are computer friendly. 451 00:17:50,520 --> 00:17:51,520 We do not compute 452 00:17:53,220 --> 00:17:55,049 for every point in the head. 453 00:17:55,050 --> 00:17:57,419 Instead, we select only 454 00:17:57,420 --> 00:17:59,549 some points of the head, for example, 455 00:17:59,550 --> 00:18:01,889 one million and compute results 456 00:18:01,890 --> 00:18:04,559 only for them to interpolate 457 00:18:04,560 --> 00:18:06,659 the areas between our selected points 458 00:18:06,660 --> 00:18:09,479 when we need to evaluate their. 459 00:18:09,480 --> 00:18:11,519 Mathematically speaking, this selection 460 00:18:11,520 --> 00:18:13,349 process is called just fertilization. 461 00:18:14,370 --> 00:18:16,769 The continuous structures of the head 462 00:18:16,770 --> 00:18:19,319 are divided into small cells and simple 463 00:18:19,320 --> 00:18:21,449 of simple geometry 464 00:18:21,450 --> 00:18:23,699 the continuous structures 465 00:18:23,700 --> 00:18:24,700 of the head. 466 00:18:26,190 --> 00:18:28,769 Usually we use tetrahydrate or IRA 467 00:18:28,770 --> 00:18:31,109 or a mixture of both. 468 00:18:31,110 --> 00:18:32,849 This works well, and it can be 469 00:18:32,850 --> 00:18:35,309 mathematically proven that the error 470 00:18:35,310 --> 00:18:37,379 from the disk fertilization is small 471 00:18:37,380 --> 00:18:38,520 enough for certain settings. 472 00:18:41,640 --> 00:18:44,279 For this approach to work well, 473 00:18:44,280 --> 00:18:46,230 we need a good 3D model of the hat, 474 00:18:48,210 --> 00:18:49,469 and we haven't talked about the 475 00:18:49,470 --> 00:18:50,470 connectivity. 476 00:18:51,510 --> 00:18:53,189 That's the last part that's missing for 477 00:18:53,190 --> 00:18:55,319 our model when building the 478 00:18:55,320 --> 00:18:56,219 3D model. 479 00:18:56,220 --> 00:18:57,869 We need this model to represent the 480 00:18:57,870 --> 00:18:59,130 connectivity well enough. 481 00:19:02,510 --> 00:19:04,999 That's the next step in the scientific 482 00:19:05,000 --> 00:19:07,399 way for getting connectivity 483 00:19:07,400 --> 00:19:09,109 values is to put the health of choice 484 00:19:09,110 --> 00:19:10,110 into an MRI. 485 00:19:11,180 --> 00:19:13,339 That is a big machine 486 00:19:13,340 --> 00:19:15,739 which can take High-Resolution images 487 00:19:15,740 --> 00:19:17,089 of the inside of the head. 488 00:19:18,440 --> 00:19:20,809 Then we take such 489 00:19:20,810 --> 00:19:23,269 a good picture from this head and cover 490 00:19:23,270 --> 00:19:25,279 this picture into this little geometric 491 00:19:25,280 --> 00:19:26,280 cells. 492 00:19:26,990 --> 00:19:29,059 There's only one connectivity value for 493 00:19:29,060 --> 00:19:30,060 one cell. 494 00:19:31,280 --> 00:19:33,409 If you want to be very exact, 495 00:19:33,410 --> 00:19:36,289 you have to use a different head model 496 00:19:36,290 --> 00:19:38,479 for every person because every head 497 00:19:38,480 --> 00:19:40,069 is unique. 498 00:19:40,070 --> 00:19:42,469 And because the brain can move around 499 00:19:42,470 --> 00:19:45,289 inside the head, you should always, 500 00:19:45,290 --> 00:19:47,299 always measure your person in the same 501 00:19:47,300 --> 00:19:48,559 position. 502 00:19:48,560 --> 00:19:50,959 But when you don't need to be precise, 503 00:19:50,960 --> 00:19:51,960 just a simple rule. 504 00:19:52,910 --> 00:19:54,919 Any head model is better than no have 505 00:19:54,920 --> 00:19:55,920 model. 506 00:19:59,550 --> 00:20:01,439 Then we are ready to estimate the 507 00:20:01,440 --> 00:20:03,659 location of neuronal activity. 508 00:20:03,660 --> 00:20:05,849 We take the connectivity model 509 00:20:05,850 --> 00:20:08,009 of our head, the location of 510 00:20:08,010 --> 00:20:09,010 the sensors, 511 00:20:10,710 --> 00:20:13,799 all the locations of all cortex points 512 00:20:13,800 --> 00:20:15,780 and put all of this into this machine. 513 00:20:17,280 --> 00:20:19,079 This machine first puts all of this 514 00:20:19,080 --> 00:20:22,019 information together into one head model. 515 00:20:22,020 --> 00:20:24,539 It made sure that all these little cubes 516 00:20:24,540 --> 00:20:27,029 and a sense of position and the cortex 517 00:20:27,030 --> 00:20:29,069 share the same coordinate system. 518 00:20:30,510 --> 00:20:32,759 The last dataset makes sure 519 00:20:32,760 --> 00:20:34,559 that we only estimate locations of 520 00:20:34,560 --> 00:20:36,150 activity inside the cortex 521 00:20:37,500 --> 00:20:39,509 when putting up these restrictions. 522 00:20:39,510 --> 00:20:42,149 You have to be sure of two things. 523 00:20:42,150 --> 00:20:44,219 First, the relocation of activity is 524 00:20:44,220 --> 00:20:45,539 actually in the cortex. 525 00:20:45,540 --> 00:20:48,119 And second, that there is no activity 526 00:20:48,120 --> 00:20:49,349 outside of the cortex. 527 00:20:51,300 --> 00:20:53,219 With the head model, the machine then 528 00:20:53,220 --> 00:20:55,559 compute something in an intermediate 529 00:20:55,560 --> 00:20:58,049 step called transfer matrix or lead 530 00:20:58,050 --> 00:21:00,359 matrix, the transfer 531 00:21:00,360 --> 00:21:02,369 matrix describes what the census would 532 00:21:02,370 --> 00:21:04,019 measure for every possible electric 533 00:21:04,020 --> 00:21:05,939 activity in the cortex. 534 00:21:05,940 --> 00:21:08,159 Building this matrix is a very time and 535 00:21:08,160 --> 00:21:09,569 memory consuming process. 536 00:21:12,340 --> 00:21:14,409 But the nice thing about this whole first 537 00:21:14,410 --> 00:21:16,659 part is that you can hear 538 00:21:16,660 --> 00:21:17,660 it offline, 539 00:21:18,850 --> 00:21:21,309 so this means you only have to do it once 540 00:21:21,310 --> 00:21:23,729 after the first huge calculation, only 541 00:21:23,730 --> 00:21:25,599 the last step requires some brain 542 00:21:25,600 --> 00:21:27,220 activity over there. 543 00:21:29,500 --> 00:21:31,209 Since this last step can happen in a few 544 00:21:31,210 --> 00:21:32,139 milliseconds. 545 00:21:32,140 --> 00:21:33,639 This is perfect for real time 546 00:21:33,640 --> 00:21:34,779 application. 547 00:21:34,780 --> 00:21:36,969 And if you always use your 548 00:21:36,970 --> 00:21:39,339 gene in the same position on your head 549 00:21:39,340 --> 00:21:41,319 and your cortex has enough points, there 550 00:21:41,320 --> 00:21:43,029 is no reason to change the transfer 551 00:21:43,030 --> 00:21:44,030 matrix. 552 00:21:46,190 --> 00:21:48,259 The next step takes this transfer 553 00:21:48,260 --> 00:21:50,539 metrics and the measured electric 554 00:21:50,540 --> 00:21:51,540 field 555 00:21:52,910 --> 00:21:55,339 a good solver for the invest problem, 556 00:21:55,340 --> 00:21:57,499 then decides which was the most 557 00:21:57,500 --> 00:21:59,479 likely activity location that cost is 558 00:21:59,480 --> 00:22:01,549 tens of voltage. 559 00:22:01,550 --> 00:22:03,859 There are different ones like Manaus 560 00:22:03,860 --> 00:22:06,469 Loretta, which give reasonable 561 00:22:06,470 --> 00:22:09,559 results, but still 562 00:22:09,560 --> 00:22:11,749 this algorithm a little to time in memory 563 00:22:11,750 --> 00:22:13,969 consuming to do this in a good 564 00:22:13,970 --> 00:22:14,970 online time. 565 00:22:16,160 --> 00:22:18,499 So in the end, you get an activity map 566 00:22:18,500 --> 00:22:20,629 that shows where the active clusters 567 00:22:20,630 --> 00:22:21,630 are. 568 00:22:23,220 --> 00:22:24,709 And to conclude my park, 569 00:22:25,910 --> 00:22:28,009 yes, we can estimate we are 570 00:22:28,010 --> 00:22:29,060 neurons are active. 571 00:22:30,130 --> 00:22:32,329 Our estimation would be much better 572 00:22:32,330 --> 00:22:34,459 if we had millions of sensors instead 573 00:22:34,460 --> 00:22:35,540 of 120. 574 00:22:36,830 --> 00:22:38,599 It would be much better if the sensors 575 00:22:38,600 --> 00:22:40,189 were inside of the brain instead of 576 00:22:40,190 --> 00:22:42,499 outside and 577 00:22:42,500 --> 00:22:44,779 still need a ton of computational power 578 00:22:44,780 --> 00:22:46,159 and memory. 579 00:22:46,160 --> 00:22:48,049 But it is possible, like Dominic will 580 00:22:48,050 --> 00:22:49,160 explain in a few seconds. 581 00:22:53,540 --> 00:22:55,099 All right. We are finished with all the 582 00:22:55,100 --> 00:22:57,259 theory and be coming back to 583 00:22:57,260 --> 00:22:58,260 the real world now. 584 00:22:59,120 --> 00:23:01,249 Don't worry, it's not as hard as 585 00:23:01,250 --> 00:23:02,250 it sounds like 586 00:23:04,820 --> 00:23:07,489 in the beginning. You can actually 587 00:23:07,490 --> 00:23:09,769 deal with it actually and deal 588 00:23:09,770 --> 00:23:11,809 with not having the whole localization 589 00:23:11,810 --> 00:23:14,359 part and deal with sensor 590 00:23:14,360 --> 00:23:15,439 data only. 591 00:23:15,440 --> 00:23:17,569 But you will see that you run 592 00:23:17,570 --> 00:23:19,159 into a wall quite soon because your 593 00:23:19,160 --> 00:23:20,659 resolution isn't high enough. 594 00:23:20,660 --> 00:23:22,789 So the whole once 595 00:23:22,790 --> 00:23:24,769 you get there, you have to consider the 596 00:23:24,770 --> 00:23:27,079 part that any talk about and 597 00:23:27,080 --> 00:23:28,400 actually create an activity map. 598 00:23:29,480 --> 00:23:31,309 So what do you need to build your own 599 00:23:31,310 --> 00:23:32,689 BCI? 600 00:23:32,690 --> 00:23:34,699 Two pieces of hardware and two pieces of 601 00:23:34,700 --> 00:23:35,649 software. 602 00:23:35,650 --> 00:23:37,279 I'll start with the hardware. 603 00:23:39,180 --> 00:23:41,909 And E.J. consists of two core parts, 604 00:23:41,910 --> 00:23:44,249 an amplifier and an analog digital 605 00:23:44,250 --> 00:23:45,659 converter. 606 00:23:45,660 --> 00:23:47,399 You need both of these parts for every 607 00:23:47,400 --> 00:23:49,559 channel, since these chips are 608 00:23:49,560 --> 00:23:51,029 relatively expensive. 609 00:23:51,030 --> 00:23:53,309 More channels automatically also means 610 00:23:53,310 --> 00:23:54,310 more expensive. 611 00:23:56,040 --> 00:23:57,899 We want a noise floor of at least 40 612 00:23:57,900 --> 00:24:00,089 decibels for our amplifier to have 613 00:24:00,090 --> 00:24:02,159 a reasonable signal to noise 614 00:24:02,160 --> 00:24:03,059 ratio. 615 00:24:03,060 --> 00:24:05,849 The same is true for the 8C. 616 00:24:05,850 --> 00:24:07,619 The absolute minimum sampling frequency 617 00:24:07,620 --> 00:24:09,869 is around 100 carats, and you should 618 00:24:09,870 --> 00:24:11,939 go for 200 hertz for 619 00:24:11,940 --> 00:24:14,189 a reasonable minimum state 620 00:24:14,190 --> 00:24:16,649 of the art research uses. 621 00:24:16,650 --> 00:24:18,419 Sample rates between thousand and two and 622 00:24:18,420 --> 00:24:20,099 a half thousand hertz. 623 00:24:20,100 --> 00:24:21,689 Not because the brain actually generates 624 00:24:21,690 --> 00:24:23,939 these frequencies, but to get a better 625 00:24:23,940 --> 00:24:26,160 position for the frequency faces. 626 00:24:27,720 --> 00:24:29,279 Finally, since we are dealing with 627 00:24:29,280 --> 00:24:31,259 microwaves at the first stage, it's very 628 00:24:31,260 --> 00:24:33,149 important to care for proper shielding 629 00:24:33,150 --> 00:24:34,229 shielding in the first steps. 630 00:24:37,660 --> 00:24:40,029 So this is the current spectrum of EEG 631 00:24:40,030 --> 00:24:42,489 devices in a sensible price range, 632 00:24:42,490 --> 00:24:44,080 excluding all the medical devices here 633 00:24:45,220 --> 00:24:47,349 in our community, we have two big open 634 00:24:47,350 --> 00:24:49,419 EEG project schematics and 635 00:24:49,420 --> 00:24:50,889 different implementations readily 636 00:24:50,890 --> 00:24:52,839 available, and it's already more than 10 637 00:24:52,840 --> 00:24:53,840 years old. 638 00:24:55,180 --> 00:24:56,410 Then we have the Open BCI 639 00:24:58,390 --> 00:25:00,579 project, which is much newer and 640 00:25:00,580 --> 00:25:01,569 sponsored by dopamine. 641 00:25:01,570 --> 00:25:03,849 So you want to choose a camp 642 00:25:05,440 --> 00:25:07,539 and we start with 40 643 00:25:07,540 --> 00:25:10,419 euros for the cheapest and worst 644 00:25:10,420 --> 00:25:12,159 version of an open EEG. 645 00:25:12,160 --> 00:25:14,619 It's so cheap because it uses the AC 646 00:25:14,620 --> 00:25:17,079 in your sound card and your computer 647 00:25:17,080 --> 00:25:19,439 only has two channels, and these channels 648 00:25:19,440 --> 00:25:21,549 will be quite bad, but it 649 00:25:21,550 --> 00:25:22,550 will work. 650 00:25:23,110 --> 00:25:25,359 The better alternative is this only 651 00:25:25,360 --> 00:25:27,519 makes open energy intensive 652 00:25:27,520 --> 00:25:29,749 has a reference electrodes, 653 00:25:29,750 --> 00:25:32,139 which means you won't get channel drift 654 00:25:32,140 --> 00:25:34,029 or saturation problems with your 655 00:25:34,030 --> 00:25:36,369 amplifiers, and its ADC 656 00:25:36,370 --> 00:25:38,679 is much better than the one 657 00:25:38,680 --> 00:25:39,969 you use for your microphone in the 658 00:25:39,970 --> 00:25:40,869 computer. 659 00:25:40,870 --> 00:25:42,789 So this is the one the cheapest device 660 00:25:42,790 --> 00:25:44,380 actually recommend for beginners. 661 00:25:46,060 --> 00:25:47,619 The best solution depends on your budget. 662 00:25:47,620 --> 00:25:49,719 Of course, what you pay for is the number 663 00:25:49,720 --> 00:25:51,849 of channels and the low amount of noise. 664 00:25:52,900 --> 00:25:54,999 The best device you can get right now is 665 00:25:55,000 --> 00:25:57,429 the open BCI, with 16 channels 666 00:25:57,430 --> 00:25:59,829 in the double configuration, but 667 00:25:59,830 --> 00:26:01,719 and forty euros will only get you the 668 00:26:01,720 --> 00:26:02,859 passive electrodes. 669 00:26:03,910 --> 00:26:05,829 Passive electrodes are cheaper, but they 670 00:26:05,830 --> 00:26:08,079 send micro walls through very long 671 00:26:08,080 --> 00:26:10,989 wires, so they are very noisy 672 00:26:10,990 --> 00:26:12,909 and you need conductive gel to make 673 00:26:12,910 --> 00:26:15,009 contact with the head surface. 674 00:26:15,010 --> 00:26:16,929 Active electrodes have the amplifier and 675 00:26:16,930 --> 00:26:19,059 the ADC directly next to the sensors. 676 00:26:20,980 --> 00:26:22,839 That's why active electrodes are more 677 00:26:22,840 --> 00:26:24,759 expensive. You need separate electronics 678 00:26:24,760 --> 00:26:25,760 for each channel, 679 00:26:27,250 --> 00:26:29,439 but the noise levels are much lower and 680 00:26:29,440 --> 00:26:31,419 not having to put gel in your hair all 681 00:26:31,420 --> 00:26:32,420 the time, it's nice to 682 00:26:34,210 --> 00:26:36,759 have. The emotive, emotive 683 00:26:36,760 --> 00:26:38,859 epoch is the most compact 684 00:26:38,860 --> 00:26:40,569 package and trim. It's what transmits 685 00:26:40,570 --> 00:26:43,449 wirelessly and encrypted by default, 686 00:26:43,450 --> 00:26:45,909 but it's completely proprietary. 687 00:26:45,910 --> 00:26:48,159 I've heard that the transmission 688 00:26:48,160 --> 00:26:49,179 has been hacked. 689 00:26:49,180 --> 00:26:50,180 Actually, 690 00:26:51,370 --> 00:26:52,269 it's big weakness. 691 00:26:52,270 --> 00:26:53,409 It's the sampling frequency. 692 00:26:53,410 --> 00:26:54,969 It only can do one hundred and twenty 693 00:26:54,970 --> 00:26:56,230 five hertz of the maximum 694 00:26:59,390 --> 00:27:01,469 and then be coming to the software and 695 00:27:01,470 --> 00:27:03,579 the software needs to do for 696 00:27:03,580 --> 00:27:05,739 basic things become on the left 697 00:27:05,740 --> 00:27:06,740 side. 698 00:27:07,510 --> 00:27:09,279 First, it needs to acquire signals from 699 00:27:09,280 --> 00:27:11,349 the e.g and put them into data 700 00:27:11,350 --> 00:27:12,350 packets. 701 00:27:13,450 --> 00:27:15,879 Second, it needs to estimate the location 702 00:27:15,880 --> 00:27:16,960 of brain activity, 703 00:27:18,820 --> 00:27:20,769 so the estimated activity needs to be 704 00:27:20,770 --> 00:27:22,539 processed at a signal level. 705 00:27:22,540 --> 00:27:24,699 I'm talking about frequency filters, 706 00:27:24,700 --> 00:27:26,769 Fourier transform formations and so on. 707 00:27:27,970 --> 00:27:30,279 For some BCI paradigms, 708 00:27:30,280 --> 00:27:32,319 these are the small programs that 709 00:27:32,320 --> 00:27:34,210 actually know what brain activity is. 710 00:27:36,190 --> 00:27:38,589 Let's start on the left field trip. 711 00:27:38,590 --> 00:27:40,509 It's based on MATLAB developed in the 712 00:27:40,510 --> 00:27:43,779 Netherlands and most feature complete. 713 00:27:43,780 --> 00:27:44,829 It's a very versatile. 714 00:27:44,830 --> 00:27:47,289 It's used in science a lot, 715 00:27:47,290 --> 00:27:49,209 but it's real time. Support is quite 716 00:27:49,210 --> 00:27:51,039 weak, so you have to encode a lot of 717 00:27:51,040 --> 00:27:52,040 things. 718 00:27:52,540 --> 00:27:54,819 Neuro Nurofen is a plug in for a field 719 00:27:54,820 --> 00:27:57,279 trip. It's the only software right now 720 00:27:57,280 --> 00:27:58,959 that can deal with the highest quality of 721 00:27:58,960 --> 00:28:00,969 had models, so it's mostly used in 722 00:28:00,970 --> 00:28:01,869 cutting edge science. 723 00:28:01,870 --> 00:28:03,909 But if you're interested, you can do this 724 00:28:03,910 --> 00:28:04,910 only with free trip. 725 00:28:06,580 --> 00:28:08,169 The most user friendly solution is 726 00:28:08,170 --> 00:28:09,849 probably open vibe. 727 00:28:09,850 --> 00:28:11,649 It doesn't do localization, but it has a 728 00:28:11,650 --> 00:28:13,809 Google. I supports many types of 729 00:28:13,810 --> 00:28:16,299 e.g., and it's very well documented. 730 00:28:16,300 --> 00:28:18,279 There's even a plethora of examples out 731 00:28:18,280 --> 00:28:20,919 there and then this Mushu. 732 00:28:20,920 --> 00:28:23,019 This is on internships running on Python, 733 00:28:23,020 --> 00:28:25,209 mainly for people who don't like who like 734 00:28:25,210 --> 00:28:26,440 Python or dislike MATLAB. 735 00:28:30,390 --> 00:28:32,219 So when you have decided which software 736 00:28:32,220 --> 00:28:34,229 to use, it's time to choose a BCI 737 00:28:34,230 --> 00:28:35,230 paradigm. 738 00:28:35,980 --> 00:28:38,039 I will give you a short overview of 739 00:28:38,040 --> 00:28:40,559 the most popular paradigms today. 740 00:28:40,560 --> 00:28:42,179 Maybe this is a good start for you guys 741 00:28:42,180 --> 00:28:43,770 to get into something more advanced. 742 00:28:46,600 --> 00:28:48,809 Steady state evoked responses are 743 00:28:48,810 --> 00:28:50,979 a brute force approach to the brain, 744 00:28:50,980 --> 00:28:53,859 you put in a certain frequency directly 745 00:28:53,860 --> 00:28:55,989 through the eyes and hope that 746 00:28:55,990 --> 00:28:58,149 at least some neurons pick it up and 747 00:28:58,150 --> 00:29:00,849 convert it to electrical activity. 748 00:29:00,850 --> 00:29:02,289 The classic pattern is a flickering 749 00:29:02,290 --> 00:29:04,389 chessboard before 750 00:29:04,390 --> 00:29:05,319 show it to you. 751 00:29:05,320 --> 00:29:07,449 Here's a little warning everyone has a 752 00:29:07,450 --> 00:29:08,529 history of epilepsy. 753 00:29:08,530 --> 00:29:10,090 Please look away for five seconds. 754 00:29:11,110 --> 00:29:12,309 Really? 755 00:29:12,310 --> 00:29:13,749 OK, go. 756 00:29:16,980 --> 00:29:19,019 Works quite easily, right? 757 00:29:19,020 --> 00:29:20,369 So 12 hours and 10 hertz, 758 00:29:22,200 --> 00:29:23,879 both checkerboards were flickering at a 759 00:29:23,880 --> 00:29:26,099 very comfortable frequency for most 760 00:29:26,100 --> 00:29:27,129 neurons in the brain. 761 00:29:27,130 --> 00:29:29,219 They are usually usually at the 10 762 00:29:29,220 --> 00:29:30,329 to a 60 hertz range. 763 00:29:31,830 --> 00:29:33,719 This frequency come in at the back of the 764 00:29:33,720 --> 00:29:35,399 brain, along with everything else you 765 00:29:35,400 --> 00:29:37,559 usually see when 766 00:29:37,560 --> 00:29:39,149 you select the back of the brain with 767 00:29:39,150 --> 00:29:40,979 your EEG software. 768 00:29:40,980 --> 00:29:43,139 You will confine this confined back 769 00:29:43,140 --> 00:29:44,459 to this exact frequency in your 770 00:29:44,460 --> 00:29:45,460 spectrogram. 771 00:29:46,350 --> 00:29:48,209 Now, the cool thing is that it can pay 772 00:29:48,210 --> 00:29:50,369 attention to one checkerboard and ignore 773 00:29:50,370 --> 00:29:51,539 the other. 774 00:29:51,540 --> 00:29:53,549 And in your measurement, the frequency 775 00:29:53,550 --> 00:29:55,199 band of the attendu checkerboard becomes 776 00:29:55,200 --> 00:29:57,239 stronger and the other one becomes weaker 777 00:29:57,240 --> 00:29:58,809 when you do that. 778 00:29:58,810 --> 00:29:59,729 Was this trick? 779 00:29:59,730 --> 00:30:01,199 Your software can detect if your 780 00:30:01,200 --> 00:30:03,029 attention is on the one checkerboard or 781 00:30:03,030 --> 00:30:04,030 in the other. 782 00:30:04,830 --> 00:30:06,989 In practice, this paradigm works like 783 00:30:06,990 --> 00:30:08,189 mental buttons. 784 00:30:08,190 --> 00:30:10,170 You just press a button by looking at. 785 00:30:13,710 --> 00:30:15,749 They don't even have to be big, and you 786 00:30:15,750 --> 00:30:17,999 can run up a thousand, take a bottle 787 00:30:18,000 --> 00:30:19,439 at the same time. 788 00:30:19,440 --> 00:30:22,169 So next paradigm P300 789 00:30:22,170 --> 00:30:23,429 or Paul paradigm 790 00:30:24,750 --> 00:30:26,429 in psychology, these are two different 791 00:30:26,430 --> 00:30:28,229 things, but in this case, they're 792 00:30:28,230 --> 00:30:29,279 combined. 793 00:30:29,280 --> 00:30:30,809 The basic idea behind the oddball 794 00:30:30,810 --> 00:30:32,969 paradigm is that your brain always 795 00:30:32,970 --> 00:30:35,129 creates the same pattern under certain 796 00:30:35,130 --> 00:30:36,599 conditions. 797 00:30:36,600 --> 00:30:38,879 The three hundredth signal I showed 798 00:30:38,880 --> 00:30:41,039 in the next slide is generated 799 00:30:41,040 --> 00:30:43,349 when we see something unfamiliar or 800 00:30:43,350 --> 00:30:44,969 something that doesn't quite belong into 801 00:30:44,970 --> 00:30:45,970 our group. 802 00:30:46,560 --> 00:30:47,899 The signal is the strongest. 803 00:30:47,900 --> 00:30:50,039 A few hundred milliseconds after the 804 00:30:50,040 --> 00:30:51,419 odd ball. 805 00:30:51,420 --> 00:30:52,979 The strongest pick is positive, which is 806 00:30:52,980 --> 00:30:54,989 what the P stands for. 807 00:30:54,990 --> 00:30:57,119 And usually the upward paradigm works 808 00:30:57,120 --> 00:30:58,379 with images or sounds. 809 00:30:58,380 --> 00:30:59,969 But of course, you can do it with other 810 00:30:59,970 --> 00:31:00,970 senses as well. 811 00:31:03,030 --> 00:31:05,069 How do you analyze that? 812 00:31:05,070 --> 00:31:07,769 On the left, there's a reference P300. 813 00:31:07,770 --> 00:31:09,689 This is how it usually looks like when 814 00:31:09,690 --> 00:31:11,369 you get get rid of all the noise. 815 00:31:12,990 --> 00:31:14,969 Detecting the video in a continuous 816 00:31:14,970 --> 00:31:17,909 signal isn't completely straightforward 817 00:31:17,910 --> 00:31:19,349 because your incoming signal would be 818 00:31:19,350 --> 00:31:21,899 contaminated by other brain activity. 819 00:31:21,900 --> 00:31:23,609 So if you use a simple amplitude 820 00:31:23,610 --> 00:31:25,679 threshold, your you will trigger 821 00:31:25,680 --> 00:31:26,680 all the time. 822 00:31:27,600 --> 00:31:29,249 A good strategy, therefore, is to use a 823 00:31:29,250 --> 00:31:30,779 reference signal. 824 00:31:30,780 --> 00:31:32,519 You take the last five hundred 825 00:31:32,520 --> 00:31:34,619 milliseconds of your continuous signal 826 00:31:34,620 --> 00:31:36,809 and correlated with the reference 827 00:31:36,810 --> 00:31:37,769 wave. 828 00:31:37,770 --> 00:31:39,869 When the correlation is above 50 829 00:31:39,870 --> 00:31:41,819 percent or so, you know that the correct 830 00:31:41,820 --> 00:31:43,230 shape was actually in your signal. 831 00:31:44,250 --> 00:31:46,259 With this tool, you can even type on a 832 00:31:46,260 --> 00:31:47,879 keyboard with your mind. 833 00:31:47,880 --> 00:31:49,739 This is currently used for locked in 834 00:31:49,740 --> 00:31:52,619 patients, for example, or quadriplegics. 835 00:31:52,620 --> 00:31:55,019 Just look up the P 300 speller. 836 00:31:58,150 --> 00:32:00,219 Last paradigm, the event related to 837 00:32:00,220 --> 00:32:01,239 synchronization. 838 00:32:01,240 --> 00:32:02,260 This is a very popular one, 839 00:32:04,030 --> 00:32:05,889 event related to synchronization is 840 00:32:05,890 --> 00:32:09,249 really easy to set up and very reliable. 841 00:32:09,250 --> 00:32:11,919 You imagine moving one of your limbs, 842 00:32:11,920 --> 00:32:13,779 for example, of wiggling your right foot. 843 00:32:13,780 --> 00:32:15,399 You can actually do it, but it's not as 844 00:32:15,400 --> 00:32:17,409 impressive as this. 845 00:32:17,410 --> 00:32:19,569 Imagination causes a certain neuron 846 00:32:19,570 --> 00:32:21,519 clusters to synchronize or to 847 00:32:21,520 --> 00:32:23,709 synchronize. You probably remember the 848 00:32:23,710 --> 00:32:25,239 waves that spread out in the beginning 849 00:32:26,710 --> 00:32:28,929 and in the other spectrum, 850 00:32:28,930 --> 00:32:31,389 the synchronization will create a gap 851 00:32:31,390 --> 00:32:33,729 in the 10 to 13 852 00:32:33,730 --> 00:32:34,730 hertz range of 853 00:32:35,980 --> 00:32:37,209 the analysis. 854 00:32:37,210 --> 00:32:39,309 For this is very simple select the region 855 00:32:39,310 --> 00:32:41,439 from the top right side of your brain or 856 00:32:41,440 --> 00:32:43,869 left side. You have to consider that 857 00:32:43,870 --> 00:32:45,669 the left foot is connected to the right 858 00:32:45,670 --> 00:32:47,559 brain half and vice versa. 859 00:32:47,560 --> 00:32:49,779 And take the 860 00:32:49,780 --> 00:32:51,549 Fourier transform and 861 00:32:52,600 --> 00:32:54,699 look at a few frequencies between eight 862 00:32:54,700 --> 00:32:57,010 and 13 has four amplitude changes. 863 00:32:58,970 --> 00:33:00,759 You know your software triggers whenever 864 00:33:00,760 --> 00:33:03,189 you think of a moving foot or waving 865 00:33:03,190 --> 00:33:04,190 hands. 866 00:33:05,560 --> 00:33:07,479 And when you look for two regions of the 867 00:33:07,480 --> 00:33:09,909 brain at the same time to left and right, 868 00:33:09,910 --> 00:33:11,709 your software can actually detect which 869 00:33:11,710 --> 00:33:12,940 hand you're waving in your mind. 870 00:33:16,530 --> 00:33:18,659 So in summary, by an 871 00:33:18,660 --> 00:33:19,799 EEG headset. 872 00:33:19,800 --> 00:33:21,930 Choose your BCI software and get started. 873 00:33:23,100 --> 00:33:24,269 We are nearly finished. 874 00:33:24,270 --> 00:33:26,309 Just do a quick recap. 875 00:33:26,310 --> 00:33:27,779 Certain clusters in your brain 876 00:33:28,860 --> 00:33:30,599 do always the same thing at the same 877 00:33:30,600 --> 00:33:32,819 place, which is important 878 00:33:32,820 --> 00:33:35,429 for selecting a location. 879 00:33:35,430 --> 00:33:36,899 Memories are stored in the connection 880 00:33:36,900 --> 00:33:37,900 between neurons. 881 00:33:38,700 --> 00:33:40,889 Memories and neurons communicate with 882 00:33:40,890 --> 00:33:42,719 electric waves and certain frequencies. 883 00:33:43,890 --> 00:33:46,019 We want to find the clusters that do what 884 00:33:46,020 --> 00:33:48,119 we want, or 885 00:33:48,120 --> 00:33:50,039 we already know how to where to search 886 00:33:50,040 --> 00:33:52,319 from. For example, from my talk or from 887 00:33:52,320 --> 00:33:53,320 any other literature. 888 00:33:54,600 --> 00:33:56,429 But either way, we select an area on the 889 00:33:56,430 --> 00:33:57,599 activity map. 890 00:33:57,600 --> 00:34:00,389 We put the signal into our BCI paradigm 891 00:34:00,390 --> 00:34:02,349 and enter the internal feedback loop with 892 00:34:02,350 --> 00:34:03,350 the user. 893 00:34:04,080 --> 00:34:05,640 Have a lot of fun. That's it for today. 894 00:34:19,000 --> 00:34:20,559 Anyone with questions? 895 00:34:20,560 --> 00:34:22,658 Please line up. There are two microphones 896 00:34:22,659 --> 00:34:24,729 here and the idea there, if you 897 00:34:24,730 --> 00:34:26,859 leave, please leave quietly and 898 00:34:26,860 --> 00:34:28,629 there's a camera already running for the 899 00:34:28,630 --> 00:34:31,269 guy or crossing the picture. 900 00:34:31,270 --> 00:34:32,769 Other questions from the internets? 901 00:34:34,000 --> 00:34:36,309 Yes. No, maybe no. 902 00:34:36,310 --> 00:34:37,310 Then number two. 903 00:34:38,260 --> 00:34:40,479 OK, if you want to to steal some device 904 00:34:40,480 --> 00:34:43,059 with a with a from the activity, 905 00:34:43,060 --> 00:34:45,218 what is the advantage of first 906 00:34:45,219 --> 00:34:47,319 inferring? Rather, activity really comes 907 00:34:47,320 --> 00:34:50,408 from instead of just training 908 00:34:50,409 --> 00:34:52,178 your algorithm on the basis of the raw 909 00:34:52,179 --> 00:34:54,218 input signals of the edge anyway. 910 00:34:54,219 --> 00:34:55,839 Can you repeat the first part of your 911 00:34:55,840 --> 00:34:57,129 sentence? What is it? 912 00:34:57,130 --> 00:34:57,339 Yeah. 913 00:34:57,340 --> 00:34:58,569 So what is the advantage? 914 00:34:58,570 --> 00:35:00,699 What is the advantage of inferring 915 00:35:00,700 --> 00:35:02,829 the location of brain activity 916 00:35:02,830 --> 00:35:05,229 instead of directly using the raw egg 917 00:35:05,230 --> 00:35:06,550 input vectors as training 918 00:35:07,930 --> 00:35:09,099 the right? 919 00:35:09,100 --> 00:35:11,709 Exac-, no, it's actually not really 920 00:35:11,710 --> 00:35:14,049 raw because it's a mixture 921 00:35:14,050 --> 00:35:16,149 of several types of 922 00:35:16,150 --> 00:35:18,189 neural activities from the from different 923 00:35:18,190 --> 00:35:19,089 parts of the brain. 924 00:35:19,090 --> 00:35:21,669 So you get a mixture, and 925 00:35:21,670 --> 00:35:23,079 when you get this mixture, you, for 926 00:35:23,080 --> 00:35:26,049 example, get a lot of visual activity or 927 00:35:26,050 --> 00:35:27,759 much preparation. 928 00:35:27,760 --> 00:35:29,829 It's all very, very much 929 00:35:29,830 --> 00:35:32,349 a convoluted and with the localization, 930 00:35:32,350 --> 00:35:33,729 you can actually split them up into 931 00:35:33,730 --> 00:35:35,649 different regions so that you only get 932 00:35:35,650 --> 00:35:37,749 the one signal that 933 00:35:37,750 --> 00:35:38,889 did what you wanted to do. 934 00:35:40,390 --> 00:35:42,429 So what is your example? 935 00:35:42,430 --> 00:35:44,949 Is controlling something with 936 00:35:44,950 --> 00:35:47,109 this imagining your 937 00:35:47,110 --> 00:35:49,029 right foot, for example, and you would 938 00:35:49,030 --> 00:35:51,249 select you'd take the electrode that's 939 00:35:51,250 --> 00:35:52,929 around here. Then you will also get 940 00:35:52,930 --> 00:35:55,119 something from some visual also 941 00:35:55,120 --> 00:35:56,889 get something from audio. 942 00:35:56,890 --> 00:35:58,869 It's not really actually a signal, so it 943 00:35:58,870 --> 00:36:00,939 won't be as noise 944 00:36:00,940 --> 00:36:03,069 free as when you do with the localization 945 00:36:03,070 --> 00:36:04,070 before. 946 00:36:04,630 --> 00:36:05,019 OK. 947 00:36:05,020 --> 00:36:07,159 Number one yeah, 948 00:36:07,160 --> 00:36:09,369 do the when you're motivated 949 00:36:09,370 --> 00:36:11,559 using it for visual 950 00:36:11,560 --> 00:36:13,899 input, you gave some numbers about 951 00:36:13,900 --> 00:36:15,999 it being, I don't know, like 300 features 952 00:36:16,000 --> 00:36:17,769 per minute or something like that. 953 00:36:17,770 --> 00:36:18,770 How do you come up with that? 954 00:36:19,990 --> 00:36:21,130 I'm waving. Oh, right. 955 00:36:25,080 --> 00:36:26,080 And number two, 956 00:36:29,220 --> 00:36:31,679 I assume the option 957 00:36:31,680 --> 00:36:33,749 of putting sensors in the 958 00:36:33,750 --> 00:36:35,939 brain is too expensive, and that's 959 00:36:35,940 --> 00:36:38,249 why it wasn't discussed 960 00:36:38,250 --> 00:36:39,259 in this. 961 00:36:39,260 --> 00:36:41,459 I think you're right, but 962 00:36:41,460 --> 00:36:43,649 I've always wondered about some kind 963 00:36:43,650 --> 00:36:46,259 of middle ground, like 964 00:36:46,260 --> 00:36:48,779 as you showed in your slides, that 965 00:36:48,780 --> 00:36:50,909 the skull is one big barrier between 966 00:36:50,910 --> 00:36:53,129 the brain signals and their sensors. 967 00:36:53,130 --> 00:36:55,229 So why hasn't anybody 968 00:36:55,230 --> 00:36:57,449 proposed driving 969 00:36:57,450 --> 00:36:59,549 needles partially or completely 970 00:36:59,550 --> 00:37:01,529 through the skull to survey the same 971 00:37:01,530 --> 00:37:03,769 thing or? 972 00:37:03,770 --> 00:37:06,109 Or anything along those lines? 973 00:37:06,110 --> 00:37:07,909 That's like relatively 974 00:37:09,140 --> 00:37:10,940 easy and cheap to do. 975 00:37:12,050 --> 00:37:13,159 OK, I know what you mean. 976 00:37:13,160 --> 00:37:14,329 It's actually being done. 977 00:37:15,730 --> 00:37:17,839 These approaches were one of 978 00:37:17,840 --> 00:37:19,640 the first step in neuroscience 979 00:37:21,230 --> 00:37:23,539 at the very beginning because you, 980 00:37:23,540 --> 00:37:25,679 you see you, you, you, you, 981 00:37:25,680 --> 00:37:27,829 you've got a skull is a big 982 00:37:27,830 --> 00:37:29,329 barrier and you need to go somewhere 983 00:37:29,330 --> 00:37:31,429 around it. So it's the easiest to 984 00:37:31,430 --> 00:37:33,169 actually put the sensors directly onto 985 00:37:33,170 --> 00:37:35,629 the brain. This is called 986 00:37:35,630 --> 00:37:37,219 electrical cortical activity. 987 00:37:37,220 --> 00:37:39,619 It's it's done widely in 988 00:37:39,620 --> 00:37:41,059 basic research. 989 00:37:41,060 --> 00:37:43,689 But the problem is that you need a 990 00:37:43,690 --> 00:37:45,889 person person that's ready to open up 991 00:37:45,890 --> 00:37:48,109 their skull, which is usually 992 00:37:48,110 --> 00:37:49,549 only done when there's something wrong 993 00:37:49,550 --> 00:37:51,709 with their brain and it 994 00:37:51,710 --> 00:37:52,979 needs to be operated or something. 995 00:37:52,980 --> 00:37:55,129 So all the basic 996 00:37:55,130 --> 00:37:57,199 research is only getting maybe half an 997 00:37:57,200 --> 00:37:59,510 hour per person of research time. 998 00:38:00,530 --> 00:38:02,779 And the thing with needles in the brain 999 00:38:02,780 --> 00:38:04,069 isn't so far fetched. 1000 00:38:04,070 --> 00:38:05,899 It's actually being done with partners 1001 00:38:05,900 --> 00:38:07,849 and patients, for example. 1002 00:38:07,850 --> 00:38:09,319 The main problem with needles in the 1003 00:38:09,320 --> 00:38:11,629 brain is that they are 1004 00:38:11,630 --> 00:38:13,729 really foreign bodies in the brain. 1005 00:38:13,730 --> 00:38:16,099 They have a hell, 1006 00:38:16,100 --> 00:38:18,229 have a median lifespan of about six 1007 00:38:18,230 --> 00:38:20,689 months, and then they slowly start 1008 00:38:20,690 --> 00:38:22,789 stop working because the brain reacts 1009 00:38:22,790 --> 00:38:25,309 to metal objects within their 1010 00:38:25,310 --> 00:38:27,439 midst and 1011 00:38:27,440 --> 00:38:29,629 and starts to cover all the metallic 1012 00:38:29,630 --> 00:38:31,339 surfaces with with fat. 1013 00:38:31,340 --> 00:38:33,619 So it isn't such so 1014 00:38:33,620 --> 00:38:36,089 much of an electrical conductor anymore. 1015 00:38:36,090 --> 00:38:37,249 Yeah, but that's true. 1016 00:38:37,250 --> 00:38:38,780 Even if you like, 1017 00:38:39,860 --> 00:38:41,989 like, say, driving the needle 1018 00:38:41,990 --> 00:38:44,239 nine tenths of the thickness of the skull 1019 00:38:44,240 --> 00:38:46,399 towards the brain, that helps 1020 00:38:46,400 --> 00:38:47,659 nothing at all. Or that? 1021 00:38:47,660 --> 00:38:49,009 How does that work? That would help, 1022 00:38:49,010 --> 00:38:50,010 of course, 1023 00:38:51,120 --> 00:38:53,089 for 90 percent of the of the variance. 1024 00:38:53,090 --> 00:38:55,159 As far as I know, there are some people, 1025 00:38:55,160 --> 00:38:57,709 you know, put 1026 00:38:57,710 --> 00:38:59,539 things like on their head, you know, as 1027 00:38:59,540 --> 00:39:00,529 piercing or whatever. 1028 00:39:00,530 --> 00:39:01,530 So. 1029 00:39:02,980 --> 00:39:05,079 Yeah, actually, you could very well 1030 00:39:05,080 --> 00:39:07,329 go completely through the brain 1031 00:39:07,330 --> 00:39:09,009 as a car. 1032 00:39:09,010 --> 00:39:10,449 It doesn't matter so much because they're 1033 00:39:10,450 --> 00:39:12,879 still the dual motor, which separates the 1034 00:39:12,880 --> 00:39:15,489 brain from the skull, and they're 1035 00:39:15,490 --> 00:39:17,679 in my very weak 1036 00:39:17,680 --> 00:39:19,899 knowledge about your, uh, 1037 00:39:19,900 --> 00:39:21,999 your clinical stuff. 1038 00:39:22,000 --> 00:39:24,549 There's no really no real information 1039 00:39:24,550 --> 00:39:26,769 in there, but it could work. 1040 00:39:26,770 --> 00:39:28,479 Yes, it hasn't been done before because 1041 00:39:28,480 --> 00:39:30,819 this type of egg 1042 00:39:30,820 --> 00:39:32,529 hasn't become so mainstream that people 1043 00:39:32,530 --> 00:39:33,530 do it for fun. 1044 00:39:35,620 --> 00:39:36,849 But it's very, very, very possible. 1045 00:39:36,850 --> 00:39:37,899 Yes. 1046 00:39:37,900 --> 00:39:39,279 Questions from the internet. 1047 00:39:39,280 --> 00:39:41,829 So question number one 1048 00:39:41,830 --> 00:39:44,379 in this comparison, the last to, 1049 00:39:44,380 --> 00:39:46,509 uh, suppress the show 1050 00:39:46,510 --> 00:39:48,669 didn't have the localization options. 1051 00:39:48,670 --> 00:39:50,529 And what can't you do if you don't have 1052 00:39:50,530 --> 00:39:51,530 that 1053 00:39:52,090 --> 00:39:54,189 last part again, you 1054 00:39:54,190 --> 00:39:55,190 know? 1055 00:39:55,700 --> 00:39:56,799 Oh, OK. 1056 00:39:56,800 --> 00:39:58,719 What can you do if you are missing a 1057 00:39:58,720 --> 00:40:00,639 little PlayStation feature in your 1058 00:40:00,640 --> 00:40:01,209 software? 1059 00:40:01,210 --> 00:40:03,279 OK? You can't actually decode both the 1060 00:40:03,280 --> 00:40:05,409 different types of activity that 1061 00:40:05,410 --> 00:40:07,659 come into one electrode, which in 1062 00:40:07,660 --> 00:40:10,029 practice means your whole signal will be 1063 00:40:10,030 --> 00:40:12,469 a mixture of different activities and 1064 00:40:12,470 --> 00:40:14,619 your activity you are searching for is 1065 00:40:14,620 --> 00:40:16,719 only one, maybe one third of the 1066 00:40:16,720 --> 00:40:18,879 whole activity, which means you have 1067 00:40:18,880 --> 00:40:20,919 a higher practical noise level, which 1068 00:40:20,920 --> 00:40:22,839 makes it harder for your algorithms to 1069 00:40:22,840 --> 00:40:24,129 actually determine what, what, what 1070 00:40:24,130 --> 00:40:25,130 happened to this part. 1071 00:40:26,770 --> 00:40:28,509 Number two in the back? 1072 00:40:28,510 --> 00:40:30,070 Well, actually, I have two questions. 1073 00:40:31,360 --> 00:40:33,669 The first question is, you said the 1074 00:40:33,670 --> 00:40:35,259 data points you get, the better the 1075 00:40:35,260 --> 00:40:38,139 quality is. Well, look, the basic thing. 1076 00:40:38,140 --> 00:40:39,879 Did you ever think about trying to 1077 00:40:39,880 --> 00:40:42,069 enhance the conductivity of hair, 1078 00:40:42,070 --> 00:40:44,469 like put some graphite shampoo 1079 00:40:44,470 --> 00:40:45,969 in it or something to to? 1080 00:40:47,020 --> 00:40:48,460 I don't know. Maybe it's possible 1081 00:40:49,480 --> 00:40:51,789 to to use it as a sensor wire because 1082 00:40:51,790 --> 00:40:53,979 your hair only needs in German. 1083 00:40:53,980 --> 00:40:56,169 Well, you know, the better 1084 00:40:56,170 --> 00:40:58,239 hair is in the in the in 1085 00:40:58,240 --> 00:40:59,829 the skin, it goes a little bit into the 1086 00:40:59,830 --> 00:41:00,879 skin. Hmm. 1087 00:41:00,880 --> 00:41:03,309 So maybe you can use it to get a 1088 00:41:03,310 --> 00:41:04,479 real good grip. 1089 00:41:04,480 --> 00:41:06,909 OK, so the skin isn't our big barrier? 1090 00:41:08,050 --> 00:41:09,939 No. But you get a lot of data points. 1091 00:41:09,940 --> 00:41:12,069 Just, I mean, it's already in there. 1092 00:41:12,070 --> 00:41:14,019 You just need the cabling to attach a 1093 00:41:14,020 --> 00:41:15,929 kidney to your arms. 1094 00:41:16,930 --> 00:41:17,619 I don't know. 1095 00:41:17,620 --> 00:41:19,689 Just due to an issue, 1096 00:41:19,690 --> 00:41:21,400 they don't conduct electricity, you know? 1097 00:41:22,540 --> 00:41:24,819 So at some graphite or something, 1098 00:41:24,820 --> 00:41:25,239 graphite? 1099 00:41:25,240 --> 00:41:26,949 OK. I don't know. 1100 00:41:26,950 --> 00:41:28,839 Oh, that's actually not bad. 1101 00:41:28,840 --> 00:41:29,840 There's a lot of 1102 00:41:31,570 --> 00:41:34,599 there's a lot of chemistry going on. 1103 00:41:34,600 --> 00:41:36,069 Is it a barbershop? I don't know. 1104 00:41:36,070 --> 00:41:38,019 So if they make curls from it, they they 1105 00:41:38,020 --> 00:41:39,579 really do a lot to the chemistry in the 1106 00:41:39,580 --> 00:41:41,679 hair. So maybe you can apply it in 1107 00:41:41,680 --> 00:41:43,449 there. Just go to some somatic. 1108 00:41:43,450 --> 00:41:45,429 OK. I don't want to be the person that 1109 00:41:45,430 --> 00:41:46,899 connects that it connects the wire to 1110 00:41:46,900 --> 00:41:48,669 every hair, but I think it would be 1111 00:41:48,670 --> 00:41:49,670 possible. Yes. 1112 00:41:50,260 --> 00:41:51,309 But you 1113 00:41:51,310 --> 00:41:52,209 just suggestion. 1114 00:41:52,210 --> 00:41:54,309 And the second thing is, you showed 1115 00:41:54,310 --> 00:41:56,739 that if I want to move my food, 1116 00:41:56,740 --> 00:41:58,959 I have to think about it and you can show 1117 00:41:58,960 --> 00:42:00,609 that as a signal. 1118 00:42:00,610 --> 00:42:01,610 OK? 1119 00:42:02,050 --> 00:42:04,239 There is also a lot 1120 00:42:04,240 --> 00:42:05,679 of tension to my body that keeps me 1121 00:42:05,680 --> 00:42:08,049 upright. I don't think about 1122 00:42:08,050 --> 00:42:10,179 and I think that is 1123 00:42:11,740 --> 00:42:13,539 done somewhere in the back here in the 1124 00:42:13,540 --> 00:42:14,829 motor cortex. No, no, no. 1125 00:42:14,830 --> 00:42:16,359 The tension thing is happening in your 1126 00:42:16,360 --> 00:42:17,360 spine. 1127 00:42:18,520 --> 00:42:20,589 OK. Roll it back. I have narcolepsy, so I 1128 00:42:20,590 --> 00:42:22,779 have something which shuts 1129 00:42:22,780 --> 00:42:25,119 that off, OK, and I don't want it. 1130 00:42:25,120 --> 00:42:27,219 So what I try to do, I 1131 00:42:27,220 --> 00:42:29,319 want to do is to put 1132 00:42:29,320 --> 00:42:31,480 my brain in the loop, diagnose when this 1133 00:42:33,070 --> 00:42:35,199 tension goes away and give 1134 00:42:35,200 --> 00:42:36,819 some, let's say, 1135 00:42:37,990 --> 00:42:40,479 yeah, give some electrical impulse 1136 00:42:40,480 --> 00:42:42,789 to to my muscles to regain 1137 00:42:42,790 --> 00:42:45,039 that, OK, which the second part 1138 00:42:45,040 --> 00:42:47,919 is basically proven to be possible. 1139 00:42:47,920 --> 00:42:50,199 And now I know the sensory input and 1140 00:42:50,200 --> 00:42:51,279 put in the loop. 1141 00:42:51,280 --> 00:42:52,599 So can I. 1142 00:42:52,600 --> 00:42:55,449 Can I find that if you lose 1143 00:42:55,450 --> 00:42:57,549 the tension of all your muscles, 1144 00:42:57,550 --> 00:42:59,949 can I can I sense 1145 00:42:59,950 --> 00:43:02,169 that with you? 1146 00:43:02,170 --> 00:43:03,789 That's a basic question 1147 00:43:03,790 --> 00:43:05,709 you want to sense if you lose control of 1148 00:43:05,710 --> 00:43:07,449 your muscles right now. Yeah. 1149 00:43:07,450 --> 00:43:09,729 Like in sleep when you have a paralysis. 1150 00:43:09,730 --> 00:43:11,799 Mm hmm. 1151 00:43:11,800 --> 00:43:12,800 The easiest. 1152 00:43:13,480 --> 00:43:15,159 I'm not really completely familiar with 1153 00:43:15,160 --> 00:43:17,919 this basic, really basic motor 1154 00:43:17,920 --> 00:43:20,259 stuff, but my easiest 1155 00:43:20,260 --> 00:43:22,239 approach would be to use an electrode 1156 00:43:22,240 --> 00:43:23,379 actually on your muscles. 1157 00:43:23,380 --> 00:43:25,390 So on your forearm, for example, 1158 00:43:26,440 --> 00:43:28,539 because the muscles in contrast 1159 00:43:28,540 --> 00:43:30,669 to the head can generate a 1160 00:43:30,670 --> 00:43:32,739 lot of electricity, like 12 or 1161 00:43:32,740 --> 00:43:34,989 so, and you 1162 00:43:34,990 --> 00:43:36,729 can really easily measure that. 1163 00:43:36,730 --> 00:43:38,439 And when the swatters goes down, then you 1164 00:43:38,440 --> 00:43:40,569 have your signal. 1165 00:43:40,570 --> 00:43:42,639 OK, so there is a basic activity I always 1166 00:43:42,640 --> 00:43:43,089 have in my mouth. 1167 00:43:43,090 --> 00:43:45,340 And it's it's a it's it's the frequency 1168 00:43:46,360 --> 00:43:48,459 and the frequency goes up and 1169 00:43:48,460 --> 00:43:49,629 down. 1170 00:43:49,630 --> 00:43:51,759 And it's when you when you relax and the 1171 00:43:51,760 --> 00:43:53,530 frequency is low and when you 1172 00:43:55,270 --> 00:43:57,399 move it, then it becomes higher. 1173 00:43:57,400 --> 00:43:58,719 OK, thank you. 1174 00:43:58,720 --> 00:44:00,849 The internets and then number one in the 1175 00:44:00,850 --> 00:44:01,850 back. 1176 00:44:03,490 --> 00:44:05,249 The question. There was a question about 1177 00:44:05,250 --> 00:44:07,359 the organization of the hyper 1178 00:44:07,360 --> 00:44:09,409 columns, are they really strictly 1179 00:44:09,410 --> 00:44:12,419 hexagonal or is that a simplification? 1180 00:44:12,420 --> 00:44:13,329 No, of course not. 1181 00:44:13,330 --> 00:44:14,880 It wasn't very idealistic. 1182 00:44:15,930 --> 00:44:18,149 Every civilization, but 1183 00:44:18,150 --> 00:44:20,789 this hyper column 1184 00:44:20,790 --> 00:44:23,459 organization only applies for 1185 00:44:23,460 --> 00:44:25,739 what's what's for the few 1186 00:44:25,740 --> 00:44:27,389 flexible regions in the brain. 1187 00:44:27,390 --> 00:44:29,549 They have a lot of fixed purpose stuff 1188 00:44:29,550 --> 00:44:31,919 in the brain, and hyper columns 1189 00:44:31,920 --> 00:44:34,049 are used 1190 00:44:34,050 --> 00:44:35,549 for for a mixture of storage and 1191 00:44:35,550 --> 00:44:37,649 computation in the general purpose 1192 00:44:37,650 --> 00:44:38,699 areas. 1193 00:44:38,700 --> 00:44:40,979 And there they are organized like this, 1194 00:44:40,980 --> 00:44:43,019 but in other areas that can be, I don't 1195 00:44:43,020 --> 00:44:45,149 know, all over 1196 00:44:45,150 --> 00:44:46,150 the place. 1197 00:44:47,160 --> 00:44:48,160 No one in the back. 1198 00:44:49,320 --> 00:44:51,419 So everything we 1199 00:44:51,420 --> 00:44:53,070 talked about right now was about 1200 00:44:54,180 --> 00:44:56,669 giving visual signals and then recording 1201 00:44:56,670 --> 00:44:58,199 them somehow, right? 1202 00:44:58,200 --> 00:45:00,659 But can I can I 1203 00:45:00,660 --> 00:45:03,149 skip the eye and give those signals 1204 00:45:03,150 --> 00:45:05,279 without wasting 1205 00:45:05,280 --> 00:45:07,859 energy on the you know that 1206 00:45:07,860 --> 00:45:08,860 the entry 1207 00:45:10,350 --> 00:45:11,189 sensor? 1208 00:45:11,190 --> 00:45:13,739 Basically, there's you want to 1209 00:45:13,740 --> 00:45:15,629 ask if you can replace our creepy eyes 1210 00:45:15,630 --> 00:45:16,630 for something better? 1211 00:45:19,230 --> 00:45:22,169 Well, yeah, that could be 1212 00:45:22,170 --> 00:45:24,159 possible. Use case. 1213 00:45:24,160 --> 00:45:26,429 Ah, yes, of course. 1214 00:45:26,430 --> 00:45:27,779 Yes, you can do that. 1215 00:45:27,780 --> 00:45:29,549 The problem, again, is that you can't 1216 00:45:29,550 --> 00:45:31,139 really stick some needles in your in your 1217 00:45:31,140 --> 00:45:33,299 brain because they won't have 1218 00:45:33,300 --> 00:45:34,529 a long lifetime. 1219 00:45:34,530 --> 00:45:36,629 And the back of our eyes 1220 00:45:36,630 --> 00:45:39,619 is actually considered part of the brain, 1221 00:45:39,620 --> 00:45:41,879 a completely same organization. 1222 00:45:41,880 --> 00:45:43,349 It's the same cells. 1223 00:45:43,350 --> 00:45:45,299 Technically, it's just an extension of an 1224 00:45:45,300 --> 00:45:47,069 extension of the brain, so the retina 1225 00:45:47,070 --> 00:45:49,169 will perform the same way like 1226 00:45:49,170 --> 00:45:51,359 the rest of the brain does. 1227 00:45:51,360 --> 00:45:53,009 There's actually some research going on 1228 00:45:53,010 --> 00:45:55,109 with blind people, 1229 00:45:55,110 --> 00:45:57,749 but hopefully blind people 1230 00:45:57,750 --> 00:46:00,329 like vision impaired people 1231 00:46:00,330 --> 00:46:02,459 who have to low contrast in 1232 00:46:02,460 --> 00:46:03,749 their in their retinas. 1233 00:46:03,750 --> 00:46:05,879 They get implants which have put 1234 00:46:05,880 --> 00:46:07,499 directly onto the retina. 1235 00:46:07,500 --> 00:46:09,689 And when you put some water on there, 1236 00:46:09,690 --> 00:46:11,879 then they can see black and white images 1237 00:46:11,880 --> 00:46:13,919 again. Funny thing, actually, we don't 1238 00:46:13,920 --> 00:46:14,920 know how color works here, 1239 00:46:17,100 --> 00:46:19,169 but why actually 1240 00:46:19,170 --> 00:46:21,239 did they use a mouse to go forward 1241 00:46:21,240 --> 00:46:23,429 and backwards for your presentation? 1242 00:46:23,430 --> 00:46:25,889 Or how precise are your 1243 00:46:25,890 --> 00:46:28,259 recognizes of what you are thinking? 1244 00:46:28,260 --> 00:46:30,389 So why didn't you use your left and 1245 00:46:30,390 --> 00:46:31,390 right thumb? 1246 00:46:32,960 --> 00:46:35,169 Um, you mean for your 1247 00:46:35,170 --> 00:46:37,649 media should have used an EEG interface? 1248 00:46:37,650 --> 00:46:39,599 OK. Actually, maybe we forgot to stick 1249 00:46:39,600 --> 00:46:40,600 with us. 1250 00:46:43,420 --> 00:46:44,659 It's precise enough. 1251 00:46:44,660 --> 00:46:45,660 It would work. 1252 00:46:47,580 --> 00:46:49,909 I would be precise enough to work. 1253 00:46:49,910 --> 00:46:52,489 It wouldn't be 100 percent accurate. 1254 00:46:52,490 --> 00:46:54,349 But since we did in the presentation, we 1255 00:46:54,350 --> 00:46:55,609 actually have the possibility of 1256 00:46:55,610 --> 00:46:56,610 correcting. 1257 00:46:57,590 --> 00:46:59,989 We would somewhat 1258 00:46:59,990 --> 00:47:02,329 be somewhat comfortable to do this, but 1259 00:47:02,330 --> 00:47:04,879 we didn't dare overregulate. 1260 00:47:06,720 --> 00:47:07,720 Number two, 1261 00:47:08,420 --> 00:47:10,789 I've been looking for the possibilities 1262 00:47:10,790 --> 00:47:13,849 of outside the university 1263 00:47:13,850 --> 00:47:16,189 brain computer interface for some time. 1264 00:47:16,190 --> 00:47:18,289 And basically, you're the first team 1265 00:47:18,290 --> 00:47:20,359 who speaks positively about 1266 00:47:20,360 --> 00:47:22,489 the emotive 1267 00:47:22,490 --> 00:47:24,689 because I was focusing on 1268 00:47:24,690 --> 00:47:26,899 the open BCI and I wanted to even 1269 00:47:26,900 --> 00:47:29,119 create a hackathon, maybe 1270 00:47:29,120 --> 00:47:31,489 in what might be in Berlin somewhere 1271 00:47:31,490 --> 00:47:33,859 in Europe and invite 1272 00:47:33,860 --> 00:47:36,379 the open BCI people because 1273 00:47:36,380 --> 00:47:39,139 I saw no, 1274 00:47:39,140 --> 00:47:41,209 no good progress about a motive 1275 00:47:41,210 --> 00:47:44,129 and basically the right electrodes. 1276 00:47:44,130 --> 00:47:46,879 Uh, can you? 1277 00:47:46,880 --> 00:47:47,880 Have you 1278 00:47:49,130 --> 00:47:50,779 experimented with them? 1279 00:47:50,780 --> 00:47:53,299 And do you have any 1280 00:47:53,300 --> 00:47:56,299 expanding experiences you want to share? 1281 00:47:56,300 --> 00:47:57,859 You could share with us? 1282 00:47:57,860 --> 00:47:59,539 Yes, I can. 1283 00:47:59,540 --> 00:48:01,609 I actually bought an amateur book when it 1284 00:48:01,610 --> 00:48:04,459 came out. I think it was in 2008 1285 00:48:04,460 --> 00:48:06,679 also, and it 1286 00:48:06,680 --> 00:48:08,749 was the cheapest and the 1287 00:48:08,750 --> 00:48:10,849 most performant device at the 1288 00:48:10,850 --> 00:48:12,949 time. And until mid 1289 00:48:12,950 --> 00:48:14,869 of next year, it still is. 1290 00:48:14,870 --> 00:48:17,389 The my major issue with it 1291 00:48:17,390 --> 00:48:19,669 is that it's not such a great 1292 00:48:19,670 --> 00:48:21,349 build quality. 1293 00:48:21,350 --> 00:48:23,779 And the company 1294 00:48:23,780 --> 00:48:26,059 behind it seemed to have 1295 00:48:26,060 --> 00:48:28,249 regretted their decision to to make 1296 00:48:28,250 --> 00:48:30,169 it available. Somehow, I'm guessing it 1297 00:48:30,170 --> 00:48:32,269 was very expensive to develop. 1298 00:48:32,270 --> 00:48:34,369 They brought it out as as a 1299 00:48:34,370 --> 00:48:36,649 gamer accessory. 1300 00:48:36,650 --> 00:48:37,960 They wanted people to 1301 00:48:39,050 --> 00:48:40,819 code some games and use it as some 1302 00:48:40,820 --> 00:48:42,919 all-purpose kind of 1303 00:48:42,920 --> 00:48:44,239 input device. 1304 00:48:44,240 --> 00:48:46,129 This didn't really work out because there 1305 00:48:47,410 --> 00:48:50,329 there is a 1306 00:48:50,330 --> 00:48:52,429 API that's behind the whole system 1307 00:48:52,430 --> 00:48:54,229 is not really that great. 1308 00:48:54,230 --> 00:48:56,449 So it was used 1309 00:48:56,450 --> 00:48:58,609 a lot for research because it's the only 1310 00:48:58,610 --> 00:49:00,889 mobile headset right now. 1311 00:49:00,890 --> 00:49:02,929 It's not. It doesn't produce a lot. 1312 00:49:02,930 --> 00:49:04,279 It doesn't have a lot of channels. 1313 00:49:04,280 --> 00:49:05,839 I mean, in research, we're using more 1314 00:49:05,840 --> 00:49:08,149 like 128 or 256 1315 00:49:08,150 --> 00:49:10,569 channels. But in the comparison, 40 1316 00:49:10,570 --> 00:49:13,539 is the noteworthy. 1317 00:49:13,540 --> 00:49:16,369 And I'm 1318 00:49:16,370 --> 00:49:19,429 the epoch hasn't really 1319 00:49:19,430 --> 00:49:21,079 progressed in the last five years 1320 00:49:21,080 --> 00:49:23,179 anywhere. I'm not sure this 1321 00:49:23,180 --> 00:49:24,180 it has any future. 1322 00:49:25,550 --> 00:49:28,279 But other than that, I have 1323 00:49:28,280 --> 00:49:31,249 made quite positive experiences with it. 1324 00:49:31,250 --> 00:49:32,689 Besides that, it only runs on Mac and 1325 00:49:32,690 --> 00:49:34,549 Windows, which is quite graphic because I 1326 00:49:34,550 --> 00:49:37,669 want to do it on my smartphone and 1327 00:49:37,670 --> 00:49:39,739 it has a relatively 1328 00:49:39,740 --> 00:49:41,029 nice data quality. 1329 00:49:41,030 --> 00:49:43,129 It has actually electrodes, which is 1330 00:49:43,130 --> 00:49:45,649 also nice for some mobile applications, 1331 00:49:45,650 --> 00:49:48,019 and it has a reasonably 1332 00:49:48,020 --> 00:49:50,539 OK C++ API, 1333 00:49:50,540 --> 00:49:52,729 which you can use to just get it out 1334 00:49:52,730 --> 00:49:54,799 of there and get it into something into a 1335 00:49:54,800 --> 00:49:56,329 software package that actually can deal 1336 00:49:56,330 --> 00:49:58,129 with brain signals. 1337 00:49:58,130 --> 00:49:59,539 OK. Number one Yeah, 1338 00:50:01,250 --> 00:50:02,119 OK. 1339 00:50:02,120 --> 00:50:04,309 This is sort of related 1340 00:50:04,310 --> 00:50:06,409 to the first question. 1341 00:50:06,410 --> 00:50:08,569 Do you know, are there any attempts 1342 00:50:08,570 --> 00:50:10,939 to make the, 1343 00:50:10,940 --> 00:50:13,129 uh, the imaging more precise by 1344 00:50:13,130 --> 00:50:15,589 using biofeedback for the 1345 00:50:15,590 --> 00:50:17,809 actual calibration 1346 00:50:17,810 --> 00:50:18,810 of the abuses? 1347 00:50:19,800 --> 00:50:22,879 Um, biofeedback. 1348 00:50:22,880 --> 00:50:25,219 Um, can you 1349 00:50:25,220 --> 00:50:26,479 clarify that? 1350 00:50:26,480 --> 00:50:28,969 Yes, there was an art project 1351 00:50:28,970 --> 00:50:30,799 a few years ago that I heard about Code 1352 00:50:30,800 --> 00:50:33,319 Brain Avatar, which used 1353 00:50:33,320 --> 00:50:35,929 the direct outputs of the e.g. 1354 00:50:35,930 --> 00:50:38,359 electrodes fed into 1355 00:50:38,360 --> 00:50:40,519 a water basin, which was then 1356 00:50:40,520 --> 00:50:43,519 projected through an overhead projector. 1357 00:50:43,520 --> 00:50:45,979 So you saw the the 1358 00:50:45,980 --> 00:50:48,079 wave interference patterns 1359 00:50:48,080 --> 00:50:50,959 that were produced by the U.S. 1360 00:50:50,960 --> 00:50:53,359 and that had some quite interesting 1361 00:50:53,360 --> 00:50:56,389 sort of biofeedback effects, too, so 1362 00:50:56,390 --> 00:50:58,549 people could actually 1363 00:50:58,550 --> 00:51:00,619 try to influence the 1364 00:51:00,620 --> 00:51:03,019 the the patterns 1365 00:51:03,020 --> 00:51:05,179 produced by their brain activities. 1366 00:51:05,180 --> 00:51:07,399 And this this system used, 1367 00:51:07,400 --> 00:51:09,169 I think, eight electrodes or something. 1368 00:51:09,170 --> 00:51:10,969 And I wondered if it's possible to 1369 00:51:10,970 --> 00:51:13,039 perhaps by adding 1370 00:51:13,040 --> 00:51:14,749 another layer which is less directly 1371 00:51:14,750 --> 00:51:17,809 accessible, perhaps increase the 1372 00:51:17,810 --> 00:51:19,969 the overall sort of 1373 00:51:19,970 --> 00:51:21,199 granularity of the 1374 00:51:22,430 --> 00:51:24,499 yes, actually just 1375 00:51:24,500 --> 00:51:26,479 a real time activity map is one of the 1376 00:51:26,480 --> 00:51:28,909 easiest problems 1377 00:51:28,910 --> 00:51:30,559 that you can you can do with a such an 1378 00:51:30,560 --> 00:51:31,639 EEG. 1379 00:51:31,640 --> 00:51:33,469 It's it's I think it's even built in in 1380 00:51:33,470 --> 00:51:35,689 most, most software packages. 1381 00:51:35,690 --> 00:51:37,939 Um, you biofeedback. 1382 00:51:37,940 --> 00:51:40,099 It's actually something of a more 1383 00:51:40,100 --> 00:51:42,289 complicated. Setup, because 1384 00:51:42,290 --> 00:51:45,019 you have a feedback loop to the computer 1385 00:51:45,020 --> 00:51:46,429 and you, for example, 1386 00:51:47,660 --> 00:51:50,509 use them as a measure of your 1387 00:51:50,510 --> 00:51:52,699 of your alpha waves or something 1388 00:51:52,700 --> 00:51:54,859 to influence something within 1389 00:51:54,860 --> 00:51:56,959 within a computer and then displays your 1390 00:51:56,960 --> 00:51:59,419 back if you're doing OK or not. 1391 00:51:59,420 --> 00:52:01,489 In today's research, 1392 00:52:01,490 --> 00:52:02,999 this is already quite advanced. 1393 00:52:03,000 --> 00:52:04,249 It's Buford PRETEXTING. 1394 00:52:04,250 --> 00:52:06,919 You can learn to meditate before that. 1395 00:52:06,920 --> 00:52:09,619 And my most amazing 1396 00:52:09,620 --> 00:52:11,839 story was that you 1397 00:52:11,840 --> 00:52:14,419 can actually learn how to shoot a ball 1398 00:52:14,420 --> 00:52:15,949 without actually learning how to shoot a 1399 00:52:15,950 --> 00:52:18,169 bow. But you in the scanner and you try 1400 00:52:18,170 --> 00:52:20,359 to move one point towards another point. 1401 00:52:20,360 --> 00:52:22,219 And when you have done that, your brain 1402 00:52:22,220 --> 00:52:24,919 has has sufficiently changed 1403 00:52:24,920 --> 00:52:26,449 so that you can actually sort of a 1404 00:52:27,520 --> 00:52:28,750 small number two. 1405 00:52:29,810 --> 00:52:31,939 When I watched a string of popular 1406 00:52:31,940 --> 00:52:34,189 questions and I 1407 00:52:34,190 --> 00:52:36,409 just wanted to know if you have 1408 00:52:36,410 --> 00:52:39,169 an outlook about the ultimate privacy 1409 00:52:39,170 --> 00:52:41,449 intrusion by reading 1410 00:52:41,450 --> 00:52:43,519 what we are thinking about 1411 00:52:43,520 --> 00:52:46,159 or something like that have any timescale 1412 00:52:46,160 --> 00:52:48,529 to say, maybe next five 1413 00:52:48,530 --> 00:52:50,119 or 10 years, 1414 00:52:50,120 --> 00:52:52,279 OK, unless we solve 1415 00:52:52,280 --> 00:52:55,219 the whole censorship in the brain issue, 1416 00:52:55,220 --> 00:52:57,439 we are not nearly close to the 1417 00:52:57,440 --> 00:52:58,849 reading thoughts stage. 1418 00:52:58,850 --> 00:53:00,949 We actually need nano 1419 00:53:00,950 --> 00:53:03,199 sensors, ideally 1420 00:53:03,200 --> 00:53:05,509 at each mini 1421 00:53:05,510 --> 00:53:07,609 column, which is around, I 1422 00:53:07,610 --> 00:53:09,709 don't know, several millions of 1423 00:53:09,710 --> 00:53:11,809 sensors streaming 1424 00:53:11,810 --> 00:53:13,909 in real time towards a very 1425 00:53:13,910 --> 00:53:16,549 advanced brain computer learning 1426 00:53:16,550 --> 00:53:18,619 interface that it has 1427 00:53:18,620 --> 00:53:20,659 to be trained first. 1428 00:53:20,660 --> 00:53:22,759 What its neurons of your what is 1429 00:53:22,760 --> 00:53:24,729 neurons of your brain is actually meaning 1430 00:53:24,730 --> 00:53:25,699 its output. 1431 00:53:25,700 --> 00:53:27,889 So it's it's a completely 1432 00:53:27,890 --> 00:53:29,149 deliberate approach. 1433 00:53:29,150 --> 00:53:31,159 You have to opt into this because 1434 00:53:31,160 --> 00:53:33,589 otherwise the technical limitations are 1435 00:53:33,590 --> 00:53:35,829 huge. And this is the time 1436 00:53:35,830 --> 00:53:37,399 I'm speaking of a timescale of at least 1437 00:53:37,400 --> 00:53:38,400 at least 20 years. 1438 00:53:39,570 --> 00:53:41,299 OK, number one. 1439 00:53:41,300 --> 00:53:42,979 OK, great. 1440 00:53:42,980 --> 00:53:45,289 So I think it would be 1441 00:53:45,290 --> 00:53:47,989 awesome if we could read 1442 00:53:47,990 --> 00:53:49,849 something more out of our brains. 1443 00:53:49,850 --> 00:53:52,039 But just the movement of 1444 00:53:52,040 --> 00:53:54,139 limbs or things that 1445 00:53:54,140 --> 00:53:55,140 we see. 1446 00:53:55,670 --> 00:53:57,809 And what 1447 00:53:57,810 --> 00:53:59,719 would increasing the number of sensors 1448 00:54:00,890 --> 00:54:03,649 say from now, hundreds to thousands 1449 00:54:03,650 --> 00:54:04,779 helping this? 1450 00:54:04,780 --> 00:54:05,780 Yes. 1451 00:54:06,140 --> 00:54:08,089 No. Not yet. 1452 00:54:08,090 --> 00:54:10,549 I mean, hundred 120 1453 00:54:10,550 --> 00:54:12,709 are the medical. And yet you can 1454 00:54:12,710 --> 00:54:14,509 only buy one with 1455 00:54:14,510 --> 00:54:16,819 256 256 now. 1456 00:54:16,820 --> 00:54:18,979 Blood? Yeah, OK. 1457 00:54:18,980 --> 00:54:20,629 And if you increase the number of 1458 00:54:20,630 --> 00:54:22,759 sensors, every sensor would help you to 1459 00:54:22,760 --> 00:54:23,989 get a better result. 1460 00:54:23,990 --> 00:54:25,649 So thousands would be yeah. 1461 00:54:26,660 --> 00:54:28,969 But the more the better. 1462 00:54:28,970 --> 00:54:31,069 So why are they all the systems that 1463 00:54:31,070 --> 00:54:33,229 we see just with these sensors 1464 00:54:33,230 --> 00:54:35,179 every few centimeters and not every 1465 00:54:35,180 --> 00:54:36,180 millimeter 1466 00:54:36,950 --> 00:54:38,989 is the main reason is actually price. 1467 00:54:38,990 --> 00:54:40,579 It's just so simple. 1468 00:54:40,580 --> 00:54:42,679 An active electrode one 1469 00:54:42,680 --> 00:54:44,839 tunnel costs around five hundred 1470 00:54:44,840 --> 00:54:46,519 and twenty euros for medical reasons. 1471 00:54:46,520 --> 00:54:48,229 Economy of scale. 1472 00:54:48,230 --> 00:54:50,479 Yes, and I'm expecting that it becomes 1473 00:54:50,480 --> 00:54:51,949 cheaper over time. But the problem is 1474 00:54:51,950 --> 00:54:54,139 that we have to be against 1475 00:54:54,140 --> 00:54:56,359 magnitudes of more sensors and 1476 00:54:56,360 --> 00:54:57,949 the price is only going to it's going 1477 00:54:57,950 --> 00:54:59,299 down slowly. 1478 00:54:59,300 --> 00:55:00,949 OK. Number two 1479 00:55:00,950 --> 00:55:01,789 Yeah. 1480 00:55:01,790 --> 00:55:04,279 Given the plasticity of the brain 1481 00:55:04,280 --> 00:55:06,439 and the tendency to kind of 1482 00:55:06,440 --> 00:55:08,539 move functions around to see 1483 00:55:08,540 --> 00:55:09,969 when parts of the brain is damaged, 1484 00:55:11,150 --> 00:55:12,709 these columns are they're easily 1485 00:55:12,710 --> 00:55:15,259 identifiable and are identifiable 1486 00:55:15,260 --> 00:55:17,329 by the signals or. 1487 00:55:17,330 --> 00:55:18,219 No, not at all. 1488 00:55:18,220 --> 00:55:20,599 You're not. The resolution of today's 1489 00:55:20,600 --> 00:55:22,789 ages are nowhere near identifying 1490 00:55:22,790 --> 00:55:23,929 single columns. 1491 00:55:23,930 --> 00:55:26,129 We still need 15 years for that and 1492 00:55:26,130 --> 00:55:28,309 the approximate according to sensor 1493 00:55:28,310 --> 00:55:29,549 increases. 1494 00:55:29,550 --> 00:55:31,789 OK, number one in the back? 1495 00:55:31,790 --> 00:55:34,070 Um, just to continue the question before 1496 00:55:35,270 --> 00:55:37,459 you said, you can make them make 1497 00:55:37,460 --> 00:55:39,949 people that are blind, see 1498 00:55:39,950 --> 00:55:41,809 black and white images kind of vision 1499 00:55:41,810 --> 00:55:44,239 impaired vision vision. 1500 00:55:44,240 --> 00:55:46,789 Sorry, but 1501 00:55:46,790 --> 00:55:48,859 can I show them pictures 1502 00:55:48,860 --> 00:55:50,420 that are not those that they see? 1503 00:55:51,710 --> 00:55:52,710 Yes. 1504 00:55:53,280 --> 00:55:55,520 So I already know how, how it's encoded. 1505 00:55:56,660 --> 00:55:58,819 We know how the retina and we know 1506 00:55:58,820 --> 00:56:00,289 how the back of the brain is encoded, 1507 00:56:00,290 --> 00:56:00,799 which would 1508 00:56:00,800 --> 00:56:02,779 be a match, and we know how to create the 1509 00:56:02,780 --> 00:56:03,229 signals 1510 00:56:03,230 --> 00:56:04,789 that we need. It's also no problem. 1511 00:56:04,790 --> 00:56:06,269 We don't know how color. 1512 00:56:06,270 --> 00:56:06,609 Let's see. 1513 00:56:06,610 --> 00:56:08,719 Yeah, but that's a side effect if 1514 00:56:08,720 --> 00:56:11,119 you want to actually do that. 1515 00:56:11,120 --> 00:56:12,679 If you actually want to input signals 1516 00:56:12,680 --> 00:56:14,599 into the brain, then you shouldn't use 1517 00:56:14,600 --> 00:56:15,919 your retinal. You should use the back of 1518 00:56:15,920 --> 00:56:17,689 your brain because there's much higher 1519 00:56:17,690 --> 00:56:19,039 resolution there. 1520 00:56:19,040 --> 00:56:19,639 Thank you. 1521 00:56:19,640 --> 00:56:19,999 OK. 1522 00:56:20,000 --> 00:56:22,189 Number two So I had 1523 00:56:22,190 --> 00:56:24,409 a conversation with a neuroscientist 1524 00:56:24,410 --> 00:56:26,629 about a year ago 1525 00:56:26,630 --> 00:56:29,329 about much of the same issues, 1526 00:56:29,330 --> 00:56:32,299 rather two of our technical difficulties 1527 00:56:32,300 --> 00:56:33,349 he encountered. 1528 00:56:33,350 --> 00:56:35,699 One is that the data 1529 00:56:35,700 --> 00:56:38,030 from decent is really high, 1530 00:56:39,890 --> 00:56:41,109 too is. 1531 00:56:41,110 --> 00:56:44,199 That's in twice with that's 1532 00:56:44,200 --> 00:56:46,149 where they implanted electrodes. 1533 00:56:47,200 --> 00:56:49,689 The areas 1534 00:56:49,690 --> 00:56:51,849 on the boy intended to move 1535 00:56:51,850 --> 00:56:54,789 around, so the needles 1536 00:56:54,790 --> 00:56:57,309 that you stuck in one day didn't exactly 1537 00:56:57,310 --> 00:56:59,739 correspond didn't exactly communicate 1538 00:56:59,740 --> 00:57:01,839 with the same area anymore after 1539 00:57:01,840 --> 00:57:03,789 my OK because the area itself moved. 1540 00:57:05,140 --> 00:57:07,449 So I don't know, um, 1541 00:57:07,450 --> 00:57:08,450 how, 1542 00:57:10,500 --> 00:57:12,579 how well this is. 1543 00:57:12,580 --> 00:57:14,349 That's just the waiting 1544 00:57:14,350 --> 00:57:16,659 areas themselves and humans 1545 00:57:16,660 --> 00:57:18,849 at least don't really move that much, but 1546 00:57:18,850 --> 00:57:21,099 single neurons can move up to 1547 00:57:21,100 --> 00:57:23,079 centimeters per hour. 1548 00:57:23,080 --> 00:57:26,379 So I would 1549 00:57:26,380 --> 00:57:28,509 say that it is an issue, but 1550 00:57:28,510 --> 00:57:29,920 with our current technology, 1551 00:57:31,570 --> 00:57:33,909 what we do is we stick at most 1552 00:57:33,910 --> 00:57:36,129 64 electrodes in the brain 1553 00:57:36,130 --> 00:57:38,379 and this at a grid of one millimeter. 1554 00:57:38,380 --> 00:57:40,059 It shouldn't be that much of an issue. 1555 00:57:40,060 --> 00:57:42,219 If it's an issue, then you have to 1556 00:57:42,220 --> 00:57:44,469 recalibrate after every 1557 00:57:44,470 --> 00:57:46,059 week or so. I don't know. 1558 00:57:46,060 --> 00:57:47,199 OK. 1559 00:57:47,200 --> 00:57:48,520 OK, thank you. I want to get to know.