1 00:00:00,152 --> 00:00:03,655 That person happens to be there and he absolutely hated Quantum 2 00:00:03,799 --> 00:00:07,630 Healing. And I was like, so what are you doing here? But 3 00:00:07,630 --> 00:00:11,462 then he was, again, I know he is a big scientist, 4 00:00:12,119 --> 00:00:15,870 but he was like saying, yeah, this is technically, this is a scam. 5 00:00:16,543 --> 00:00:20,375 And we talked like for a few hours. I couldn't convince him. He 6 00:00:20,375 --> 00:00:24,126 couldn't convince me, but that, that's the mentality of some people. Again, 7 00:00:24,158 --> 00:00:25,649 I cannot be sure if it's right or wrong. 8 00:00:28,670 --> 00:00:32,265 Curious. What a strange little factoid, because I've heard that too 9 00:00:32,924 --> 00:00:36,764 through completely different contexts where there's a lot of 10 00:00:37,037 --> 00:00:40,760 disbelief in quantum annealing, yet it's proven commercially at 11 00:00:40,760 --> 00:00:44,245 least to be one of the, one of the more popular 12 00:00:44,358 --> 00:00:48,156 options. Whether it's real or not might be 13 00:00:48,172 --> 00:00:51,649 up for debate. That's fascinating. Speed is 14 00:00:51,828 --> 00:00:55,404 just a byproduct. Alireza Khodayi, senior quantum computing 15 00:00:55,436 --> 00:00:59,203 scientist at Nelnet, on why a consistently biased forecast 16 00:00:59,203 --> 00:01:02,874 beats a lucky one and where quantum actually pays off in finance. 17 00:01:03,291 --> 00:01:06,786 Quantum Podcast, turn it up fast. Candace and 18 00:01:06,802 --> 00:01:10,569 Frank blowing my mind at last. Quantum Podcast, they're 19 00:01:10,569 --> 00:01:14,400 breaking the mold. Science has got beats to fold. Hello 20 00:01:14,576 --> 00:01:18,007 and welcome back to Impact Quantum, the podcast where we explore the emerging 21 00:01:18,119 --> 00:01:21,597 industry of quantum computing, where you don't need to be a PhD in 22 00:01:21,646 --> 00:01:24,531 physics, you just need to be a little bit curious. And with that is the 23 00:01:24,531 --> 00:01:27,983 most With me is the most quantum curious person I know, 24 00:01:28,368 --> 00:01:32,013 Candace Cahouli. How's it going, Candace? It's great. I'm really excited to 25 00:01:32,013 --> 00:01:35,690 record again, and we have a really interesting guest today. 26 00:01:36,734 --> 00:01:40,041 We're going to speak with Alireza Kodai, 27 00:01:40,989 --> 00:01:43,975 and I— please excuse me for any mispronunciation there. 28 00:01:44,746 --> 00:01:48,583 And he is the senior quantum computing scientist at Nelnet. 29 00:01:49,161 --> 00:01:52,872 How are you today? I'm fine, thank you. How about you? 30 00:01:53,610 --> 00:01:56,791 I'm good. The sun is shining. No complaints here. 31 00:01:57,434 --> 00:02:01,112 Yeah. Also, thank you for having me. No problem. That's a very 32 00:02:01,225 --> 00:02:04,855 cool title, by the way. Senior Quantum Computing Scientist. Yeah. 33 00:02:04,903 --> 00:02:08,325 Thank you. Thank you. So what do you do as a— in that role? 34 00:02:08,839 --> 00:02:11,939 Because it sounds descriptive and vague all at the same time. 35 00:02:13,176 --> 00:02:16,630 Yeah, sure. So I start with the 36 00:02:16,630 --> 00:02:20,325 quantum computing part. When you think about it, quantum computing, 37 00:02:21,149 --> 00:02:24,950 it's a newer paradigm that we have after 38 00:02:25,031 --> 00:02:28,133 all classical computing that we have been using for 39 00:02:28,133 --> 00:02:31,386 probably 3 decades now, seriously. 40 00:02:33,206 --> 00:02:36,733 So with quantum computing, we are— there are myths about quantum 41 00:02:36,733 --> 00:02:40,574 computing. I would say the first advantage is not necessarily 42 00:02:40,574 --> 00:02:43,355 the speedup. It's about the 43 00:02:43,869 --> 00:02:46,902 problems that classical computing start to 44 00:02:47,572 --> 00:02:51,326 reaching the ceiling of the computational power, so they cannot 45 00:02:51,326 --> 00:02:54,056 solve it as effectively as we expect. 46 00:02:54,842 --> 00:02:58,279 And there are few fields 47 00:02:58,391 --> 00:03:01,249 so far that quantum computers 48 00:03:02,148 --> 00:03:05,857 is getting proofs that probably they 49 00:03:06,002 --> 00:03:09,165 are the better chance. And those fields include 50 00:03:09,727 --> 00:03:12,376 pharmaceutical, molecular, 51 00:03:13,066 --> 00:03:16,498 design, and also in logistics, aviation, 52 00:03:16,755 --> 00:03:19,305 finance, all those sort of things. 53 00:03:20,299 --> 00:03:24,100 But when you think about it, in any of these fields, the algorithms we do 54 00:03:24,709 --> 00:03:28,190 essentially boils down to a few mathematical 55 00:03:28,462 --> 00:03:32,295 problems, most of which is optimization. There 56 00:03:32,311 --> 00:03:35,615 are other things like Monte Carlo simulations and things like that. 57 00:03:36,272 --> 00:03:39,176 So basically what I'm doing is applying the 58 00:03:40,187 --> 00:03:43,557 quantum computing and quantum algorithms to financial 59 00:03:43,621 --> 00:03:47,233 problems, if that makes sense. So it— 60 00:03:47,313 --> 00:03:50,539 okay, so it does make sense. So how do you 61 00:03:50,619 --> 00:03:53,701 determine which quantum algorithms have genuine 62 00:03:54,150 --> 00:03:57,649 commercial potential versus 63 00:03:58,243 --> 00:04:01,806 those that are scientifically fascinating? 64 00:04:03,122 --> 00:04:06,636 Yeah, that's a great question. And this is 65 00:04:06,732 --> 00:04:10,518 actually one of the challenges that I have. And I 66 00:04:10,518 --> 00:04:13,855 guess every practitioner in quantum computing has, 67 00:04:14,513 --> 00:04:18,026 because quantum computing, most of it so far have been in 68 00:04:18,139 --> 00:04:21,909 labs and R&D initiatives. Now we are 69 00:04:21,909 --> 00:04:25,727 getting transitioned into real-world commercial application 70 00:04:25,824 --> 00:04:29,498 of it, which you need to explain the money 71 00:04:29,530 --> 00:04:33,232 value of any quantum algorithm you want to use 72 00:04:33,232 --> 00:04:37,062 in any industry for people that are 73 00:04:37,062 --> 00:04:40,748 not necessarily quantum computing scientists or even close to that. 74 00:04:41,293 --> 00:04:45,058 People like CEOs, CFOs, these people just know 75 00:04:45,074 --> 00:04:48,568 about the business, not necessarily the technicality of 76 00:04:48,600 --> 00:04:52,366 quantum computing. But back to your question, at least 77 00:04:52,366 --> 00:04:56,099 in finance, we have certain problems that 78 00:04:56,677 --> 00:05:00,463 If you make a little bit, like a fraction of a 79 00:05:00,463 --> 00:05:04,281 percent improvement, that translates to a lot of 80 00:05:04,378 --> 00:05:07,731 money. You can think about the modeling of 81 00:05:07,827 --> 00:05:11,372 risk in different forms. You can think about the 82 00:05:11,597 --> 00:05:15,079 deriving prices for things like bonds or 83 00:05:15,079 --> 00:05:18,881 securities. And a lot of these examples necessarily 84 00:05:18,993 --> 00:05:21,753 not a better prediction, I would say, 85 00:05:22,687 --> 00:05:26,517 It's more about more consistency. For example, my 86 00:05:26,613 --> 00:05:30,270 own PhD was on forecasting volatility 87 00:05:30,286 --> 00:05:34,088 using quantum algorithm. And one of the results that I've 88 00:05:34,088 --> 00:05:37,649 seen that makes me interested to continue that research was 89 00:05:38,403 --> 00:05:42,237 when you use quantum algorithm and you do the forecasting of 90 00:05:42,237 --> 00:05:46,038 volatility, you can have a 91 00:05:46,103 --> 00:05:49,920 consistent bias forecast, but it's consistent. So you know 92 00:05:50,434 --> 00:05:54,269 if it is overestimating or underestimating. then you 93 00:05:54,269 --> 00:05:57,707 can offset it. But if you use classical algorithm 94 00:05:58,045 --> 00:06:01,885 with the same problem, sometimes you get very accurate forecast, 95 00:06:01,933 --> 00:06:05,532 sometimes you get very bad forecast. So when you get the 96 00:06:05,532 --> 00:06:08,666 result, you are not sure how much you can depend on it. 97 00:06:09,389 --> 00:06:12,892 So I would say in any case, when you see a problem 98 00:06:13,004 --> 00:06:15,784 that is, that is mathematically 99 00:06:16,121 --> 00:06:19,351 problematic, that makes it inconsistent results, 100 00:06:19,912 --> 00:06:23,596 or you have mathematical intractabilities like 101 00:06:23,906 --> 00:06:27,578 it is not differentiable, it is getting to plateau on optimization space, 102 00:06:28,656 --> 00:06:32,102 you would refer to quantum. Okay, so it was a long 103 00:06:32,520 --> 00:06:35,195 answer to a short question, but yeah, feel free to ask me more. 104 00:06:38,048 --> 00:06:41,810 First question I have is, I'm coming hot 105 00:06:41,810 --> 00:06:45,475 off the heels of being at KubeCon last week, and the 106 00:06:45,685 --> 00:06:49,456 question there is, was all about how do you 107 00:06:49,456 --> 00:06:52,583 get AI proofs of concept into production, right? 108 00:06:53,449 --> 00:06:57,056 I would imagine, and that's got its own challenges 109 00:06:57,056 --> 00:07:00,616 attached to it, I would imagine that we're having similar 110 00:07:00,744 --> 00:07:04,592 kind of discussions, or will have these, plenty of these discussions when it gets to 111 00:07:04,673 --> 00:07:08,312 how do you get a quantum computing solution into production, right? 112 00:07:08,441 --> 00:07:12,225 It goes from one thing to have a proof of concept, the 113 00:07:12,225 --> 00:07:15,918 POC. How do you get this thing into production when— Not a 114 00:07:15,918 --> 00:07:17,408 lot of people understand it. 115 00:07:19,720 --> 00:07:23,414 Yeah, the fact is— sorry, go ahead. So 116 00:07:23,479 --> 00:07:27,287 if we expect a production or we define production at the 117 00:07:27,287 --> 00:07:31,000 level of classical computing, honestly, I don't think you get any 118 00:07:31,000 --> 00:07:34,695 production with quantum computing. But if we just narrow down the 119 00:07:34,728 --> 00:07:38,355 definition to, hey, I have this algorithm, I want to 120 00:07:38,355 --> 00:07:41,773 use it in a real-world application, and The 121 00:07:41,853 --> 00:07:45,681 production is not here, like something like you have, I 122 00:07:45,713 --> 00:07:49,123 don't know, fault tolerancing, or if something goes wrong, there is a 123 00:07:49,300 --> 00:07:52,658 failover because most of the 124 00:07:52,803 --> 00:07:56,597 quantum vendors, they have their own stack, they 125 00:07:56,678 --> 00:08:00,140 have their own software library, and they are not 126 00:08:00,140 --> 00:08:03,965 necessarily uniform. You can do things in Qiskit 127 00:08:04,077 --> 00:08:07,535 or PennyLane, but essentially when you have a 128 00:08:07,535 --> 00:08:11,231 quantum software running on a machine, There is 129 00:08:11,295 --> 00:08:15,015 no guarantee that you can run the same quantum software 130 00:08:15,416 --> 00:08:19,168 on another machine with the same level of quality or efficiency 131 00:08:19,280 --> 00:08:22,792 or whatever metrics you have, right? Which is one of the big 132 00:08:22,792 --> 00:08:26,544 aspects of production, meaning that you go in 133 00:08:26,800 --> 00:08:30,472 production level, you don't rely, you don't depend on 134 00:08:30,520 --> 00:08:34,080 the hardware level. You can maintain a quality of service 135 00:08:34,336 --> 00:08:38,040 that is guaranteed, right? With quantum stacks so 136 00:08:38,040 --> 00:08:41,331 far, I would say we cannot guarantee that because the 137 00:08:41,395 --> 00:08:44,430 results would be different based on 138 00:08:45,489 --> 00:08:48,733 many things. The noise pattern that we have in quantum 139 00:08:48,941 --> 00:08:52,634 hardware, the length of the circuit, the number 140 00:08:52,714 --> 00:08:56,423 of qubits, so many variants. But we are 141 00:08:56,423 --> 00:09:00,004 getting there. And I have seen many trends 142 00:09:00,100 --> 00:09:03,777 that they work on things like the quantum 143 00:09:04,114 --> 00:09:07,923 compilers, quantum orchestration level that basically 144 00:09:07,923 --> 00:09:11,095 sit on top of the quantum hardware 145 00:09:11,774 --> 00:09:15,512 to abstract away differences in hardware 146 00:09:15,577 --> 00:09:18,765 and get to the point that you can send a job 147 00:09:19,235 --> 00:09:22,941 to be run on a quantum machine regardless of what is the 148 00:09:22,941 --> 00:09:26,617 underlying stack is. So we are, we are 149 00:09:26,651 --> 00:09:29,971 getting there. You talked 150 00:09:29,971 --> 00:09:33,126 about consistency and then you 151 00:09:33,289 --> 00:09:36,528 mentioned risk. So can quantum 152 00:09:36,625 --> 00:09:39,939 computing help us understand uncertainty? 153 00:09:42,860 --> 00:09:46,443 Yes. When you think about quantum computing, it's essentially a 154 00:09:46,572 --> 00:09:50,220 probabilistic computational paradigm, 155 00:09:51,088 --> 00:09:54,416 which is very different from classical, which is like 156 00:09:54,481 --> 00:09:56,358 deterministic computational paradigm. 157 00:09:59,202 --> 00:10:02,658 And for any quantum algorithm, for any— I would say even any 158 00:10:03,132 --> 00:10:06,613 quantum operation that you do. I give you a very 159 00:10:07,062 --> 00:10:09,792 simple example. So when you use classical 160 00:10:10,338 --> 00:10:13,742 computers, you can do 2+ 2, they have a 161 00:10:13,774 --> 00:10:17,211 definitive answer, right? Quantum computers are not 162 00:10:17,323 --> 00:10:20,663 built for that kind of definitive computation. 163 00:10:21,337 --> 00:10:25,079 It's more about you do your operation and then 164 00:10:25,175 --> 00:10:29,013 at the end of operation, you measure your qubit, 165 00:10:29,045 --> 00:10:32,815 which is the equivalent of a bit. And the result 166 00:10:32,815 --> 00:10:35,989 of the measurement comes in the terms of probability 167 00:10:36,726 --> 00:10:40,365 or a probabilistic distribution function, PDF, right? 168 00:10:40,846 --> 00:10:44,469 So it's essentially a probabilistic computational 169 00:10:44,581 --> 00:10:48,092 paradigm, and hence it natively embeds 170 00:10:49,070 --> 00:10:52,676 all those nuances of probability paradigms. For 171 00:10:52,676 --> 00:10:56,267 example, I earlier mentioned Monte Carlo simulation. 172 00:10:56,844 --> 00:11:00,515 If you want to do it with classical computers, you have to do it 173 00:11:00,691 --> 00:11:04,477 a lot of times, right? Sampling, basically. sampling the discrete points, 174 00:11:05,425 --> 00:11:08,684 while with quantum computers you have the whole 175 00:11:08,813 --> 00:11:12,490 continuous space covered in one 176 00:11:12,490 --> 00:11:16,038 qubit and then you measure it. So essentially you 177 00:11:16,343 --> 00:11:20,165 are not doing discrete steps, you just do 178 00:11:20,694 --> 00:11:24,387 the operation once and then you get the result. In that sense, you 179 00:11:24,387 --> 00:11:27,727 can say it's more speed, but again, as I said, it's 180 00:11:28,652 --> 00:11:32,428 I guess speed is just a byproduct of what 181 00:11:32,444 --> 00:11:35,533 we do in quantum computers. We have a lot more 182 00:11:36,067 --> 00:11:37,426 advantages than speed. 183 00:11:40,112 --> 00:11:43,381 Interesting. Is your 184 00:11:43,526 --> 00:11:47,161 LinkedIn headline image kind of implies that you're in a finance 185 00:11:47,161 --> 00:11:50,922 company? What sort of opportunities and risks does 186 00:11:51,100 --> 00:11:53,989 quantum computing open up for the finance world? 187 00:11:55,945 --> 00:11:58,960 I start with opportunities. I would say the best opportunity 188 00:11:59,698 --> 00:12:03,482 for financial practitioners, and I'm saying this because I have worked 189 00:12:03,482 --> 00:12:05,535 with a variety of 190 00:12:06,289 --> 00:12:08,630 financial domain 191 00:12:08,630 --> 00:12:12,222 experts, risk analysis, 192 00:12:12,286 --> 00:12:15,830 whatever. The first one I already told you, the consistency of the 193 00:12:15,830 --> 00:12:18,348 results. They would love to see something 194 00:12:18,348 --> 00:12:21,764 consistently biased 195 00:12:22,149 --> 00:12:25,132 rather than having something that, you know, inconsistent. 196 00:12:26,479 --> 00:12:29,334 But then beside that, I would say 197 00:12:30,328 --> 00:12:34,113 there are certain mathematical problems that if you 198 00:12:34,113 --> 00:12:37,625 do it with classical computing inference, 199 00:12:38,764 --> 00:12:42,532 you don't get even— and in some cases you don't get— you cannot get 200 00:12:42,532 --> 00:12:46,237 any result. And optimization is one example. 201 00:12:46,558 --> 00:12:49,845 If you have like a non-convex optimization problem, 202 00:12:50,631 --> 00:12:53,807 no matter how much GPU or how much 203 00:12:53,983 --> 00:12:57,704 computational power you throw at it, you cannot 204 00:12:57,704 --> 00:13:01,424 solve it because it's mathematically not possible to do it with 205 00:13:01,617 --> 00:13:05,193 the algorithms we do in classical. Or another thing 206 00:13:05,289 --> 00:13:08,994 that frequently happens in finance, and especially in risk 207 00:13:09,555 --> 00:13:13,019 forecasting, is the plateaus that comes in 208 00:13:13,163 --> 00:13:16,706 optimization space in which you lose the gradient. 209 00:13:16,754 --> 00:13:20,104 Again, no matter how much GPU you have, when you lose 210 00:13:20,537 --> 00:13:23,951 the gradient on a space, you cannot continue. So you lose— 211 00:13:24,945 --> 00:13:28,680 you end up with non-convergence of optimization. 212 00:13:29,338 --> 00:13:32,127 So these are certain mathematical problems that 213 00:13:33,185 --> 00:13:36,423 you cannot solve efficiently or you cannot solve at all 214 00:13:37,273 --> 00:13:40,883 with classical computer. And then you have 215 00:13:40,916 --> 00:13:44,654 to approximate, you have to get proxies close to the 216 00:13:44,654 --> 00:13:48,371 solution, but you are not getting close enough 217 00:13:48,660 --> 00:13:52,422 to the actual solution. And again, this is 218 00:13:52,534 --> 00:13:56,119 finance. As I said, I get a small percentage of improvement 219 00:13:56,718 --> 00:14:00,130 translates into probably more money. Yeah. 220 00:14:01,812 --> 00:14:05,611 Interesting. So what would meaningful quantum advantage 221 00:14:05,692 --> 00:14:06,559 look like in finance? 222 00:14:09,943 --> 00:14:12,847 That's a tricky question because I know there are 223 00:14:13,618 --> 00:14:16,703 proponents and opponents of quantum advantage in general. 224 00:14:17,699 --> 00:14:20,829 But again, depending on how you define the quantum advantage, 225 00:14:21,493 --> 00:14:23,096 the way I would define it is 226 00:14:25,314 --> 00:14:28,162 something that gives you a competitive advantage to your 227 00:14:29,502 --> 00:14:33,231 opponents in the market, right? And again, we are not 228 00:14:33,231 --> 00:14:36,605 talking about a big advantage. It could be very small, like marginal. 229 00:14:37,422 --> 00:14:40,982 Advantage. And as you know, in financial 230 00:14:41,511 --> 00:14:45,279 market, everyone that has survived, they 231 00:14:45,424 --> 00:14:49,256 are almost at par, right? So imagine all those big bankers 232 00:14:49,352 --> 00:14:53,185 or regional bankers, and anyone that survives 233 00:14:53,297 --> 00:14:56,985 means that they partition market between them. So now 234 00:14:57,145 --> 00:15:00,994 imagine one of those parties have a little 235 00:15:00,994 --> 00:15:04,281 bit of competitive advantage with regard to a better 236 00:15:05,022 --> 00:15:08,816 estimation of the risk or a better pricing. So imagine you, you 237 00:15:09,073 --> 00:15:12,806 are a bank, right? You want to give the loan and then you 238 00:15:12,999 --> 00:15:16,668 want to suggest a better interest rate considering 239 00:15:16,845 --> 00:15:20,643 all the factors that everyone else does. But then you want 240 00:15:20,771 --> 00:15:24,232 your interest rate to be more appealing to the customer, right? 241 00:15:24,617 --> 00:15:28,014 So you have a better way of estimating the risk. 242 00:15:28,334 --> 00:15:32,070 So your risk premium is more tailored to 243 00:15:32,070 --> 00:15:35,809 your customers. So most probably they, they take your offer, 244 00:15:36,081 --> 00:15:39,547 right? And then you can get a bigger portion of the 245 00:15:39,547 --> 00:15:43,285 market. And this is just like a simple example. Another example is 246 00:15:43,478 --> 00:15:47,055 if you are issuing securities and you have 247 00:15:47,328 --> 00:15:51,114 a better way of, you know, incorporating everything in it— what is the 248 00:15:51,114 --> 00:15:54,933 market expectation, what is the risk, whatever— then 249 00:15:54,933 --> 00:15:58,445 you're offering in terms of price is much more 250 00:15:58,494 --> 00:16:02,157 appealing than the others. So I would say, 251 00:16:03,880 --> 00:16:06,820 at least for myself, the expectation of quantum is not 252 00:16:07,688 --> 00:16:11,190 abruptively big. It is just 253 00:16:11,769 --> 00:16:15,163 whatever you can do on top of whatever you are already 254 00:16:15,261 --> 00:16:19,070 offering. A better accuracy, a wider outreach 255 00:16:19,507 --> 00:16:23,326 because you have included many factors in your conclusion. 256 00:16:24,166 --> 00:16:27,675 in your offering, things like that. I would say it's changing 257 00:16:27,675 --> 00:16:31,464 future, but for now, again, back to what we discussed, 258 00:16:31,528 --> 00:16:35,012 because we don't have that much of production-level 259 00:16:35,638 --> 00:16:39,266 software, production-level model, that's as we can do now. 260 00:16:40,294 --> 00:16:43,826 So we talk about quantum computing and we talk about AI 261 00:16:44,275 --> 00:16:47,711 as separate technologies. What happens when we begin 262 00:16:47,808 --> 00:16:49,798 combining quantum algorithms 263 00:16:55,474 --> 00:16:57,248 Wow, you get the big brother. 264 00:16:59,120 --> 00:17:02,826 But joke aside, I have been in this 265 00:17:02,939 --> 00:17:06,086 conversation so many times from the 266 00:17:06,635 --> 00:17:09,798 different perspectives. One question I always get is like how 267 00:17:10,508 --> 00:17:14,251 AI can help quantum computing to progress and vice 268 00:17:14,315 --> 00:17:17,814 versa, how quantum computing can get AI progress, 269 00:17:18,520 --> 00:17:22,209 and then now you are throwing in the finance, and then it becomes even more 270 00:17:22,209 --> 00:17:25,369 complicated. But I guess it is not 271 00:17:25,914 --> 00:17:29,603 mutually exclusive. All those fields can 272 00:17:29,603 --> 00:17:33,132 help each other, especially AI and quantum. 273 00:17:33,613 --> 00:17:36,548 We already have these quantum machine learning, 274 00:17:37,623 --> 00:17:41,456 and it's not just a concept, it's an actual thing, right? There are many 275 00:17:41,472 --> 00:17:44,568 algorithms, machine learning algorithms that run on 276 00:17:45,182 --> 00:17:48,664 quantum from scratch or quantum 277 00:17:48,696 --> 00:17:52,309 enhanced or quantum inspired like tensor networks. 278 00:17:52,935 --> 00:17:56,757 So each of these is like a different paradigm of introducing how 279 00:17:56,950 --> 00:18:00,707 quantum can contribute to AI. The other way around is 280 00:18:00,964 --> 00:18:04,417 also correct. I have seen many algorithms, for 281 00:18:04,417 --> 00:18:08,046 example, estimating or forecasting the 282 00:18:08,046 --> 00:18:11,595 noise pattern in quantum machines so we can have a 283 00:18:11,659 --> 00:18:14,999 better or a more reliable quantum 284 00:18:15,452 --> 00:18:19,286 Computation. What else? From AI to quantum, I have seen 285 00:18:19,286 --> 00:18:23,075 many— not many, actually. I have seen a few efforts for 286 00:18:23,926 --> 00:18:27,409 using the AI to design the quantum 287 00:18:27,409 --> 00:18:30,476 circuit and then try AI 288 00:18:31,150 --> 00:18:34,746 to come up with the best circuit given some metrics, the number of 289 00:18:34,843 --> 00:18:38,262 qubits, whatever. And then so you don't have to 290 00:18:39,354 --> 00:18:43,015 do like a trial and error as human. Right. So these 291 00:18:43,112 --> 00:18:46,179 are the ways that these 2 fields can help each other. 292 00:18:46,902 --> 00:18:50,709 But of course, finance can get help and get contribution 293 00:18:50,757 --> 00:18:54,339 from both sides. As a matter of fact, I'm working on that 294 00:18:54,677 --> 00:18:58,371 intersection as well. I've been working on a few projects 295 00:18:58,371 --> 00:19:01,568 that use a combination of AI and 296 00:19:01,905 --> 00:19:05,374 quantum forces to further enhance 297 00:19:05,599 --> 00:19:09,132 financial algorithms. But I would say the problem 298 00:19:09,453 --> 00:19:13,147 of working in finance in industry is that you cannot 299 00:19:13,147 --> 00:19:16,920 publish many of this, right? And this is not just me. I know 300 00:19:17,000 --> 00:19:20,838 many friends that work in other financial institutes have 301 00:19:20,886 --> 00:19:24,659 big achievements, but all we can do is just talking about it. 302 00:19:24,707 --> 00:19:27,870 You cannot, you know, talking about how we did it. Yeah, I mean, it's a 303 00:19:27,999 --> 00:19:31,355 very competitive industry, right? When quants first came 304 00:19:31,628 --> 00:19:35,408 into being, quants would figure things out years before anyone else 305 00:19:35,408 --> 00:19:39,047 and they would not share it. I would imagine it's the same 306 00:19:39,047 --> 00:19:42,880 setup. You're describing a very similar dynamic here, right? You may have figured out 307 00:19:42,961 --> 00:19:46,401 how the market works, right? How to predict it, but no one's 308 00:19:46,401 --> 00:19:49,825 necessarily going to want to shout that from the rooftops. 309 00:19:52,080 --> 00:19:55,774 Yeah. Yeah. That's so true. And it 310 00:19:55,807 --> 00:19:59,550 breaks my heart. I did a PhD. I shared the knowledge. I 311 00:19:59,679 --> 00:20:03,465 read the knowledge shared by the others. But I guess that, as 312 00:20:03,481 --> 00:20:06,996 you said, dynamics in industry, especially in finance industry, is 313 00:20:07,028 --> 00:20:10,384 different. I would say even a patent 314 00:20:10,384 --> 00:20:13,578 application, having a patent is another thing 315 00:20:14,413 --> 00:20:17,463 in most cases in finance because you don't want to share it. 316 00:20:18,186 --> 00:20:21,958 You are more running on the trade secrets rather 317 00:20:22,039 --> 00:20:25,522 than patents. Again, difference from other 318 00:20:25,891 --> 00:20:28,267 industries like pharmaceutical chemistry. 319 00:20:29,675 --> 00:20:33,519 Most of the engineering fields, they just share but patent it. Finance 320 00:20:33,519 --> 00:20:37,092 is not like that. But I guess eventually people, 321 00:20:38,830 --> 00:20:42,462 even though separately, come up with the very similar ideas 322 00:20:42,640 --> 00:20:46,079 at the end, and then probably they converge at some point. 323 00:20:46,192 --> 00:20:49,690 Everyone uses whatever algorithm for quantum 324 00:20:49,690 --> 00:20:53,406 optimization, so it's not a secret anymore. You will see it. It 325 00:20:53,406 --> 00:20:57,008 just appears everywhere. I guess that also has 326 00:20:57,104 --> 00:21:00,510 happened for optimization using discrete algorithms. 327 00:21:01,362 --> 00:21:04,961 Everyone knows about that, which is a good and bad thing 328 00:21:05,057 --> 00:21:08,832 at the same time. I remember probably 5 years ago, 329 00:21:08,865 --> 00:21:12,672 it created a hype. Hey, let's apply QAOO to 330 00:21:12,704 --> 00:21:15,757 optimization and boom, you get your best algorithm. 331 00:21:16,544 --> 00:21:20,287 And then later it turned out that it's a hype. You can 332 00:21:20,335 --> 00:21:23,406 have some advantages, but not as big as you 333 00:21:23,953 --> 00:21:27,750 might expect it. But again, yeah, this is a work in 334 00:21:27,798 --> 00:21:31,395 progress, how you contribute, how you let the others know 335 00:21:31,475 --> 00:21:35,148 without losing your secrets. That's a big question. 336 00:21:38,587 --> 00:21:42,103 I'd be so curious, what does a typical day look like for 337 00:21:42,103 --> 00:21:44,955 someone doing the kind of work you're doing? 338 00:21:47,371 --> 00:21:50,956 I would say, of course, R&D is a big part of it, right? So 339 00:21:52,708 --> 00:21:56,483 You learn from the field and 340 00:21:56,483 --> 00:22:00,162 then you come up with some ideas, some hunches that, hey, if we can 341 00:22:00,162 --> 00:22:03,472 apply this and that to such and such problem, 342 00:22:04,178 --> 00:22:07,135 we will get some advantage, right? So you have to 343 00:22:07,649 --> 00:22:11,151 experience that as well and repeat it enough times 344 00:22:11,504 --> 00:22:15,087 to be sure that, hey, yeah, there is some advantages. It is 345 00:22:15,199 --> 00:22:18,781 not just by chance or it is not because of noise. Which is 346 00:22:18,893 --> 00:22:22,550 funny. Sometimes you get some advantages because of noise, 347 00:22:23,303 --> 00:22:26,639 and then when you repeat it, it just fades out. But then 348 00:22:27,393 --> 00:22:31,082 the biggest part of it is not technical. It's 349 00:22:31,082 --> 00:22:34,850 about convincing people in 350 00:22:34,930 --> 00:22:38,010 leadership, executives that don't 351 00:22:38,298 --> 00:22:41,073 necessarily care what is your new solution or new 352 00:22:41,891 --> 00:22:45,643 approach is about. It's about how much it can save 353 00:22:45,740 --> 00:22:49,285 money or how much it can generate money. All right. 354 00:22:49,589 --> 00:22:52,815 So you have to translate all those technical 355 00:22:52,815 --> 00:22:56,619 advantages into some keys, some PKIs 356 00:22:57,534 --> 00:23:01,321 that they can understand and they can appreciate. At the end 357 00:23:01,385 --> 00:23:05,157 of the day, it is not important how much I am excited as 358 00:23:05,253 --> 00:23:08,591 a quantum computing scientist. It is important, at least in 359 00:23:08,639 --> 00:23:12,282 industry, it is important that you be able to explain 360 00:23:13,037 --> 00:23:16,840 those results in a simple language understandable by 361 00:23:17,049 --> 00:23:20,598 business side. And you can provide enough 362 00:23:20,743 --> 00:23:24,166 proof for that. So it is at least as 363 00:23:24,343 --> 00:23:27,704 important and as big as the technical part. Sometimes it's much 364 00:23:28,204 --> 00:23:31,995 more even challenging. And as far as I can see, 365 00:23:32,092 --> 00:23:34,884 people like me, they always have 366 00:23:35,954 --> 00:23:38,270 challenge and sometimes trouble in that part. 367 00:23:39,842 --> 00:23:43,330 Anyone that reads a technical paper can come up with ideas, 368 00:23:43,430 --> 00:23:46,792 execute them, get some results, but then how you 369 00:23:46,792 --> 00:23:50,564 translate it to something understandable and money-wise, 370 00:23:50,919 --> 00:23:54,429 at least in finance again. So yeah, I 371 00:23:54,429 --> 00:23:58,276 imagine translating these terms for mere mortals to someone 372 00:23:58,276 --> 00:24:02,026 who's in the C-suite who may not be a quantum physicist, particularly in finance. 373 00:24:03,140 --> 00:24:05,848 Oh yeah, that's got to be a tough— that's got to be a tough part 374 00:24:05,848 --> 00:24:09,363 of the job. So to be honest with you, I just 375 00:24:09,379 --> 00:24:12,536 predicted this when I was doing my PhD, 376 00:24:13,290 --> 00:24:17,058 I happened to start an MBA and 377 00:24:17,058 --> 00:24:20,747 got an MBA with a specialization in finance 378 00:24:21,388 --> 00:24:25,028 for one big reason, that I'd be able to translate 379 00:24:25,798 --> 00:24:29,598 what I get from technicality results into some finance. So 380 00:24:30,304 --> 00:24:33,383 I tried to make myself prepared for this, 381 00:24:34,377 --> 00:24:37,953 but even with 2 advanced degrees, I'm sometimes— I'm more— 382 00:24:38,017 --> 00:24:41,369 not sometimes, most of the times I'm challenging to 383 00:24:41,834 --> 00:24:45,606 explain it. And then I realized this is not just me. So I 384 00:24:45,606 --> 00:24:48,544 happen to be in a few external 385 00:24:49,315 --> 00:24:53,088 consortiums and roundtables that try 386 00:24:53,152 --> 00:24:55,785 to progress quantum computing in finance. 387 00:24:56,845 --> 00:25:00,682 And I can name the name, but I don't think it's proper for 388 00:25:00,714 --> 00:25:04,262 here, but I will assure you that big 389 00:25:04,487 --> 00:25:08,260 names, like very big names that you know in finance, they have 390 00:25:08,260 --> 00:25:11,278 the same problem. There are people, they have decades of 391 00:25:11,876 --> 00:25:15,642 quantum teams. I know a bank, a few banks that have been invested 392 00:25:15,642 --> 00:25:19,476 in this billion dollars. They have quantum teams for 393 00:25:19,476 --> 00:25:22,942 decades. Still, they have this challenge, how to translate the results 394 00:25:23,583 --> 00:25:27,065 that talks to some financial experts and financial 395 00:25:27,113 --> 00:25:30,963 executives. So we have a big way to 396 00:25:30,963 --> 00:25:34,653 go in that regard. That makes sense, right? And you look at the 397 00:25:34,717 --> 00:25:38,567 big names that have publicly announced they have a quantum division, I can't 398 00:25:38,583 --> 00:25:41,436 imagine that every one of those CEOs or board members is 399 00:25:41,932 --> 00:25:45,747 familiar with the inner workings of quantum physics. 400 00:25:45,795 --> 00:25:49,225 But kudos on you for getting the degree to be ready for it because you 401 00:25:49,225 --> 00:25:53,007 saw that coming. A lot of technologists don't see that if they 402 00:25:53,007 --> 00:25:56,790 can't communicate what they know, they're left on an island on 403 00:25:56,838 --> 00:26:00,636 their own. Yeah. Again, probably it was by chance that I thought about that 404 00:26:00,700 --> 00:26:04,515 and got my MBA. You're definitely right. Yeah, I have seen many 405 00:26:04,547 --> 00:26:08,346 people, very intelligent, very smart. Very good in their 406 00:26:08,458 --> 00:26:11,954 field, but then they can communicate in a way that they 407 00:26:12,708 --> 00:26:16,461 educated for. They can communicate in terms of technical papers, 408 00:26:16,493 --> 00:26:19,957 journal articles, but doesn't communicate with 409 00:26:20,101 --> 00:26:23,292 business people. Even in startups, I've been 410 00:26:23,533 --> 00:26:27,141 consulting some startups, the same problem, right? 411 00:26:27,397 --> 00:26:30,974 You have a best algorithm and then you want to get some funds. 412 00:26:31,807 --> 00:26:35,205 So how you communicate to the funders? Right. That's a big 413 00:26:35,205 --> 00:26:38,974 challenge as well. That's a good point. That 414 00:26:38,974 --> 00:26:41,626 was the whole joke of The Big Bang Theory, right, is that they could not 415 00:26:41,626 --> 00:26:45,418 communicate with normal people, right? Yeah, exactly. Sorry, Candace, I cut you off. 416 00:26:45,467 --> 00:26:49,188 No, no problem at all. Was there a particular problem in finance that 417 00:26:49,188 --> 00:26:52,355 made you think this is exactly the kind of 418 00:26:52,404 --> 00:26:55,964 challenge quantum computing might eventually help us solve? Yes. 419 00:26:59,414 --> 00:27:03,250 I mentioned that in an earlier conversation, but any— 420 00:27:03,427 --> 00:27:07,126 I think any problem that boils down to a 421 00:27:07,126 --> 00:27:10,002 mathematical intractable 422 00:27:11,561 --> 00:27:15,257 problem, it's better to be challenged with quantum. And I 423 00:27:15,257 --> 00:27:18,006 give you a few instances. I gave a few 424 00:27:18,778 --> 00:27:22,507 already. So if you have a mathematical problem that 425 00:27:22,507 --> 00:27:25,734 gets non-convex at some point, or gets 426 00:27:25,734 --> 00:27:29,284 disjoint in their space, or 427 00:27:29,412 --> 00:27:33,138 gets the— cannot be differentiated mathematically, like you 428 00:27:33,203 --> 00:27:37,009 cannot get the derivatives on it. There are a few features I can 429 00:27:37,379 --> 00:27:40,719 actually name, like mathematical features. If you see those 430 00:27:40,767 --> 00:27:44,478 mathematical red flags in your problem, 431 00:27:45,248 --> 00:27:48,910 then you can think about quantum computers because those red flags 432 00:27:49,103 --> 00:27:52,874 eventually cause your classical algorithm 433 00:27:52,874 --> 00:27:55,934 to a stop at some point and then get to the 434 00:27:56,623 --> 00:28:00,261 quantum. And from the classical point of view, 435 00:28:00,565 --> 00:28:04,411 it also makes sense. Every time in mathematics that 436 00:28:04,411 --> 00:28:07,295 we fail to get a deterministic answer, 437 00:28:08,241 --> 00:28:11,141 you sort of just resort to 438 00:28:11,237 --> 00:28:14,987 statistics instead of mathematics. Now that we cannot get a 439 00:28:14,987 --> 00:28:17,919 definitive answer, we just talk about probability. 440 00:28:18,804 --> 00:28:22,199 of this answer being the answer that we are looking for, that kind of 441 00:28:22,718 --> 00:28:26,284 talking. And then quantum computing, as I said, is the 442 00:28:26,284 --> 00:28:29,510 best probabilistic computational paradigm. 443 00:28:30,219 --> 00:28:33,668 I guess in summary, when you don't get a definitive answer 444 00:28:34,023 --> 00:28:37,821 in any problem, including in finance, quantum 445 00:28:37,821 --> 00:28:38,873 computing is your friend. 446 00:28:42,567 --> 00:28:45,498 And the other way around, I have to say, if you know you can get 447 00:28:45,854 --> 00:28:49,665 the deterministic answer for any problem, 448 00:28:49,923 --> 00:28:53,726 you won't bother quantum computing. If you're doing like arithmetic 449 00:28:53,726 --> 00:28:56,642 or deterministic algebra equation solution, 450 00:28:57,639 --> 00:28:59,142 you're better off with the classical. 451 00:29:02,555 --> 00:29:05,937 Interesting. What's the big 452 00:29:05,937 --> 00:29:09,674 challenge to getting funding, right? 453 00:29:09,820 --> 00:29:13,646 For not just for a startup, but for within a bank. I would imagine the 454 00:29:13,711 --> 00:29:17,393 big players in the finance world are already invested in quantum because they 455 00:29:17,393 --> 00:29:20,805 see the writing on the wall. But there's probably a lot of smaller 456 00:29:20,934 --> 00:29:24,605 players that are— they're still trying to ingest AI and figure out what AI means 457 00:29:24,621 --> 00:29:26,933 for them. What does that pitch look like roughly? 458 00:29:29,250 --> 00:29:33,083 Oh, to be honest, I wouldn't pitch a 459 00:29:33,228 --> 00:29:37,029 thing that promised a lot from quantum because the first thing you 460 00:29:37,158 --> 00:29:40,419 get in answer is that, why not AI? 461 00:29:41,717 --> 00:29:45,053 Even the problem you're saying is already solved by classical computing 462 00:29:45,391 --> 00:29:49,019 and you can see it's working. So why bother with quantum? 463 00:29:49,581 --> 00:29:53,162 Which I don't necessarily agree with, but answering to those questions 464 00:29:54,157 --> 00:29:57,722 require your audience to have some technical 465 00:29:58,750 --> 00:30:01,351 knowledge that you cannot create within a pitch, 466 00:30:01,688 --> 00:30:05,333 right? So my strategy is not 467 00:30:05,622 --> 00:30:09,371 getting to that conversation from the start. And I 468 00:30:09,371 --> 00:30:12,612 have seen this strategy from many other people. 469 00:30:13,582 --> 00:30:16,808 For now, at least, we can think about 470 00:30:18,244 --> 00:30:22,051 quantum processing unit, QPU, as something like 471 00:30:22,051 --> 00:30:25,492 GPU. So you resort to it when it is 472 00:30:25,525 --> 00:30:27,901 needed. And when, as I said, when your classical 473 00:30:28,742 --> 00:30:32,477 computation was having some problem that 474 00:30:32,541 --> 00:30:36,335 you could say it's good for quantum resort. 475 00:30:36,817 --> 00:30:39,741 The same thing in pitches, right? So if I go and say, hey, 476 00:30:40,656 --> 00:30:43,500 I'm building this company that can give you, 477 00:30:44,593 --> 00:30:47,549 I don't know, an algorithm that does 478 00:30:48,304 --> 00:30:51,694 arithmetic problems, not arithmetic. So I'm giving you a concrete 479 00:30:51,694 --> 00:30:54,602 example. I can solve the Black-Scholes model 480 00:30:55,437 --> 00:30:59,149 with quantum and with classical. So if I go build a 481 00:30:59,149 --> 00:31:02,730 startup saying that, hey, I do it classically, 482 00:31:02,923 --> 00:31:06,407 but then I have a QPU involved that solves it even 483 00:31:06,680 --> 00:31:10,310 better in cases that classical gets 484 00:31:10,310 --> 00:31:14,083 stuck. They buy it much more than if 485 00:31:14,083 --> 00:31:17,712 I say I'm going to have a startup that have a 486 00:31:17,777 --> 00:31:20,731 product that solves the Black-Scholes model 487 00:31:21,550 --> 00:31:25,356 with a quantum approach from scratch. That would be 488 00:31:25,404 --> 00:31:29,254 super hard to justify, honestly. Because as I say, we will say, hey, yeah, 489 00:31:29,334 --> 00:31:33,084 we can already solve it with classical. And, but by the way, 490 00:31:33,100 --> 00:31:36,866 Black-Scholes model is for deriving the price of 491 00:31:36,866 --> 00:31:39,574 security. So it's a very common popular 492 00:31:40,135 --> 00:31:43,644 problems that almost everyone knows in finance. 493 00:31:43,756 --> 00:31:47,554 But yeah, I guess again, the summary is I 494 00:31:47,586 --> 00:31:50,774 would start with quantum as an 495 00:31:50,790 --> 00:31:54,508 additional computational unit rather than from 496 00:31:54,588 --> 00:31:56,895 scratch quantum. Again, that breaks my heart. 497 00:31:58,337 --> 00:32:01,702 That's where we are now. So during a financial 498 00:32:01,702 --> 00:32:05,355 crisis, institutions need to understand their exposure very 499 00:32:05,467 --> 00:32:08,831 quickly. Could quantum computing eventually help 500 00:32:08,976 --> 00:32:12,805 financial organizations respond faster to rapidly 501 00:32:12,821 --> 00:32:16,618 changing markets? Definitely. And I'm glad that you asked 502 00:32:16,666 --> 00:32:20,479 this because that was one of the conclusions of my 503 00:32:20,591 --> 00:32:23,523 dissertation. One thing that 504 00:32:24,917 --> 00:32:28,749 I concluded was the— what I called it the agility 505 00:32:29,326 --> 00:32:32,980 of the algorithm, meaning that, so when you think 506 00:32:33,029 --> 00:32:36,860 about the volatility, volatility, what is volatility? It 507 00:32:36,860 --> 00:32:40,450 is the changes in the price of an 508 00:32:40,546 --> 00:32:44,313 asset, right? And then when you look 509 00:32:44,313 --> 00:32:47,535 at the volatility as a time series, like how 510 00:32:47,696 --> 00:32:51,463 volatility changes for a certain asset, for a certain stock 511 00:32:52,330 --> 00:32:56,119 over the time, right? You can see variations, right? 512 00:32:56,761 --> 00:33:00,597 And then you are trying to forecast, meaning that you 513 00:33:00,597 --> 00:33:04,353 are looking at the past data and then try to forecast the coming 514 00:33:04,947 --> 00:33:08,447 day's volatility, right? I use the same algorithm 515 00:33:09,362 --> 00:33:13,118 solved with classical and quantum. Everyone knows in finance, 516 00:33:13,150 --> 00:33:16,906 again, GARCH, G-A-R-C-H. So you can implement 517 00:33:16,906 --> 00:33:20,563 GARCH with classical computing. And you can say it 518 00:33:20,579 --> 00:33:24,312 has a certain accuracy, it has a certain metrics 519 00:33:24,424 --> 00:33:28,206 when it's solved with classical, and probably you can hit that 520 00:33:28,206 --> 00:33:30,930 accuracy better or worse with the quantum. 521 00:33:31,796 --> 00:33:35,321 But something that is for sure is your quantum 522 00:33:35,321 --> 00:33:39,119 algorithm reacts to the fluctuation of the 523 00:33:39,183 --> 00:33:42,276 price, of fluctuation of the seen volatility, 524 00:33:42,949 --> 00:33:46,285 much more faster than classical. Meaning that 525 00:33:47,281 --> 00:33:50,863 I, my quantum algorithm responds to those 526 00:33:50,959 --> 00:33:54,637 fluctuations faster than classical. I actually have 527 00:33:54,653 --> 00:33:58,059 some graphs and results, and I'm not sure if 528 00:33:58,589 --> 00:34:01,962 this podcast would have any visual components, but— 529 00:34:02,620 --> 00:34:06,266 Well, we could share it and we could just give people a 530 00:34:06,298 --> 00:34:10,137 link to the YouTube version. We do that. Okay. Yeah, that's a good 531 00:34:10,153 --> 00:34:13,605 idea. So you will see how, so 532 00:34:13,605 --> 00:34:17,395 imagine that you have these fluctuations and then your classical 533 00:34:17,395 --> 00:34:21,105 GARCH is more like averaging out those fluctuations 534 00:34:21,121 --> 00:34:24,671 because it doesn't have that speed to capture everything. 535 00:34:25,265 --> 00:34:28,798 On the other hand, quantum is just almost 536 00:34:28,798 --> 00:34:32,604 exactly fluctuating, your forecast fluctuating with 537 00:34:33,038 --> 00:34:36,651 the realized volatility, which something that I didn't 538 00:34:36,651 --> 00:34:40,357 expect honestly. And then I had to do it too 539 00:34:40,357 --> 00:34:44,124 many times to be sure that this is not a noise artifact or 540 00:34:44,124 --> 00:34:47,683 whatever, but then I realized no, this is actually a real thing. 541 00:34:48,196 --> 00:34:51,546 It's more agile in capturing it, and also 542 00:34:52,765 --> 00:34:55,730 because you are measuring the circuit at the end, you can measure it with 543 00:34:56,468 --> 00:35:00,043 different ways. I managed to incorporate the regimes 544 00:35:00,379 --> 00:35:03,233 into measuring, so you can say hey, 545 00:35:04,645 --> 00:35:08,337 I'm an expert. I believe that market is in bullish 546 00:35:08,401 --> 00:35:11,772 regime now. With that context, what is the 547 00:35:11,772 --> 00:35:15,576 forecast of volatility? The same market may be analyzed by 548 00:35:15,640 --> 00:35:19,124 another expert as not being bullish. It could be bearish in their 549 00:35:19,172 --> 00:35:22,816 opinion. Then you can run the same quantum 550 00:35:22,848 --> 00:35:26,492 algorithm on the same time series and 551 00:35:26,797 --> 00:35:30,457 you get a different forecast with the context of 552 00:35:30,649 --> 00:35:34,283 bearish. So that's another— context still matters. Oh, 553 00:35:34,348 --> 00:35:37,809 yeah. Yeah. That's another thing you cannot get from 554 00:35:37,906 --> 00:35:41,693 classical large. So, yeah, because we are doing measurement and 555 00:35:41,888 --> 00:35:45,423 we are doing— we are dealing with a natively 556 00:35:46,023 --> 00:35:49,342 probabilistic computational paradigm. 557 00:35:50,003 --> 00:35:52,409 Right. Interesting. 558 00:35:54,346 --> 00:35:58,058 It's giving a long-term weather forecast. Sorry, Ken. It's okay. No, it's okay. When, like, 559 00:35:58,123 --> 00:36:00,208 hey, this is what it's like. This is what the winter is going to look 560 00:36:00,322 --> 00:36:03,091 like in a non-El Niño year. This is what it's going to look like in 561 00:36:03,091 --> 00:36:05,985 an El Niño year. This is what it could look like in a super El 562 00:36:06,034 --> 00:36:09,876 Niño year, which is exactly right now. Yeah, like a scenario, 563 00:36:09,892 --> 00:36:12,937 given like scenarios, and then what happens 564 00:36:12,970 --> 00:36:16,774 based on that scenario. Yeah. How is it different than Monte Carlo 565 00:36:16,774 --> 00:36:19,772 simulations, or it's not the same but similar? 566 00:36:21,549 --> 00:36:25,321 What, the forecasting of volatility? Yeah, based on that. Based 567 00:36:25,321 --> 00:36:28,214 on context, is that— When you do 568 00:36:29,119 --> 00:36:31,844 volatility forecasting, you have to 569 00:36:32,668 --> 00:36:36,122 include some feature of 570 00:36:36,155 --> 00:36:39,867 volatility, like volatility clustering or 571 00:36:39,932 --> 00:36:42,955 memory to volatility because of the market behavior. So imagine 572 00:36:43,829 --> 00:36:47,386 the volatility, like the volatility memory of market. What does it 573 00:36:47,467 --> 00:36:50,735 mean is if you have a volatile day 574 00:36:50,903 --> 00:36:54,714 today, your tomorrow would 575 00:36:54,746 --> 00:36:57,743 be much more likely volatile 576 00:36:58,416 --> 00:37:01,685 than if you had a calm today, right? So it's kind of 577 00:37:02,438 --> 00:37:05,979 market has a memory to volatility for a while, right? 578 00:37:06,460 --> 00:37:10,258 If you have volatile today, then probably 579 00:37:10,290 --> 00:37:13,992 coming next days would be volatile as well, right? That, 580 00:37:14,040 --> 00:37:17,725 that, and as a matter of fact, the GARCH model has a term for it. 581 00:37:18,600 --> 00:37:22,446 In Monte Carlo, what you do is you are picking 582 00:37:22,622 --> 00:37:26,299 the points, sampling the points based on 583 00:37:26,396 --> 00:37:29,768 a uniform distribution. You don't have any— what is the 584 00:37:29,768 --> 00:37:33,477 technical term for it? So you don't have any assumptions. You 585 00:37:33,477 --> 00:37:36,576 would say that's the same likelihood if 586 00:37:37,170 --> 00:37:40,430 today volatility is the same 587 00:37:41,008 --> 00:37:44,733 as the next day volatility, right? So you are not 588 00:37:44,733 --> 00:37:48,057 assuming anything different. You're just taking randomly taking 589 00:37:48,249 --> 00:37:52,017 points. While in GARCH model, and 590 00:37:52,017 --> 00:37:55,590 basically in any statistical model, you have baked in 591 00:37:55,719 --> 00:37:59,389 some assumptions, some, some things that you have 592 00:37:59,389 --> 00:38:03,123 seen in macroeconomics or microeconomics, you have 593 00:38:03,155 --> 00:38:06,969 baked them into your model. And then what you 594 00:38:06,985 --> 00:38:10,462 want to do is trying to find the parameters of model. 595 00:38:10,847 --> 00:38:14,437 In other words, you already know what is the mathematical 596 00:38:14,437 --> 00:38:18,043 expression of the volatility. You just don't know 597 00:38:18,059 --> 00:38:21,669 what are the coefficients. To put— to plug into that 598 00:38:21,669 --> 00:38:25,492 mathematical expression. In Monte Carlo, on the other hand, 599 00:38:26,231 --> 00:38:29,748 you are looking for the PDF, probability distribution 600 00:38:29,845 --> 00:38:33,636 function, of a problem, right? You don't have any 601 00:38:33,636 --> 00:38:37,314 mathematical expression. You just want to see what 602 00:38:37,314 --> 00:38:41,024 that curve looks like, and then what is the mean of it? What is the 603 00:38:41,024 --> 00:38:44,414 variance like? Statistically speaking, it's pure 604 00:38:44,574 --> 00:38:48,139 statistic. Rather than mathematical 605 00:38:48,171 --> 00:38:51,704 modeling with unknown coefficients, like different 606 00:38:51,704 --> 00:38:54,819 degrees of uncertainty. I mean, you are totally 607 00:38:54,899 --> 00:38:58,608 uncertain about some process, you do Monte Carlo. 608 00:38:59,234 --> 00:39:02,686 When you think about it, but you don't have complete 609 00:39:02,686 --> 00:39:05,689 understanding, then you try to have a 610 00:39:06,219 --> 00:39:08,836 mathematical model, but with some uncertain 611 00:39:09,527 --> 00:39:13,172 components in it. And then try to find those components 612 00:39:13,480 --> 00:39:17,249 with whatever things that you do. It could be optimization, 613 00:39:17,282 --> 00:39:21,018 could be Monte Carlo, could be anything. I hope 614 00:39:21,018 --> 00:39:24,419 I did a good job. I think so. I think I understand. I think I 615 00:39:24,419 --> 00:39:27,498 understand. Yeah. Awesome. 616 00:39:29,179 --> 00:39:32,817 If you could challenge one widely held assumption 617 00:39:32,817 --> 00:39:36,590 about quantum computing in science— I'm sorry, 618 00:39:36,687 --> 00:39:38,956 in finance. What would it be? 619 00:39:42,000 --> 00:39:45,702 You cannot tell. I'm just putting myself into a lot of risk now 620 00:39:45,831 --> 00:39:48,305 because I know this subject has 621 00:39:49,857 --> 00:39:52,813 too many lovers and haters. I would say you cannot tell 622 00:39:54,781 --> 00:39:57,847 in an isolated case if 623 00:39:58,266 --> 00:40:02,110 quantum optimization has or has not an advantage, 624 00:40:03,115 --> 00:40:06,723 because I have seen people dividing on this. People think that 625 00:40:06,723 --> 00:40:10,540 quantum optimization has absolute advantage in finance. 626 00:40:11,614 --> 00:40:15,383 And some people say, no, it absolutely doesn't have any advantage or proven 627 00:40:15,383 --> 00:40:18,991 advantage in finance. I would say it is not 628 00:40:18,991 --> 00:40:21,877 true in either case. You should really look at your problem. 629 00:40:22,824 --> 00:40:25,983 You look at your mathematical properties of the 630 00:40:26,592 --> 00:40:30,313 optimization space you try to solve, and then start 631 00:40:30,313 --> 00:40:34,117 imagining if it has or has not. More specifically, 632 00:40:34,198 --> 00:40:37,839 I have seen, again, I don't want to name the names, but I talked with 633 00:40:37,839 --> 00:40:41,202 a gentleman. He's a very good scholar and in 634 00:40:41,299 --> 00:40:45,127 industry, in academia. He believed that 635 00:40:45,821 --> 00:40:49,514 quantum annealing used for optimization 636 00:40:51,072 --> 00:40:54,350 is a scam. Basically, that was what he said. Really? Yeah. 637 00:40:55,249 --> 00:40:59,062 And he says anything that quantum annealing can do in optimization, We 638 00:40:59,062 --> 00:41:01,579 can do it with imaginary time 639 00:41:02,974 --> 00:41:06,789 transformation of whatever. People 640 00:41:06,837 --> 00:41:10,637 are tough in that. And then we, of course, on the other hand, 641 00:41:10,669 --> 00:41:14,388 we have people that say, hey, quantum approaches, whether it be 642 00:41:14,612 --> 00:41:17,161 annealing, whether it be gate-based, I don't know, 643 00:41:18,540 --> 00:41:22,307 amplitude estimation methods are silver bullet 644 00:41:22,307 --> 00:41:26,019 to any optimization problem. So I stay in 645 00:41:26,067 --> 00:41:29,845 middle and I guess I'm getting punches from both sides. 646 00:41:31,853 --> 00:41:35,636 Interesting. No, I think 647 00:41:35,636 --> 00:41:39,388 there's an interesting use case there because there were an interesting story there, 648 00:41:39,420 --> 00:41:43,006 not use case, where a lot of people hate on a particular company whose their 649 00:41:43,071 --> 00:41:46,672 sole job is, I'm not going to name names, but their sole, 650 00:41:46,867 --> 00:41:50,406 their sole claim to fame is quantum annealing. And they get a lot of hate 651 00:41:50,471 --> 00:41:54,304 in a lot of the Reddit forums and various places on the internet. And I 652 00:41:54,304 --> 00:41:58,088 never understood it. But I guess if people don't believe 653 00:41:58,088 --> 00:42:01,581 in the process of annealing, yeah, natural. Plus 654 00:42:02,814 --> 00:42:05,650 for now they're one of the top names in the field as a company. So 655 00:42:05,650 --> 00:42:08,165 it's like the top guy always gets made fun of. 656 00:42:09,880 --> 00:42:13,404 Yeah. As a matter of fact, I got into that conversation in some of the 657 00:42:13,404 --> 00:42:17,153 conferences specifically for quantum annealing. And 658 00:42:17,153 --> 00:42:20,758 then that person happens to be there and he absolutely hated 659 00:42:20,758 --> 00:42:24,170 quantum annealing. And I was like, so what are you doing here? 660 00:42:24,923 --> 00:42:28,280 But then he was Again, I know he is a big 661 00:42:28,280 --> 00:42:31,707 scientist, but he was like saying, yeah, this is 662 00:42:31,965 --> 00:42:35,569 technically, this is a scam. And we talked like for a few hours. 663 00:42:36,670 --> 00:42:40,328 I couldn't convince him. He couldn't convince me. But that, that's the mentality 664 00:42:40,442 --> 00:42:43,093 of some people. Again, I cannot be sure if it's right or wrong. 665 00:42:45,807 --> 00:42:49,456 Curious. What a strange little factoid, because I've heard that 666 00:42:49,489 --> 00:42:53,052 too through completely different contexts. where 667 00:42:53,360 --> 00:42:56,978 there's a lot of disbelief in quantum annealing, yet it's 668 00:42:56,978 --> 00:42:59,298 proven commercially at least to be one of the, 669 00:43:00,832 --> 00:43:03,994 one of the more popular options. Whether it's real or not 670 00:43:05,269 --> 00:43:08,883 might be up for debate. That's fascinating. Yeah, 671 00:43:08,980 --> 00:43:12,495 again, as I said, for some problems like— what is the name— 672 00:43:13,171 --> 00:43:16,748 barren plateaus. If you have a lot of barren 673 00:43:16,877 --> 00:43:20,449 plateaus in your optimization space, Quantum 674 00:43:20,449 --> 00:43:24,139 annealing may help you, may help. There are other factors to count in. 675 00:43:24,754 --> 00:43:28,266 So you cannot say it never helps. You cannot say it always helps. 676 00:43:28,890 --> 00:43:32,551 But then people, I guess they are not, I don't know what they 677 00:43:33,018 --> 00:43:36,792 think. They looking for a smaller version of 678 00:43:36,809 --> 00:43:40,453 justification. And in the sense that they 679 00:43:40,453 --> 00:43:43,322 just lose track of other factors that may involve. 680 00:43:44,321 --> 00:43:47,013 But yeah, definitely you're right. Annealing is one of the more 681 00:43:47,856 --> 00:43:51,460 arguable quantum methods. And it's not just 682 00:43:51,460 --> 00:43:55,292 annealing. In gate-based models, we have other algorithms that— I'm 683 00:43:55,373 --> 00:43:58,676 giving an example. TensorFlow— sorry, tensor networks 684 00:43:59,542 --> 00:44:03,246 are not quantum at all, right? They are 685 00:44:03,246 --> 00:44:06,998 based on GPUs. But then there are people saying that whatever 686 00:44:06,998 --> 00:44:10,446 you do in certain quantum algorithms, you can do with tensor 687 00:44:10,927 --> 00:44:14,679 networks. So that bunch of quantum gate-based 688 00:44:15,208 --> 00:44:18,741 Algorithms are not necessary, and 689 00:44:18,838 --> 00:44:22,291 then the same thing happens there. Interesting, fascinating. 690 00:44:23,340 --> 00:44:23,453 Yeah, 691 00:44:27,093 --> 00:44:30,151 great questions. I put myself in a lot of danger here. 692 00:44:32,998 --> 00:44:36,427 You have to go to the Quantum Witness Protection Program. Oh yeah, like Saul 693 00:44:36,556 --> 00:44:40,404 Goodman, right? He's in Nebraska, right? Yeah, yeah, exactly. Yeah, 694 00:44:40,500 --> 00:44:44,290 so we have the history of being. and protection problem. People 695 00:44:44,857 --> 00:44:48,698 trying to throw tomatoes at me if they see me. 696 00:44:50,510 --> 00:44:53,957 Maybe they'll be quantum tomatoes. They'll either burst down into nothing or 697 00:44:54,281 --> 00:44:55,304 burst down into tomatoes. Yeah. 698 00:44:59,777 --> 00:45:03,418 I'm gonna hush now so Candace can ask a question. No, I'm— I've asked. I've— 699 00:45:03,418 --> 00:45:06,864 I'm filled with interesting information. I think this has been 700 00:45:07,383 --> 00:45:10,690 absolutely fascinating. I'm really happy that we got to dig into 701 00:45:10,738 --> 00:45:14,528 finance now. I think it's really great. Yeah, it's often 702 00:45:14,592 --> 00:45:17,771 cited as one of the top use cases for quantum 703 00:45:17,884 --> 00:45:20,742 finance, pharma, and a few other things. 704 00:45:21,480 --> 00:45:25,270 Logistics and recently aviation. And then there 705 00:45:25,270 --> 00:45:28,770 is a defense which I never touched, but I heard about that. 706 00:45:29,428 --> 00:45:32,383 But yeah, we probably want to have a second episode on 707 00:45:33,041 --> 00:45:36,860 quantum finance. Sure. I'm just joking, but yeah, if you're— We'd love 708 00:45:36,860 --> 00:45:39,847 to have you back. This is a fun conversation. We'd love to have you back. 709 00:45:40,290 --> 00:45:43,518 And in there, I can tell what is quantum finance, which is 710 00:45:44,518 --> 00:45:47,875 a little bit different than quantum algorithm in finance, 711 00:45:48,779 --> 00:45:52,009 right? There's a thing as quantum finance. Interesting. 712 00:45:53,180 --> 00:45:55,228 Where can folks find out more about you? 713 00:45:58,262 --> 00:46:01,663 Oh, didn't get the question. What do you mean? Oh, I'm sorry. Where can folks 714 00:46:01,663 --> 00:46:03,589 find out more about you and what you do? your own? 715 00:46:07,862 --> 00:46:11,139 Again, I didn't get it. So you're asking what is my work look like? 716 00:46:11,672 --> 00:46:15,499 Where can we— where can people find out more about you and— Oh, sorry, I 717 00:46:15,870 --> 00:46:19,688 was hearing work. Oh, it's okay. My LinkedIn, I post once in 718 00:46:19,688 --> 00:46:23,392 a while in my LinkedIn about the things that I can 719 00:46:23,392 --> 00:46:26,507 publish publicly. Like recently I 720 00:46:26,719 --> 00:46:30,447 published a few things. One was the 721 00:46:30,512 --> 00:46:33,116 contest I won in INQ 722 00:46:33,908 --> 00:46:35,817 competition, and it was like for 723 00:46:37,370 --> 00:46:40,662 predicting— or not predicting, it was for 724 00:46:40,855 --> 00:46:44,357 identifying the current market regime based on a 725 00:46:44,422 --> 00:46:47,813 library of regimes that we have. So imagine if I ask you, or you ask 726 00:46:47,813 --> 00:46:50,434 me, what regime we are living now 727 00:46:51,954 --> 00:46:55,464 for a market of whatever stock, I would 728 00:46:55,480 --> 00:46:59,241 say Based on my quantum algorithm that runs on IonQ, 729 00:46:59,851 --> 00:47:03,011 I would say 20%, we are in bearish. 730 00:47:04,038 --> 00:47:07,679 I don't know, 15%, we are in bullish. So there are probability 731 00:47:07,968 --> 00:47:11,177 assigned with market regimes and 732 00:47:11,802 --> 00:47:15,556 I do it in a quantum way. Sure. Yeah, I published the 733 00:47:15,556 --> 00:47:19,342 code. I published the working panel so that there's a 734 00:47:19,422 --> 00:47:22,791 link in my LinkedIn. You can go click on that. It gets to the Colab. 735 00:47:23,372 --> 00:47:27,062 execution panel, you can just select 736 00:47:27,351 --> 00:47:30,257 from the few options that I put there from the stocks, 737 00:47:31,140 --> 00:47:34,641 and then select it, and then it gives you the probability 738 00:47:35,395 --> 00:47:39,233 of what regimes they are in as, as today. And 739 00:47:39,233 --> 00:47:42,942 also it gives you a few quantum things that, you know, what, what 740 00:47:42,942 --> 00:47:46,266 circuit has been used, what are the parameters used in the circuit for 741 00:47:46,940 --> 00:47:50,360 Montenegro people. Interesting. Yeah, sure. 742 00:47:50,489 --> 00:47:53,989 Another thing that I, again, published in my LinkedIn is 743 00:47:54,664 --> 00:47:58,277 an interactive reading experiment. So 744 00:47:58,277 --> 00:48:01,745 imagine you have these technical paper or 745 00:48:01,745 --> 00:48:05,359 survey paper that has a lot of 746 00:48:05,359 --> 00:48:09,084 material, including mathematics, graphs, tables, 747 00:48:09,277 --> 00:48:12,986 along with the text. And then a 748 00:48:13,115 --> 00:48:16,872 lot of papers referred or cited as well. When I was 749 00:48:16,969 --> 00:48:20,373 a student, my main problem was like, I read the first paragraph, 750 00:48:20,999 --> 00:48:24,708 then I have to think about it, then I have to go chase the referred 751 00:48:25,173 --> 00:48:28,561 citation, and then that citation happened to be a 752 00:48:28,561 --> 00:48:31,467 10-page paper by itself. Then I have to chase 753 00:48:32,543 --> 00:48:36,188 where in that paper the original paper means such and 754 00:48:36,188 --> 00:48:39,463 such. It was a lot of effort also for 755 00:48:39,463 --> 00:48:43,043 mathematics or graphs or whatever. Many papers, 756 00:48:43,075 --> 00:48:45,979 because they have a limitation of a space, they don't give you the full 757 00:48:46,524 --> 00:48:49,909 steps that led to that graph or led to that 758 00:48:50,278 --> 00:48:54,032 equation. Then I was always wondering how they derive this equation, 759 00:48:54,128 --> 00:48:57,930 how they come up with this. So I solved all these problems 760 00:48:58,075 --> 00:49:01,540 in my platforms. When you publish a 761 00:49:01,604 --> 00:49:05,085 paper in my platform, it goes and chases and 762 00:49:05,198 --> 00:49:09,047 reads all the citations, gets an idea, of what 763 00:49:09,047 --> 00:49:12,428 is the paper corpus underlying your paper. 764 00:49:13,133 --> 00:49:16,771 It visually sees the graphs and mathematics. It is just 765 00:49:16,771 --> 00:49:20,617 not reads the text. So you can ask it to 766 00:49:21,466 --> 00:49:25,216 specific questions about the graph. What does this point mean? What does 767 00:49:25,312 --> 00:49:29,110 this trend mean? Can we support this graph with 768 00:49:29,270 --> 00:49:32,811 the literature? If anyone else has come up close with this 769 00:49:33,420 --> 00:49:36,578 result? Or for the mathematics, you can ask 770 00:49:37,187 --> 00:49:40,923 simply derive this mathematics, like it's just reverse engineering 771 00:49:41,035 --> 00:49:44,850 the equations that you have. So it gives you, if it can 772 00:49:45,123 --> 00:49:48,858 find it in the references, it gives you a step-by-step derivation 773 00:49:48,954 --> 00:49:52,786 of the math that is given there. Again, I published it, one 774 00:49:52,802 --> 00:49:56,569 of my papers, and I also put the link there. I reached 775 00:49:56,569 --> 00:49:59,487 out to people encouraging them to 776 00:50:00,417 --> 00:50:04,093 convert their already written PDF papers into 777 00:50:04,270 --> 00:50:07,753 this interactive mode. I have a few interested people, but honestly, 778 00:50:07,897 --> 00:50:11,717 not as much as I expected. But I'm still trying 779 00:50:11,749 --> 00:50:15,569 to, to basically encourage the idea that reader should 780 00:50:15,617 --> 00:50:19,133 not stop at the point that author stopped in the paper, right? 781 00:50:19,919 --> 00:50:23,258 I may have my own conclusion at the end of the paper, not 782 00:50:23,338 --> 00:50:27,142 necessarily the reader agrees with, right? So the reader 783 00:50:27,142 --> 00:50:30,658 can investigate my paper, come up with different 784 00:50:30,690 --> 00:50:34,223 ideas, or even counterpropose some of the 785 00:50:34,223 --> 00:50:37,387 results that I claim. And that platform even 786 00:50:38,061 --> 00:50:41,097 makes them able to do that. They can counterpropose, they can 787 00:50:41,113 --> 00:50:43,794 contribute, they can add to that paper. 788 00:50:44,983 --> 00:50:48,227 And if it's legit, platform itself calls the 789 00:50:48,660 --> 00:50:52,450 original author and lets them know that, hey, we have this, 790 00:50:52,514 --> 00:50:56,308 it seems to be interesting, you can collaborate. Things like 791 00:50:56,308 --> 00:51:00,098 that. Interesting. Yeah, I can give you the 792 00:51:00,162 --> 00:51:03,646 link for the audiences to take a look at it. 793 00:51:05,483 --> 00:51:07,917 That's great. Yeah, definitely. Yeah, 794 00:51:09,593 --> 00:51:13,160 excellent. Any parting thoughts? Quantum finance, 795 00:51:13,225 --> 00:51:16,992 at first it seemed a little scary, but honestly, the way you've been 796 00:51:16,992 --> 00:51:20,100 explaining it, it's— again, I'm just completely 797 00:51:20,132 --> 00:51:22,876 fascinated by it. I think that this is really interesting. 798 00:51:24,233 --> 00:51:26,778 It's not going to tell you how to predict the stock market. It's not about 799 00:51:26,778 --> 00:51:30,295 that. Getting rid of those silly assumptions, but I really 800 00:51:30,311 --> 00:51:33,956 appreciated hearing everything. Thank you so much. Yeah, sure. Yeah. 801 00:51:34,004 --> 00:51:37,633 Thank you for having me again. It's a great pleasure to be able 802 00:51:37,633 --> 00:51:40,812 to talk about things that, as you said, 803 00:51:41,550 --> 00:51:45,019 people think it's scary, but it's not really scary. As soon as you get the 804 00:51:45,019 --> 00:51:47,507 concept, what is it principally about? 805 00:51:48,970 --> 00:51:52,229 you get comfortable with. Yeah, absolutely. 806 00:51:53,068 --> 00:51:55,166 With that, we'll roll the outro.