Challenge Briefing and Live Q&A — Transcript
Full transcript
- 0:02So, there's a bit of a miss messages,
- 0:04but um yeah, I you saw my
- 0:08>> Yeah. Yeah. Okay. Don't worry, Jeff.
- 0:09We'll talk about it later. It's good.
- 0:11>> Good. Good. Thank you. Thank you very
- 0:13much.
- 0:15>> Of course.
- 0:26Just waiting for more people to join.
- 1:30Hi Verdon. Good morning. Good afternoon.
- 1:33Thanks everyone for joining.
- 1:38>> Hello.
- 1:41>> Good morning.
- 1:44I think we should wait for a couple of
- 1:45minutes for more people to join. I I got
- 1:47rece I keep receiving emails at like
- 1:494:00 a.m. my time that people were
- 1:51already waiting to join. So I see the
- 1:53enthusiasm which is really exciting.
- 1:58It's
- 2:16maybe we can start in about
- 2:19two
- 2:22two minutes.
- 3:07Yes, and thank you Eugene for raising
- 3:09the question. uh if everyone's looking
- 3:11to join a team still uh yes I think it's
- 3:15yeah if anyone wants to form teams
- 3:17outside of this call of course you're
- 3:18more than welcome to get uh to know each
- 3:21other and uh yeah so if anyone can wants
- 3:24to reach out to Eugene you're more than
- 3:26welcome
- 3:47Okay. Should we perhaps
- 3:50start in one minute? Is that okay for
- 3:52everyone?
- 3:55>> Yes, we're done.
- 3:56James. Okay.
- 3:57>> Yeah.
- 3:58>> Yeah. Fab Fabian um as well is on the
- 4:02call. I think actually he might want to
- 4:04start with a couple of slides as well.
- 4:08>> Oh, he would like to share some slides
- 4:10or
- 4:12>> Yeah. Is that Can we just check if it's
- 4:14possible?
- 4:14>> Yes, I I'm here. I don't know if you see
- 4:17my video actually.
- 4:19>> Uh
- 4:19>> hi James.
- 4:21>> Yeah, I can see you now.
- 4:24>> Hi. Hi Febian. Hi.
- 4:25>> Hi. Good morning.
- 4:27>> Good morning. Good morning. Um, let me
- 4:29see if I'm gonna stop sharing my slides
- 4:33and then uh
- 4:35um
- 4:40stop sharing. Uh do you can you share
- 4:42your from your side?
- 4:46>> Yes. I I will just quickly set up the
- 4:49the sharing of my the the presentation
- 4:53mode which always screws up my windows.
- 4:57So, but now I think I got it.
- 5:00So, let me just
- 5:07Okay.
- 5:10Very good.
- 5:16Yes. So, yes. So, uh I think do you want
- 5:19to do another introduction or should I
- 5:21just jump in
- 5:23>> from our side?
- 5:25>> Um you can jump in and then we can we
- 5:28can do the industry challenge later. But
- 5:30yeah, because we're done, we'll take
- 5:31over.
- 5:32>> Mhm.
- 5:32>> Perfect. So, yeah. So, I I'm I'm from my
- 5:35name is Fabian. I'm based out of
- 5:37Switzerland, Zurich. I'm I'm at AWS like
- 5:41in a team that's called advanced
- 5:42solutions lab. We're working with with
- 5:45customers on on quantum computation and
- 5:47I'm I'm basically here because also like
- 5:50AWS is kind of like like helping out
- 5:52here on this quantum challenge and with
- 5:54some credits because as you may know
- 5:57like Amazon bracket is our quantum
- 6:00service and I quickly just take a few
- 6:02slides to kind of show what we're doing
- 6:04at AWS in in the area of quantum
- 6:06computation.
- 6:13Yes. So we have basically three pillar.
- 6:15Main pillar of course is our service
- 6:17Amazon bracket as I already said where
- 6:18we our goal is to give access of quantum
- 6:21computers to a broad community as as as
- 6:25scientists as you that want to
- 6:27experiment with quantum computers. Then
- 6:30we have the AWS center for quantum
- 6:32computation which is our research
- 6:34facility where we do active research on
- 6:37quantum algorithms hardware and also
- 6:40networking and we have the Amazon
- 6:42advanced solutions lab that's where I'm
- 6:43in. We're working actually with
- 6:45enterprise customers uh in order to kind
- 6:48of help them to get started with quantum
- 6:50computation run research engagement with
- 6:52them and also build PC's out on Amazon
- 6:56bracket.
- 6:59So maybe first on Amazon bracket. Amazon
- 7:02bracket is part of AWS cloud. It's
- 7:04integrated in it. It provides you with
- 7:07several ways on connecting to it and to
- 7:09to leverage quantum computers. it uh you
- 7:12can you can work like from manage
- 7:14notebooks where you get the compute in
- 7:16on AWS uh in the AWS cloud through
- 7:20Amazon bracket or you can use your local
- 7:22ID for the for code development
- 7:26and then you can basically access like
- 7:29different simulators we have like state
- 7:31vector simulator tensor network
- 7:32simulator and densy matrix simulators uh
- 7:35simulators running on AWS cloud so we
- 7:38also have a simulator that runs directly
- 7:40on your laptop or on the notebooks if
- 7:42you uh like do that for for testing. But
- 7:46if you want to do it on a larger scale
- 7:48like uh this managed simulators actually
- 7:51help you to do that it's basically in
- 7:54the cloud starting like spinning up the
- 7:56compute that is required. You will do
- 7:58your computation and once the
- 8:00computation is done it you will uh also
- 8:03kind of depreciate all the kind of
- 8:05resources that has been used and we'll
- 8:07give you the the the results back. Of
- 8:10course that that will incur some costs
- 8:12on your side because these are managed
- 8:14compute. Then we have uh like like
- 8:17quantum hardware which we come into it
- 8:19as well. The quantum hardware usually
- 8:21don't live in our AWS uh like like uh
- 8:25data centers. They are usually accessed
- 8:28through API and standing at the
- 8:30different locations.
- 8:33So we have different quantum hardware
- 8:35that we we provide for you. is like the
- 8:38from IMQ with ion trap devices. We have
- 8:41like superc conducting with regetti and
- 8:43IMQ and we also have quer which is a
- 8:47like a neutral atom device which is a
- 8:50special purpose kind of Hamilton
- 8:52simulation uh QPU.
- 8:56So you will see like we have different
- 8:58numbers of cubits like for instance we
- 9:01also like the the INQ comes with kind of
- 9:03error mitigation techniques and then we
- 9:06have up to 108 cubits for
- 9:07superconducting with the newest rigetti.
- 9:14Good. And maybe highlighting one thing
- 9:16like we are actually approaching the
- 9:18area of kind of like uh quasi fall
- 9:22tolerant quantum computation
- 9:24which we also engaged in with our
- 9:26strategic partner quer which will not be
- 9:29at the first place like any accessed
- 9:32through the Amazon bracket will be kind
- 9:33of like tailor because these are kind of
- 9:35very advanced. We are we are planning to
- 9:38start off in 2028 with the first kind of
- 9:41quasi fall tolerant quantum comput
- 9:45uh computer computing device which gives
- 9:48you about 25
- 9:5156 cubits together with about 10 to the
- 9:54six uh operations that you can do in
- 9:56depth. So we are very excited actually
- 9:59on this journey and it also shows how
- 10:01how we progressing in that field.
- 10:07Good. So the Yeah. So not much to say on
- 10:10on on the AWS center for quantum
- 10:13computing. I mean their their uh
- 10:16location is is close to is it uh at
- 10:20close to Caltech and they also working
- 10:23on uh our own hardware.
- 10:27Maybe quickly to our team what we doing.
- 10:29So it's the Amazon advanced solutions
- 10:32lab. So we basically work with customers
- 10:34in order to search for uh quantum
- 10:37applications.
- 10:39As you know this these are and this is
- 10:41also part of this challenge. Finding
- 10:44industry relevant quantum application is
- 10:46still one of the holy grails and quantum
- 10:48computation. We are still working uh
- 10:50with custom on that especially bridging
- 10:52the gap between academia and and and the
- 10:55industries and for that we combine like
- 10:58different kind of like fields like
- 11:00quantum computing at the heart of it
- 11:03where we actually bring deep expertise
- 11:05and then but combining it with machine
- 11:07learning and optimization so we can
- 11:09benchmark the different approaches which
- 11:11is very important because like if you
- 11:14cannot show that it brings a quantum
- 11:16advantage against classical then then
- 11:19then the the the valid validity of
- 11:22results are is always questionable. So
- 11:24we we really want to combine these
- 11:26experties
- 11:27and I I I think like I'm stopping here
- 11:31and handing back to
- 11:33>> S. Yes. Thank you. Thank you so much
- 11:35Fab. Perfect. Thank you so much. Uh now
- 11:38I will share um our slides for this um
- 11:43challenge. So
- 11:46um
- 11:47yes, welcome everyone. Thank you so much
- 11:49for joining and thank you Fabin for your
- 11:51introduction. Um yeah so the agenda for
- 11:55today is more about uh getting to know
- 11:57what the project is about more in diving
- 12:00into the technical um implications of
- 12:03this deliverables and then of course for
- 12:05those who want to participate ask
- 12:07questions you're actually going to be
- 12:08asking questions for the team that's
- 12:10actually running this challenge. So just
- 12:12a brief hello um Fabian thank you yeah
- 12:15for your introduction. Now from my side
- 12:18um business development manager at Kai
- 12:20Ventures uh Verdon is a CTO and Tobias
- 12:24is our global head of technology and we
- 12:26also have James from ADB side on this
- 12:29call.
- 12:30Hello everyone and um James I will leave
- 12:35it to you to actually do an introduction
- 12:36on the challenge itself.
- 12:39>> Excellent. Uh thank you SA. So hi
- 12:41everyone, my name is James. I come from
- 12:43ADB uh from our digital innovation team.
- 12:46And this challenge specifically we're
- 12:48looking at something called a joint loss
- 12:50model or how we can replicate joint loss
- 12:52modeling using quantum systems. Uh very
- 12:56very quickly a joint loss model can be
- 12:58used to describe a set of uh
- 13:00interrelated or dependent events. Um,
- 13:04and often times when you design a
- 13:06traditional joint loss model, um, you're
- 13:08looking at, uh, what the likelihood of
- 13:10two or more related processes triggering
- 13:12simultaneously,
- 13:13um, might be. And specifically to us at
- 13:16at ADB, we're interested in this because
- 13:19uh these style of uh events,
- 13:22probabilistic events are very common
- 13:24across uh the problems that we look at
- 13:28uh that are related to um the financial
- 13:31services that we offer as a development
- 13:33bank. So you can think problems in the
- 13:36insurance industry uh such as insurance
- 13:38link securities or catastrophe bonds but
- 13:41also other models that we might look at
- 13:44such as bond market and looking at the
- 13:47secondary or tertiary uh spillover
- 13:50effects of certain bond market triggers
- 13:52that might happen. So there's a lot of
- 13:55macroeconomics um as well as a sort of
- 13:58smaller more traditional economic
- 14:00problems that we look at where we're
- 14:02trying to understand and trying to
- 14:03assess how quantum or quantum computing
- 14:06techniques could be leveraged in in this
- 14:08case. Specifically for this problem uh
- 14:11we are looking at a joint loss scenario
- 14:14for rainfall and for earthquake being
- 14:16triggered simultaneously across two
- 14:18countries in Carrick. And with this
- 14:22problem, because we're dealing with
- 14:23earthquakes and we're dealing with with
- 14:25rainfall events, there is the issue that
- 14:28we do have a very strong data
- 14:31constraint. We don't have a lot of
- 14:32historical data, maximum 30 to 50 years
- 14:35of high quality data. Um the
- 14:39dependencies are poorly constrained
- 14:41because of uh the geography, the style
- 14:43of geography between these two
- 14:45countries. And we're really looking at
- 14:47sort of very interesting uh novel
- 14:51perhaps very far out out of the box blue
- 14:54sky ideas where we could leverage new
- 14:57technologies. And in this case we're
- 14:59looking at quantum to improve the style
- 15:02and the way we produce these models in
- 15:04such a way that could potentially be
- 15:06reproduced across the bank and also
- 15:08deliver deliver us some form of quantum
- 15:11advantage. whether that be sort of
- 15:13improvements in efficiency or
- 15:15improvements in sampling or so on and so
- 15:17forth.
- 15:19So thank you. Thank you very much.
- 15:21>> Thank you. Yep. Now to Verdan on more
- 15:26technical side. Thank you James.
- 15:29>> Okay. Um yeah thanks a lot James for
- 15:31that uh quick introduction about what
- 15:33the challenge means. Uh so if you anyone
- 15:35of the participants today is coming from
- 15:37more quantum background or from a more
- 15:40technical computer science math
- 15:42background uh the introduction is good
- 15:44but you must be like wondering where is
- 15:46the math give me the equation and I'll
- 15:48you know model around it I'll build the
- 15:50solution around it. So that is where I
- 15:52come in. uh this entire problem that
- 15:54James described the joint loss modeling
- 15:56problem can be broadly described with
- 15:59the equation that is uh presented over
- 16:00here right and to take an example and uh
- 16:04mind you like today this would be a very
- 16:06quick overview of a lot of uh technical
- 16:08stuff but obviously as you proceed
- 16:10forward with a challenge and you want to
- 16:12work hands-on all the slides all the
- 16:14equations that are mentioned over here
- 16:15along with clear descriptions of what
- 16:17can be done what has already been done
- 16:19for you to benchmark and uh relate your
- 16:21resources with uh is available on the
- 16:24web page on the challenge description.
- 16:26So in fact the entire code repository
- 16:29has been shared by ADB. Uh it is
- 16:32available on uh the web page. Uh so yeah
- 16:35diving into the equation right what does
- 16:37this entire problem mean? Uh let's first
- 16:40consider an example problem. We see
- 16:42there are two countries. One is Kirius
- 16:45the other one is Tajak. And there are
- 16:48two kind of disasters that we are
- 16:50considering over here. So one is
- 16:51earthquakes and the other one is extreme
- 16:53rainfall.
- 16:55Now each of these uh countries has some
- 16:58kind of uh principle of three-year terms
- 17:01right of 80 million US which is split
- 17:05into 70 million for earthquakes and 10
- 17:08millions for rain. Now this equation
- 17:11over here kind of uh shapes this entire
- 17:13joint loss modeling into a mathematical
- 17:16format and and the way I look at it uh
- 17:19is going from inside out. So you start
- 17:21making sense of it. So let's maybe first
- 17:23fix uh the the x term over here that's
- 17:26the joint outcome. So a variable x kind
- 17:30of defines what kind of parallel unit
- 17:33over here uh within a certain period. In
- 17:36our example, that could be a coupled
- 17:38unit between a country and a disaster.
- 17:41So it could be something like kyus
- 17:43earthquake or rain in kyus. Uh similarly
- 17:47earthquake in Tajjak or rain in Tjak and
- 17:50and so on and so forth. Right? So for
- 17:53this example, those would be the set
- 17:55variables. But if you take up another
- 17:58joint loss modeling example or you take
- 17:59some different kind of data sets that
- 18:01are available publicly, uh those
- 18:03variables might change a little bit.
- 18:05That's kind of the essence over here.
- 18:07Now for each of these variables X uh
- 18:10when we take P of X that's basically the
- 18:12probability of that disaster happening
- 18:14or that peril actually occurring. Uh
- 18:18that probability will give you how
- 18:20frequent that outcome is for a given
- 18:23country. Ideally we would want that to
- 18:25be as low as possible. But from uh this
- 18:28principle point of view, we understand
- 18:30that this probability will help us
- 18:33decide how how we want to model our
- 18:36economics around that. Right? Very
- 18:39similar to the probability, we also have
- 18:42f of x over here which basically gives
- 18:44you a solid number for any kind of
- 18:47payouts that uh eventually turn into the
- 18:50quantity that we might care about.
- 18:52Right? So for example, it could be the
- 18:54total loss. It could be yeah just a yes
- 18:57and no answer for whether the countries
- 18:59are being paid for the disaster
- 19:01happening or something along those
- 19:02lines. Right?
- 19:04And um let's see what what else is
- 19:07there. And then we have the contract
- 19:08terms, right? So the big L the sum of CI
- 19:11is the total loss portfolio of that
- 19:13entire uh bond. Right? Now all of this
- 19:18jargon kind of starts making sense if
- 19:20you look at from a bird's eye point of
- 19:22view when we say that for each country
- 19:24we want to take uh different kind of
- 19:26probabilities of how frequently a
- 19:28disaster might be occurring and based on
- 19:31that we need to compute the loss models
- 19:34we need to compute the payout terms for
- 19:36each of these countries.
- 19:38uh a few more uh guidelines or a few
- 19:41more I would say rules for modeling this
- 19:44is number one uh that the earthquakes
- 19:48are actually paid in quarter steps so
- 19:50not annually they are accumulated four
- 19:53times a year up to the total sublims
- 19:58the precipitation that remains as an
- 20:00annual thing so all the data all these
- 20:02rules all these guidelines are again
- 20:04shared on the repository so you can have
- 20:07a look and make more sense of it. I know
- 20:10just talking about this might be very uh
- 20:13technically jargon for some of you and
- 20:14that's that's totally okay but this is
- 20:16just to give you all of you a reference
- 20:18point of what modeling this means like
- 20:21now once you have this entire model the
- 20:24good thing is that uh the benchmarks are
- 20:26already been provided by ADB so you can
- 20:28actually compute this small data set you
- 20:30can actually do the classical routine
- 20:32already and compare it with how ADB has
- 20:35done it or what the baseline benchmarks
- 20:38are and the interesting thing is that we
- 20:41want all of you to go beyond that. So
- 20:42not just replicating what is already out
- 20:44there but we all want to uh you know
- 20:47start building crazy ideas to James
- 20:49point where you want to uh have out of
- 20:52the box ideas from all of you. Uh we are
- 20:55sharing some uh approaches that we might
- 20:57feel are uh are suitable enough. There
- 21:00have been some past literature on some
- 21:02of these approaches using quantum
- 21:03amplitude amplification or uh say
- 21:07sampling methods or quantum machine
- 21:08learning for instance. But uh this is a
- 21:12challenge and you are more than welcome
- 21:14to use your own resources come up with
- 21:16some new approaches that you might want
- 21:18uh to try out over here and uh yeah get
- 21:22us solution right. Uh the end goal over
- 21:25here is kind of two-part. So number one,
- 21:27we want to see benchmarks against the
- 21:30classical solutions that we're already
- 21:31providing you with, right? So any kind
- 21:33of solution that you come up with, it
- 21:35has to have a onetoone benchmark of uh
- 21:38how it compares with the current
- 21:39solution. And number two, we want to see
- 21:42scalability, right? Because the data set
- 21:44that we're providing today and the
- 21:45approaches that we're providing today
- 21:46might work in a limited data set, but
- 21:49what happens when we are actually doing
- 21:51it at a production scale? uh what
- 21:53happens when we want to use quantum
- 21:55potentially for solving real world
- 21:57problems and not just you know toy
- 21:59challenges. So we do want to see a real
- 22:02validation of how scalable a solution is
- 22:05and to make that easier we are also
- 22:07providing credits through AWS. So Fabian
- 22:10shared a little bit on that. Uh we will
- 22:12be providing credits to select teams for
- 22:14scaling up their work right. So whether
- 22:16that means running it on GPU resources
- 22:18or running it on real quantum computer
- 22:21uh you let us know right uh so that is
- 22:24where kind of the challenge uh takes
- 22:26shape uh so if you move to the next
- 22:28slide uh sea
- 22:31>> yep so here are the deliverables and
- 22:32I'll pass it on to sa to just go through
- 22:35all of that y
- 22:37>> uh yeah sure um the registration is open
- 22:39as you know on the on the website um we
- 22:43have two phases the phase one is more
- 22:45about like the proposal if you can uh
- 22:47yeah work on uh plan of action what to
- 22:51expect and then of course the team will
- 22:52evaluate more deeply if you're um if
- 22:55you're okay to move to phase two and
- 22:56then once you actually been selected um
- 23:00on phase two uh you will be actually
- 23:01working on uh working in the prototype
- 23:04uh you have to provide a technical
- 23:05documentation analysis business case and
- 23:09then in the end the final report which
- 23:11then will be judged actually by a few
- 23:13judges um and I will walk you through
- 23:15the timeline uh later in the in this
- 23:18workshop. Uh just to be mindful of time,
- 23:21I think we need to speed up a little
- 23:22bit. So I know there are a few more
- 23:23slides. Verdon, maybe you can um walk
- 23:26through in terms of evaluation itself
- 23:29and then we want to have more time for
- 23:31any questions that people have.
- 23:34>> Great. uh so now we expect a lot of
- 23:36submissions to come our way and that is
- 23:38always a good problem to have but uh to
- 23:40make it very transparent and to make
- 23:42sure that there is a clear guidelines on
- 23:44what you might expect from a
- 23:45participant's perspective
- 23:47>> uh this is what the evaluation criteria
- 23:48looks like right so the first one would
- 23:50be problem and approach uh it should
- 23:53have a clear onetoone matching with our
- 23:55benchmark so that is it's it's you know
- 23:57easier for us to also evaluate u we are
- 24:00keeping the challenge pretty open but
- 24:02that does not mean that we are straying
- 24:04away from the challenge at hand, from
- 24:05the problem statement at hand, right?
- 24:07And then obviously uh how does your
- 24:10approach, how does the solution match
- 24:11with the problem? Uh we believe me, we
- 24:15we have heard many times a lot of people
- 24:17claiming that hey quantum can do x y and
- 24:19zed that's all good but uh how much of
- 24:22that is really scalable, right? Uh so
- 24:25getting that proof of concept ready with
- 24:27clear prototypes with clear results uh
- 24:30will be the second metric that we'll be
- 24:32judging this on. Uh then obviously comes
- 24:34the technical trigger where we will be
- 24:36you know diving deep and trying to break
- 24:38your code and and understanding what
- 24:40works what doesn't work. Uh so make sure
- 24:42that it's uh technically solid whatever
- 24:45you end up submitting. Now beyond that
- 24:48in the first phase of submission we also
- 24:51want to see proof and and validation of
- 24:53whether your team has what it takes to
- 24:57continue into the next phase of our
- 24:58entire challenge. So we need to
- 25:01understand what your work plan looks
- 25:02like right this could include a road map
- 25:04this could include I don't know like a
- 25:06GAN chart with clear milestones clear
- 25:09risks what you expect to happen good but
- 25:12also you should have a clear view of
- 25:14what uh might not work out right and
- 25:16what are some uh backup options that you
- 25:19might want to consider already uh so
- 25:22getting that work plan ready would be
- 25:23really helpful for us to evaluate you as
- 25:25a team moving into the next phase and
- 25:28similar to work plan your resource those
- 25:29estimations. So this ties back with my
- 25:32earlier point of saying that all of the
- 25:34teams or at least the selected teams
- 25:36would be getting hardware credits to
- 25:38scale up their resources, right? So how
- 25:41much hardware credits do you want?
- 25:42Right? Do you need GPUs? Do you need
- 25:43CPUs? If you want quantum processors,
- 25:45what kind of QPUs do you want? Do you
- 25:47want to run it on superconducting
- 25:49devices? Do you want to run it on uh
- 25:51some of the simulators available? Do you
- 25:52want to run it on neutral atom devices?
- 25:54Right? you you need to uh let us know
- 25:56with clear justification of what you
- 25:58want and why you want it and why would
- 26:02uh giving you this kind of resources
- 26:03make your solution successful so that
- 26:06our partners like AWS can kindly grant
- 26:08you those kind of credits and make the
- 26:10scalability possible
- 26:12>> perfect
- 26:12>> and uh finally your team skills right we
- 26:16don't have kind of any uh barrier over
- 26:19here anyone and everyone can participate
- 26:20but again we do want to look uh at your
- 26:23skills what you're bringing to the table
- 26:25to make sure that if you're
- 26:26participating, we can bet on you to be
- 26:29the best team possible.
- 26:31>> Thank you.
- 26:33And uh maybe James, do you want to
- 26:35finish off with the most important slide
- 26:38in terms of the timeline so people do it
- 26:40on time? Yeah, thank you.
- 26:42>> Yep. Thank you. Uh so we have now just
- 26:45less than a couple of weeks uh for
- 26:47registration submission. Registration is
- 26:49via the ADB challenges websites which I
- 26:51will share a post uh a link in the uh in
- 26:58the chat shortly. I know there's also
- 26:59been some questions around the GitHub
- 27:01page. We won't be sharing GitHub page.
- 27:03We'll be sharing the raw repository with
- 27:06you for you to download. This is a zip
- 27:07file and then uh from 23rd 30th so the
- 27:11towards the end of this month there'll
- 27:12be short listing of the selected teams
- 27:15for the challenge. So we'll be choosing
- 27:16between five and eight teams and the
- 27:18expectation is for the following weeks
- 27:21uh there will be development up until a
- 27:24pitching event that will be hosted in uh
- 27:27January late January next year.
- 27:30Um there's a prize uh or
- 27:36seed funding attached to this of $20,000
- 27:39to one selected team. Um and we're
- 27:43looking for submissions anywhere from an
- 27:45individual to uh five members uh or be
- 27:50associative with an organization or
- 27:51more. Um and all submissions please make
- 27:54sure they are in English. Um thank you
- 27:58very much and yeah uh we will be sharing
- 28:01this repo uh very very shortly. If you
- 28:04have any questions, uh please feel free
- 28:06to reach out to myself or Kai and we'll
- 28:09be happy to uh make sure that they're
- 28:13addressed and uh provide any assistance
- 28:16if if required in getting yourselves set
- 28:18up and running. So yeah, please go to
- 28:22the link um application link is there.
- 28:24I'll post something in the chat as well.
- 28:26But thank you.
- 28:27>> Thank you. Thank you. And I think now
- 28:28it's the perfect time to move to any
- 28:29questions anyone has. And now let me
- 28:32just uh go to the um
- 28:36stop sharing. I'll just go to I think
- 28:39there are a few questions that were
- 28:40asked
- 28:42uh in the chat. Uh so yeah you Eugene as
- 28:44I said yeah have you're free to see if
- 28:47anyone wants to um wants to join you. Um
- 28:54and
- 28:55um let me see could we share the link?
- 28:57Yeah, the link will be shared now um by
- 29:01uh James, but the GitHub GitHub page
- 29:04will be actually uh shared on the
- 29:07website itself, right? I think there's a
- 29:08link to that, James, right? Correct me
- 29:10if I'm wrong.
- 29:12>> Uh we we won't be showing the GitHub
- 29:13page. U
- 29:14>> just in complications, we'll be sharing
- 29:16the get the repo there and the challenge
- 29:19page. If you scroll to the bottom of the
- 29:21challenge, should be a challenge
- 29:22briefing. And uh if you're interested,
- 29:24please make sure to register on that
- 29:27platform uh on the challenges ADB
- 29:30platform. Sign up, create an account,
- 29:31and then uh register your interest in
- 29:34this challenge.
- 29:36>> Yeah, perfect. Thank you. Thank you. So
- 29:38yes, we are open for questions. We have
- 29:41around 9 minutes left of this webinar.
- 29:43So anyone please ask questions.
- 29:48Feel free to unmute yourself and jump
- 29:50in. Thank you.
- 29:55Hi, this is Abijit Basu. Um, and we are
- 29:58already working on a Himalayan glacious
- 30:00problem of similar nature. Um, we
- 30:02haven't um so we are trying to figure
- 30:05out what's the ultimate application that
- 30:07you're targeting. So quantum mechanics
- 30:10problem and statistics is great but
- 30:14curious is a is a very typical to a
- 30:16particular subject matter whether it's
- 30:17parametric insurance or bond. So do you
- 30:20guys have a clear line of thought where
- 30:22these applications are going to be
- 30:24because based on that how you create
- 30:26this use this uh quantum mechanics to
- 30:31solve a problem could be very different.
- 30:35>> Yeah. Uh so if you don't mind I'll I'll
- 30:37take this question. So look ultimately I
- 30:40think the goal for us is as a financial
- 30:42institution is we want to provide better
- 30:44products for our customers. Uh that's
- 30:47really sort of summary of what our goal
- 30:49is at the end of the day. Um so the way
- 30:52we look at this is this this is a
- 30:54potentially a way for us to improve the
- 30:56quality of the models that we produce
- 30:58and with that comes some advantages to
- 31:01the to the products that we can offer.
- 31:03Potentially it could be uh tighter
- 31:05spreads uh or better pricing efficiency.
- 31:08Um and to us it becomes sort of also a
- 31:11question of better risk, better
- 31:12controls, um better sort of the
- 31:14fiduciary uh understanding uh of the
- 31:18relationship between the products we're
- 31:19offering and the customers we're
- 31:21serving. So
- 31:23I mean for us we're we're looking at uh
- 31:26insurance and the insurance industry.
- 31:28We're specifically looking at insurance
- 31:30link securities.
- 31:32Um so we lots of information online
- 31:34about how insurance link securities are
- 31:37structured. Um and also uh if you go to
- 31:41Arteimus actually it's a very good
- 31:43industry website you'll be able to see
- 31:45recent insurance link securities
- 31:47offerings. Um and with that you can also
- 31:49see some of the documents that might be
- 31:51attached to them that describe uh the
- 31:53triggers the type of triggers um the
- 31:56tail risk all of these different sort of
- 31:58parameters.
- 32:00Um so that's that's sort of I'd say the
- 32:02primary use case that we've attached to
- 32:04this. The sort of secondary use case is
- 32:06can we improve our internal models at
- 32:08the bank which uh have similar or share
- 32:11similar attributes to these joint loss
- 32:13modeling problems and that could include
- 32:15spillover um in bond markets which I
- 32:18think is a very good example. Say you
- 32:20have a default that triggers in country
- 32:22in one country, country country A. Uh
- 32:24what is the sort of codependency or the
- 32:27the potential joint loss um and the
- 32:31spillover impact on countries B, C, D
- 32:34and E who might be holding um some bonds
- 32:38in country A. So you very quickly have
- 32:40these very large um graphs uh problems
- 32:43that are systemically quite challenging
- 32:45to model um using traditional techniques
- 32:48like Monte Carlo sampling. And that's
- 32:51where we were looking at sort of is
- 32:53there some approach here that is
- 32:55potentially more costefficient uh
- 32:58potentially uh provides some insight
- 33:01some breakthrough using a quantum
- 33:03technique that we can't uh achieve using
- 33:06just say a high performance computer or
- 33:09a locally run model
- 33:12>> we'll be leaning
- 33:14>> oh sorry please
- 33:15>> yeah so followup question so what I'm
- 33:18trying to do is so what we have disced
- 33:20so far is the mathematical side of the
- 33:23statistical side of it, quantum side of
- 33:24it. But ultimately this is going to be
- 33:26applied in a use uh business
- 33:28environment, right? Which is the
- 33:30financial risk or financial uh
- 33:32underwriting. Now there are two pieces
- 33:35to it. One is this models themselves are
- 33:37proprietary sometimes like once you
- 33:39build the model it has IP attached to
- 33:41it, right? So uh what is the IP
- 33:43protection that we get of sharing our
- 33:45model? one. The second is when you apply
- 33:48to the business outcomes there is a
- 33:50there is a basically a benefit to the
- 33:52organization ADB or otherwise how do you
- 33:55comp compensate us for that kind of
- 33:58model. So we have already created this
- 34:00models for parametric insurance. So when
- 34:02you create a model for parametric
- 34:03insurance is very different than
- 34:04underwriting a bond because the
- 34:06parameters are very different right the
- 34:08the facilities are very different the
- 34:10policies are very different. So if you
- 34:12could help me understand the first two
- 34:14questions and then kind of the third
- 34:15question is how well are you prepared to
- 34:18take quantum mechanics into your policy
- 34:20framework that ADB has if you have not
- 34:23used quantum mechanics and I'm I'm
- 34:24assuming you don't have the policy
- 34:26framework for that and how do you plan
- 34:27to inform it uh because doing the math
- 34:30is the simplest part of it getting the
- 34:33business side uh with the IP protection
- 34:35covered is second piece of it and third
- 34:37piece of it is how you support it with
- 34:39policy framework that your organization
- 34:40is ble to take advantage of it.
- 34:42>> Yeah.
- 34:43>> Okay. So, the first question I I won't
- 34:45answer myself, but I can certainly have
- 34:47a uh colleague follow up. Um
- 34:52so, if you actually if I don't have your
- 34:54email already, if you could just share
- 34:56your email with me, we can I can share
- 34:57with you sort of some background on
- 34:59that. There should be some information
- 35:01available on the ADB challenges website
- 35:03around IP.
- 35:06The second question um look ultimately
- 35:08if if the solution is good, if it's high
- 35:10quality, if we see a clear advantage,
- 35:13there is potentially a pathway where we
- 35:16could look to deploy the solution
- 35:19inhouse. Um that is not without some
- 35:22complexities. Look, it's not just
- 35:25something you can just take a new
- 35:26service and bolt it into a into a large
- 35:29enterprise environment. But um look we
- 35:32have the purview that if solutions are
- 35:34proved to be useful then there and there
- 35:37is an interest across um the
- 35:39organization then uh we would seek to
- 35:42engage on something longer term uh
- 35:45potentially um but again this is not a
- 35:48given this is not a guarantee this is
- 35:50shouldn't be sort of the basis of an
- 35:52expectation of this project. Um this is
- 35:54just uh you know one of the many
- 35:56potential avenues uh that uh that could
- 36:00be open to uh potential uh participants
- 36:03of this challenge. Um on your third
- 36:07question look we're our our sort of
- 36:10journey in quantum is still sort of
- 36:11relatively early um and this technology
- 36:14is very nent as well. So we're very much
- 36:18of the view we have a a very strong
- 36:20curiosity to engage and understand uh
- 36:23across our organization how quantum
- 36:25could be leveraged uh to provide an
- 36:27advantage to our operations and services
- 36:29that we provide as an institution. But
- 36:33uh you know today I'd say we're maybe
- 36:37not at the level degree of maturity of
- 36:39other technologies such as machine
- 36:41learning for example and AI where we
- 36:44might have mature and sophisticated
- 36:46policies. But I think part of this
- 36:48process is to work in harmony with those
- 36:51different different uh sort of uh
- 36:55directions which is can we find useful
- 36:58applications for this technology? Can we
- 37:00find interested business units
- 37:02internally? Um, and all three of those,
- 37:05you know, that alongside internal policy
- 37:07would move in lock step together as we
- 37:10build confidence, understanding, um, and
- 37:12as well as prove out the the usefulness
- 37:15of this technology. So that's something
- 37:18that we we expect will happen over the
- 37:20coming years as we continue to get more
- 37:22comfortable with uh with deploying
- 37:24quantum in operations and with any new
- 37:27technology because this is such a
- 37:29critical application I would expect
- 37:31there to be lots of questions around
- 37:32sort of you know how does this work is
- 37:34it an interpretable technique um
- 37:37mechanistically you know how does it
- 37:39work under the hood and because
- 37:41colleagues will be I think with any new
- 37:44technology skeptical to understand
- 37:46schedule bet technology until they fully
- 37:48understand uh the uh the fundamentals.
- 37:52>> Yeah, thank you James. Uh just to be
- 37:54mindful of time, we have one minute left
- 37:56uh for this webinar. Um so if anyone has
- 37:59any more more questions, you're more
- 38:00than welcome to email us. Um the webinar
- 38:04will be shared online on the website and
- 38:07also our emails in case and um also we
- 38:11are thinking maybe doing a follow-up uh
- 38:13webinar to have more questions like this
- 38:15answered and more deep dive because I
- 38:17know we were quite short in time but
- 38:18this is more of a kickoff and uh yes
- 38:21we're super super happy that everyone
- 38:23was able to join I know due to time
- 38:25differences and thank you everyone
- 38:27participating. Thank you Fabian, Vardan
- 38:30and James for your inputs and yes this
- 38:33is just the beginning and we are very
- 38:36happy. So yeah any questions just yeah
- 38:38let us know I can um yeah take them on
- 38:41and yes thank you thank you everyone um
- 38:44for your time. Thank you. Speak speak
- 38:46later. Thank you.
- 38:48>> Thank you.
- 38:49>> Thank you. Bye bye. by
About this transcript
This page contains the full transcript of Challenge Briefing and Live Q&A by ADB Digital Sandbox, generated from the public captions YouTube serves with the video. The transcript has 6,056 words across 893 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.