Sign California Banking Claim Computer

Industry sign banking california claim computer

hello everyone a very good morning to those who are joining us from singapore and asia and good afternoon or good evening to all of you joining us from other parts of the world thank you for being here with us for today's online event on banking on quantum the value of quantum computing in finance presented by sg novate and partnered with the high commission of canada in singapore my name is jin from sg08 and as a singapore government-backed investor we have been building up and driving deep tech innovations in ai healthcare quantum tech and autonomous technologies across various industries at sg innovate much of our work is to connect singapore with the global deep tech ecosystem to learn how we can better harness technology today we have our panel of experts from singapore and canada to discuss why and how companies in finance should start thinking about their quantum strategies and the importance of the role of all stakeholders in supporting an industry's quantum readiness we encourage you to engage with our speakers during the session by submitting your questions in the q a box located on the lower panel of your screen without further ado i would like to invite our moderator for this discussion professor paul griffin to start us off paul please hello thank you jim thanks for inviting me uh hello to everybody yes i'm paul griffin from the singapore management university and we have four great panelists today who can give us some very broad and deep explanations of quantum computing and also we investigate how this can be used in finance so i'm going to let them introduce themselves what's uh first of all uh maybe uh rafael jannik if you could uh give a brief introduction to yourself thank you paul um so as paul mentioned i'm a rooftop yannick i work at the xanadu and at xanadu we're really looking to build quantum computers that are useful and available to people everywhere part of this means that we have built the photonic quantum computer that is available through the cloud but we're also looking at other quantum technologies um to really enable the adoption of of quantum computers by background academically is in physics but i would call myself a rehabilitated academic i spent the last 10 years working in uh various companies large and small really pushing on machine learning specifically computer vision systems for medical industrial applications and then a year and a half ago joined santa due to really focus on bringing its you know deep tech uh its quantum computers uh and make them available to people great thanks yeah so uh next up uh andrew firstman can you give the visit about yourself yeah hello thanks so much for having me um i am the ceo of a company called one qubit we're really thinking about applications for quantum computing um prior to that was co-founder of satologic large number of small satellites in low-earth orbit and my background is uh academically in economics philosophy political science and uh findings great thanks so a good good broad set so uh next up uh phil phil kay can you introduce yourself yeah for sure thanks paul uh i'm phil i'm uh i work with the national research council canada which is a canadian government agency the largest uh federal r d agency in canada uh my academic background stretches back to the early 2000s i was the first phd student to graduate from uh canada's institute for quantum computing based in waterloo i then went and worked at the with for the federal government uh for uh canada's cryptologic agency uh the communication security establishment for almost 15 years um and quantum quantum technologies was a theme that pervaded my my uh my entire career at cse i then went and worked at a company called d-wave a canadian quantum computing company based in vancouver for about a year and a half uh and uh this past spring landed at uh nrc so so very happy to be at nrc and happy to be here great thank you so much and the last but not least uh jane thompson can you introduce yourself hi my name is jane i work for a singapore-based quantum tech startup called horizon quantum computing which does claim software my background is actually quantum information theory uh until very recently i worked a lot in quantum algorithms and worked in singapore in particular the center for quantum technologies so i wouldn't say as rehabilitated academic yet but maybe getting great thanks so uh i think yeah everyone on the call obviously has heard about quantum computing that's why you're here i think it's probably good to start off with kind of you've probably heard all the hype and all the advantage of supremacy so let's kind of start off with getting a kind of a you know reality check so i'm going to i invite phil first of all to give us a his view on what is the current state of quantum computing you know really uh and then we'll open up to other panelists to to adding into that sure absolutely um so whenever um i start a conversation like this about quantum computing one of the first things i like to do is to uh explicitly scope what we're talking about and so what i want to do is just quickly review um the distinction between computational annealing and quantum computing uh very different things sometimes they both go by the name of quantum computing depending on who you're talking to but qualitatively different approaches to uh to computing so computational annealing is is um is d-waves technology and it's a very special purpose uh computing approach to computing uh that is specifically it's a special purpose type of uh computer that's specifically tailored to solving optimization problems specifically quadratic optimization problems um so um and uh what d-wave does is tries to use quantum effects it tries to use qubits to do um to do to do annealing better and faster so um one of the one of the challenges uh for for computational annealing though is that there isn't a strong there isn't the same kind of theoretical foundation underlying it and and the there isn't a clear or proven path um from uh you know current annealing technology to something that is scalable fault tolerant uh uh and and provides quantum advantage um there there is some evidence that perhaps quantum mechanics is going to help with with computational annealing but again that that is still uh the jury is still out on that quantum computing what i'll call quantum computing on the other hand uh um what many would call gate model quantum computing uh is an entirely different animal uh it uh in in quantum computing you have a very clear and proven um path to scalable fully scalable fault tolerant quantum computing uh a very challenging path and and you know as we'll see the uh the um the technology path to be followed is is uh very very challenging but there is nonetheless a clear a clear path forward just um you know proven results that that there's nothing fundamental that's going to stop us from one day building fully scalable quantum computers that we know for certain problems will vastly outperform anything that classical computers can do so where are we now with building quantum computers uh of course it's as i said it's very very challenging uh the types of qubits the types of quantum bits you need uh that the level of control you need over over the quantum bits in a quantum computer is is significantly more challenging than what um is needed for a computational annealing uh quantum device uh and so where a d-wave machine has 5000 quantum bits um uh the largest uh gate model quantum computer at this point i believe is ibm's with about 65 uh quantum bits and that you know that is a slow and steady march they've promised ibm has actually promised recently a thousand cubits by 2023 which i think is a hugely ambitious goal but um if they think they can they can achieve that then that's going to be a very significant uh result and i'll tell you why a thousand is kind of an important number in a minute uh meanwhile there's a honeywell which um is also doing quantum computing although their physical qubits are based on a different technology where ibm is using superconducting technology to build their qubits honeywell's using trapped ions um their quantum computer is 10 quantum bits currently i believe uh and then there's um another approach to implementing your qubits you can use photonic qubits based on states of light um and xanadu does this and i believe xanadu's quantum computers is about 12 quantum bits uh one thing i should briefly point out though is that counting qubits isn't uh you know isn't the the full story uh if you remember back in the 90s when everyone was was looking at clock speed as the way to measure how fast your your desktop computer was and then they realized that uh you know you could there are other things that um could measure the power of a computer um how how high quality your qubit is how much noise your qubit has how long your qubit can last for example how many gates can you put your qubit through before it before it loses its properties these are important characteristics too and so there is a measure called quantum volume that ibm designed to try to account for this and to give you a a bit of an idea where as i mentioned ibm's quantum computer is um um is 65 cubits and honeywell's quantum computer is only 10 cubits um honeywell brags or price both say um a quantum volume of 64 and ibm actually i think honeywell's up to 128 now quantum volume uh and um um ibm's around 64. so um you know six times the number of qubits but only double the quantum volume um so the clear pro oh and and uh as anna do i'm not sure what the quantum volume is rafael can probably talk to that later but um uh all that to say you know the the path to um to full scalability is very difficult one of the key milestones we need to get there is what's called a fault tolerant qubit uh these qubits that i'm talking about when i'm saying you know 10 cubits 65 cubits these are physical qubits so a physical um you know trapped iron or a physical superconducting um loop um to implement a quantum bit because they're physical devices they're prone to errors they're prone to imperfections and so we call these noisy qubits uh and um they're they're qubits that are prone to errors if you want to be able to the problem with noise is that it it uh it compounds and um it it uh propagates and so the bigger you make your computer if you're taking a you know a noisy qubit and then you have 10 noisy cubits and then you have 100 noisy qubits eventually the errors that are that are being generated in those qubits are going to overwhelm the computation and you won't be able to do anything so in order to fully scale a quantum computer you need to do error correction and you need to implement what's called a fault tolerant quantum bit and that's basically a quantum bit that is air corrected to the point uh uh that you can do operations on it uh as long as you want uh and you can you can scale as many as you want the there's a lot of overhead required though to turn a physical qubit into a fault tolerant qubit and it's about the order of depending on on the approach you use it can be about a thousand times uh overhead which means you need a thousand physical qubits to implement one logical qubit so that gives you an idea of the these these significant challenges to to build scalable error-free quantum computers uh what we have in the meantime are called uh are referred to as nisk nisq near-term intermediate scale quantum computers and these are the quantum computers that have sort of fewer than a thousand uh of these physical imperfect qubits and that's where we are today the one of the main challenges on the application side is that we're trying to figure out what can we do with these nisk quantum computers are there any applications that for which a noisy intermediate scale quantum computer with maybe a couple of hundred um physical noisy qubits can outperform uh what we can do classically and there's some evidence there's some hope that there are going to be some applications um probably you know some of the i think some of the the leading candidates for um for promising uh applications of near-term intermediate quantum computers are in quantum chemistry simulations um so um being able to simulate physical systems chemical systems at the quantum level which is important for you know a variety of you know materials design or other applications um for um simulating electron transport all kinds of things in in things that i don't understand in biology and um protein um folding all that kind of stuff uh finance i'm not sure i think other uh other speakers today are going to speak more to um with the potential for applying near-term intermediate quantum computers to finance um i'll just say a couple of brief words about um some of the um the um the the software stacks the the application space and some of the hype that's that's now being generated we're seeing an explosion of commercialization uh you know all the companies who are putting quantum computers on the cloud now have their own software stacks they have their own programming languages so ibm's got got kiss kit xanadu's got penny lane microsoft has its q sharp languages a language there's a company called zapata that's building whole workflows that are sort of hardware agnostic there's all kinds of amazing cool stuff going on but the same time that's also generating a lot of hype and and and to a large extent i think a a set of inflated expectations around um the people are seeing all of this advanced sort of technology work being done on the software side and that sort of leads them to believe that oh well that the underlying algorithms must already be there and so um and of course a lot of what we see in the media sometimes fuels those expectations so i think you know especially in applications like finance and stuff um there does need to be a bit of a reality check and where really are we and what is proven and known about what uh what we're going to be able to achieve and what is speculation and um not strictly blind speculation but what is speculative i'll say and um early stage work that's that's um that's exploring the space well i think i feel like i feel yeah uh yeah well actually why why are we on this habit of soccer just overview of the kind of situation and hardware and things so uh i think uh maybe jane if you can uh say a bit more as you're working for horizon which is kind of a software-based company if you want us to expand a bit more on the uh uh on the software side so to cut um but maybe i can talk in the context of the software and that's what i would know a little bit more about as opposed to the devices and the current state-of-the-art so um and maybe in the context of finance as well to put that into the discussion so we should be really excited about the potential for integrating finance tools into quantum software stack um this is actually very promising and there is good academic reasons for being interested in this and believing in its potential so for instance this comes from basic capabilities of quantum computers maybe not so much the hype but what we know that they're really good at doing and particularly we know quantum computers are very good at simulation and prediction tasks so they have really strong um algorithms in quantum search and quantum work so these power things like faster hitting times to simulate from the long-term behavior of a system more efficiently going mechanically uh quantum monte carlo methods for estimating a parameter to a certain level of precision with quadratically fewer samples quantum mechanically so these are these are really solid and uh a lot of the common finance applications are being built on these uh things like conditional value at risk value at risk portfolio performance estimation optimization they are actually built on the solid tools of the simulation walks so hopefully the software will pull in these elements i believe it is already pulling in these elements uh because this gives us a very strong foundation also matrix algebra uh really promising techniques in matrix algebra uh quantum algorithms of solving linear equations singular value decomposition uh quantum principle component analysis eigenvalue and eigenvector resolution so um th re you get your kind of machine learning advantages a lot of them are advantages and things like classification prediction uh even regression curve fitting inference of relationships and data sets all of this will actually come through from this tool chain and again um i can see a lot of really good progress in the phone software companies on this front as well and in addition to that you obviously have a role to play in terms of optimization on quantum computers for solving financial problems so this is like this would take me an hour to talk about all by yourself but i mean you're looking at variational quantum algorithms quantum approximate optimization algorithms and maybe even things like adaptive quantum search which are coming through and allowing you to do a lot of optimization as long as you can design a good cost function which is a good proxy for what you're trying to minimize and also as efficiently as valuable on the computer so this this software um algorithm suite is very strong the applications and the use cases is actually in the software domain a really big um point of progress at this point in time and of course as phil said the the hardware is is also like matching our expectations against the hardware is obviously going to be something which we should be very interested in doing as otherwise we'll just overhype and i definitely agree with that so i can you ask me for example and the question you sent me uh so if you're still interested in the example yeah i think that'd be uh example and how easy is it for someone to actually start programming these quantum computers if you've got a background saying just python machine learning can you just sit down and start banging out like some other people here you're very smart and um you should actually really have no problems because you're good at handling data structures and you're good at handling abstract objects and really quantum computers just really sophisticated tools for harnessing these particular capabilities so um i suppose this question might be tailored almost to me because i'm coming from horizon and what we actually do is we we allow you to program in classical conventional computer code and then we automatically accelerate your code by just so automatically inferring or recognizing where you could have used the quantum algorithm to do it faster so um and this is a very good question for horizon in particular but um obviously i actually think that you're going to be really good in a really good position to use the tools of most of the quantum software companies because you have the good analytical knowledge if you're able to do this so but i did prepare you an example so i'm glad you're glad you went which is the um gaussian process regression and why would you be interested in using quantum software tools in this context i'm told that this is actually solo clip this is a technique for actually inferring functional relationships on training datasets so you can be a training data set and i am further functional relationship between the points in and in particular i want to do this because i want to take another point which is not in my data set which i am curious about i want to make a prediction about how it's going to perform so you in particular see applications in finance for this apparently in the context of future prices of commodities and to give you an idea about why you should be interested in harnessing quantum computers to do this uh as a case study if i was to take a thousand data points and train my function off this i'll get a certain level of accuracy and this will be good but if i do this on a classical computer and then i want to get a higher accuracy by invoking a million dollar points instead of my training set then i'm going to take more than a billion times slower to run the computation than classical computer uh there might be some special cases where i might be able to get it down to 2 000 times slower but the general paradigm is i'll be a billion times slower to train up a million dollar points than i am to train up a thousand dollar points and this is just you the scaling in classical computational routines it's a scaling issue in which you add a new data point to the training set and your classical algorithm gets slower and that's kind of normally in halfway uh quantum mechanically you don't see the slowdown quantum mechanically you can train after thousands other points and then you can want to increase your accuracy and train up a million dollar points you'll only slow down by a fact of four so what you're actually looking at is um really strong advantages in terms of scaling for this kind of basic matrix linear algebra operations on the quantum computer which is what's actually the backbone of this routine so uh if you have really good data analytics skills you should be really quite excited about what you can do by just upgrading your routines by running seven key bottleneck steps on the quantum computer and i'm absolutely convinced that you are in a good position to try this of course i should caveat what i'm saying which is that phil is absolutely right uh i won't run this on today's miss corner computers um obviously the hardware has an error rate and a number of qubits which isn't compatible with running such an ambitious uh million-point uh training program for fitting a function uh but there is a very good promise that the hardware is actually leveling up at a very incredible rate i'm sure xanadu is going to tell you about their progress i've been reading quite closely and i'm quite excited to hear about this and um the software is very much leveling up to anticipate that the quantum computing will become a much bigger player in many different industries and that's something which everybody should be able to access so i do believe that you have every opportunity given the impressive collection of software expertise out there to actually give it a go if you're good at data analytics and data structures great okay thanks that's great so we've got an idea about the hardware we've heard about software i'm going to hand it over now to andrew and ask about the wider financial applications you say you've got a you know background economics and finance so uh how how is quantum computing being applied now will be applied is it just scaling speed up or is it something something kind of uh maybe i shouldn't say maybe more exciting but even more exciting that quality computing can do in the area of finance over to you andrea thanks well i would answer your first and easiest question first by saying that what are quantum computers doing in finance now and the answer is um receiving funding from financial institutions um but in terms of providing actual value um it's really like saying you know um what do people do with the first transistor and on the one hand the answer is almost nothing but on the other hand you know you can see this as an incredibly important moment in the history of computation and we're a little bit past there right now but we're still sort of at that um you know in one to tens of qubits and uh and so there's a lot of things that we would like to be able to do but the hardware is just not quite there yet um and so probably another interesting question is well how will we use quantum computers in the future in order to solve problems and uh you know i don't have a crystal ball and there is a lot of work to be done there but what we do have is the history of financial uh and computational finance and so one of the things that i find really interesting is looking back to how computers um were initially utilized in sort of big ways within say financial markets and so when i think back to the ideas of um you know if anyone's taken any basic finance in school you might have learned about the black scholes algorithm for um pricing financial options or financial instruments in the future and what's interesting is that what was actually utilized in order to succeed there was somebody kind of recognized that the trajectory of the stock market looks a little bit like the trajectory of say a rocket and that the correction of that trajectory as it's sort of going up and down you know utilizing techniques that were designed in order to simulate rocket trajectories there was an understanding that there was an applicability and so you know if you traditionally studied like ito's lemma and things like this probably you were a physicist doing rocket engineering but eventually people realize those same tools are extremely helpful for us in doing this analysis and simulation within finance and so i think that we're probably going to follow a pretty similar path where what you're going to see is quantum computers are initially going to be utilized for material simulation and simulation of the quantum world meaning essentially we want to know things about the way quantum information interacts and quantum information naturally interacts within the physical world and so we're going to build very important tools for doing things like understanding how chemistry emerges from physics and i suspect that what's going to end up happening is that some of the tools that make it so that quantum computers are applicable within the world of physical world simulation are going to become repurposed and being applied to financial applications and so for me one of the ways that i think about the question of how close are we to uh utilizing quantum computers within finance and how do we know when it's getting closer i think one of the answers will be we will probably start seeing quantum computers applied to the discovery of new materials specifically things like catalysis understanding you know what matter can you put into a physical chemical reaction in order to either speed up or slow down that process these things feel a little bit more within the wheelhouse of quantum computers but once we start to succeed with those you know very direct explorations of quantum information using these quantum information devices that will really open the floodgates to be able to start thinking about some of these slightly less direct applications of quantum computing and i think many of the applications um that were discussed you know just previous are exactly the sorts of things that we're looking for but it is important to note that when we were talking about the number of qubits you know there were two things that i think were very important one was if you have something like a thousand qubits of a certain type of qubit that's actually kind of like having one error corrected qubit and so whenever we talk about um how many qubits people will need to do these sort of financial applications we're probably talking about not these noisy intermediate scale quantum devices but instead sort of full scale fault tolerant error corrective devices and on top of that you probably for these kinds of financial applications need on the order of thousands if not tens of thousands or even millions of qubits and so we are very far away i think from utilizing uh circuit model quantum computers directly to solve financial problems um the other thing that's really interesting to think about here is the fact that um when we think about uh you know competing against regular computers because that's probably you know the competition it's not as though it's quantum computers or pen and paper we now have this much more difficult task of competing against the moving target of classical devices it's also worth noting that quantum computers are inherently very slow today compared to the clock speeds of these classical machines and so it's not only a challenge to find out things that you can do with the quantum computer it's important to find things that you should do with a quantum computer because there's no other good way to be able to solve that problem and so that actually makes this exploration very challenging because essentially you have to go to a finance organization and say what's a really critical thing that you wish that you could do or else your com your company is you know going to be in dire jeopardy that's impossible for you to do any other way today um and of course there are no such problems because if we had those sorts of problems that were you know stopping us from having a company we wouldn't have a company and so we're really sort of trying to get people to stretch and think about you know if you had this new type of device what sorts of things would be really valuable to your company that you're not able to do today and i think that's going to come from not so much you know physicists in a lab thinking about finance but financial practitioners having access to some of these devices and getting much more deeply engaged and so i think the best argument for having people in finance start thinking about quantum computers today is not the expectation that they're going to be able to make more money in the markets today but instead because there's a lot of thinking that needs to happen on that idea of how can we harness these special capabilities of these quantum devices in ways that provide real value for people in the finance industry and frankly that's still an open question okay so yeah thank you very much i think we'll take that open question to rafael to to give him his comments on what's happening in your space and also i'd like to bring in the idea of the quantum internet of actually not just running quantum computers individually you know what's happening in the maybe you can talk a bit about the the photonic side and the connecting up computers and maybe even you know what we're going to end up when we do have a quantum internet and whether that's 10 years 20 years 50 years yeah absolutely maybe first just to comment on what's already been mentioned you know one thing that's kind of been implicitly mentioned is that a quantum computer is wildly different than the classical computer and not only in the way that it operates but actually in the things that it's good at it's something completely different and unique which means that the problems it will tackle and be able to solve um are not problems necessarily that we're thinking about today so a lot of the algorithms that do exist that are being applied to financial um problems might have very familiar sounding names but actually come with a whole bunch of caveats that most people in finance don't normally think of so you know quantum monte carlo is a perfect example uh but through our interactions you're with different financial institutions a lot of the things that they would naturally love to do with their monte carlo runs uh you can't do with a quantum monte carlo so it might give you this quadratic speed up but there's other limitations uh so that that's the negative side is that often you know just knowing that this exists and has a speed up is not the whole story um it's also like the details and the implementation so even with a fault tolerant universal quantum computer um you know the clock speed and how many of these qubits you need actually to build a single file dollar qubit means that any of that speed up might actually disappear but this this is kind of the negative on the positive uh there's whole classes of problems we don't think of in finance anymore because they're not tractable by classical computers so things that people have absolutely never thought to try to solve or have forgotten that that were problems and worked around them that with the advent for fault power and universal quantum computing could potentially become tractable again and and there's almost like a call to action for people in finance to start learning about quantum today um in order to start unlocking what those those applications are on the really negative side i think probably the thing that a lot of people would have heard which is maybe the stick that as quantum companies use to get people motivated to start investing in quantum is this idea that one very well known algorithm is one that can break public key encryption so shor's algorithm has been demonstrated to be able to tackle elliptical curve encryption problems which is what the majority of public key encryption schemes that we use to secure data today um is based on um so this is omething that that we're facing um you know there's ways around it and and one of those ways are kind of the beginnings of what we would call quantum networking or eventually the quantum internet so uh quantum key distribution is the technology where we can leverage similar principles that are used to build quantum computers to actually exchange information on a public channel in such a way that we can basically understand when it's being intercepted and in this way we can exchange keys uh that nobody will theoretically be able to break and and this is really the first form of quantum communication that's been around now probably for well over a decade and there's been various efforts uh around the world to deploy these networks and we're starting to see more and more financial institutions invest in uh proofs of concept of how to secure their communications with such technologies and then if you keep on pushing that forward this idea of a quantum internet the ability to actually exchange uh information in its quantum state between two quantum computers emerges but for that uh the challenge to build those types of networks is actually probably more difficult than building a fault tolerant universal quantum computer so um it's still a ways away and the really challenging thing is that much with like the internet uh you know many decades ago we don't actually know what all the use cases are of of a quantum internet today we can think about a handful whether it be networking quantum computers quantum sensing networks um but it gets it's really challenging to actually think about what the broader applications will be great thanks yeah we got some questions coming in just about on time as well so i'll start off with uh one quick two questions in one from uh from annal uh saying that uh i'm sure some of the panelists can answer this yeah what is a real use case being worked on now in in banking and what is the actual business uh business point of view so i mean yes we're all interested and maybe we can do fast in the future but right now is there a real a real use case with a real business case behind that use case that's being worked on does anybody have an answer for that i mean if nobody's going to jump in i'll i'll happily go ahead so um we have carried a series of projects um um it really depends on what the the purpose is the majority of the projects that we engage on and i think most do are focused around quantum readiness so actually getting um the knowledge within a financial institution so they can set direction usually i like to think of it as a de-risking future investment um so those ones are usually more further looking out and have been focused things on one like quantum monte carlo for place pricing of complex derivatives or for looking at uh using monte carlo for economic models for for regulators or even meeting regulatory requirements but those are things you're realistically not going to be running on hardware today but it will give you a good understanding on what are the hardware specs that you would need to actually achieve this on the other side there are applications that maybe within the next 12 months uh could see quantum advantage so things where you could actually start running on hardware and seeing performance that is better than what you could do classically and these are usually focused on optimization combinatorial optimization problems so things like looking at portfolio rebalancing or even things like attempting to detect arbitrage opportunities within um certain markets um so you know the applications are fairly broad but i'd say they fall into one one of two main categories right now for most of the work that we've seen either the quantum monte carlo modeling type of projects or the combinatorial optimization type of work there is a whole other aspect that people do uh get involved in which is quantum machine learning and it's sanity we love it we we have uh our quantum machine learning penny package penny lane which works on any hardware with any back end that you're interested in but quantum machine learning is incredibly difficult to prove when you actually have an advantage which portion of that hybrid network is doing the heavy lifting whether it's the classical part or or the quantum part so i'd say we like to use quantum machine learning as a gateway drug to get people into the world of quantum from especially if they're coming from machine learning but usually if you're looking for um you know near-term roi real roi you'd probably be focusing on those other types of problems okay thanks can i jump in uh with with just a comment on the quantum machine learning i like that description of it as a gateway drug but uh i guess i just want to point out that one risk i see with that is that because machine learning is so important uh to you know such a wide range such a wide range of applications and it is you know it's what everyone's thinking and talking about now i worry sometimes when quantum computing gets hitched to um the ml sort of bandwagon because it to me it sets up an expectation that if it is not met because at least to my knowledge it is not yet really known what the advantage of quantum computing will bring to machine learning to to ai and if quantum computing fails to deliver on that promise then people you know that that may have an unfortunate effect of of um undermining confidence in quantum computing for all sorts of other things that uh you know besides ml that um you know that it might succeed at yeah maybe just add to what phil said uh you know one of the the reasons that quantum machine learning is so popular is because we don't have fault tolerant universal quantum computers so when we have these nescario devices uh it's difficult to program them kind of with uh ahead of time with a given algorithm it's a lot easier to treat them as a black box where you have a handful of knobs that you can actually turn to get that output that you want so then the idea that you can train this black box looks very familiar to people who come from a quantum machine or from a machine learning side where they're used to building a neural network where they're just going to use gradient descent in order to optimize it and get out the answer that they want and in a very very similar way you can treat near-term devices where you just throw data at it and turn the knobs until you start getting the results that you want but like phil said what the advantage is is still a big field of research i guess you get kind of machine learning that's really worth complex okay well i mean you get kind of machine learning which has to do with the solving linear equations that'll speed up but then you usually have a question about how you're going to handle the qram and the data load in or you get kind of machine learning and the variational methods but then unfortunately you hit get hit with the baron plateaus problem um and you uh there's certainly a lot of debate as to where exactly it will be i'm useful uh in certainly in this devices it's very promising to actually use it as a way to actually begin to appreciate what might be handled on a quantum computer efficiently but you said definitely have some circumspection about any conclusions that you draw okay thanks yeah one thing that i think is really interesting in that space is that if we think about um when we when we say machine learning we almost always mean deep learning today and of course there are a number of other different types of artificial intelligence or machine learning methods which have been proposed over the years that have different requirements than say just the matrix multiplication that forms the workload from most of the the work that we have right now so an interesting question would be are there other methods including for i'm particularly interested in sampling methods that could be used on quantum devices in order to animate some of the less favorable methods of machine learning or let's say methods of machine learning that are out of favor because their strengths aren't matched with the capabilities of current hardware and so um one of the things that's really interesting and i think which sits on top of something that was what phil said to really kick this off is when people talk about quantum computers they mean a number of different things including quantum annealers and universal circuit model quantum devices but we might also be able to use application specific devices that have been produced in order to take advantage of one very specific quantum function in order to animate for example a method of machine learning and so i think that there is also um other things other than sort of universal quantum computers which may be helpful within expanding machine learning because again quantum computers are not just fast classical devices they're sort of very different computing devices that have very different strengths and very different weaknesses and if we can tailor our methods to match the strengths and weaknesses of some of those devices that might be a little bit easier than trying to think about how we can try and outdo what classical computers do so well on the same problems which is almost impossible actually this is really fascinating because i had a chat with david walpole about exactly this earlier this year he's the guy who invented all the no free lunch theorems and he said that he was there when they were first developing the machine learning paradigm and uh andrew lung set an assignment for his students and realized that uh back propagation could be accelerated on conventional gpu graphics cards and then he said there was a watershed movement where machine learning all went to deep learning uh well not deep lining but back propagation and he said that before this was actually a much more rich texture this was exactly his point his point was instead of trying to follow the classical paradigm uh it's only a classical paradigm because of the cheap readily available chip already widely distributed and that was only when they found this that it became the classical paradigm and his point was actually what happens if you mix up what happens if you accept some of the other paradigms as well yeah that's exactly that's exactly along the lines of some of the thinking that we've been doing which is really to say that the success of deep learning kind of required this miracle where the method was built after the chips that could accelerate it already existed and if we're going to rely on miracles in order to advance the state of the art we might be waiting for quite some time but if you can think about what would you need to achieve computationally in order to accelerate some of these out of favor methods i think that's a very rich area of exploration and probably a conversation we should have together beyond this yeah it's a really interesting question yeah i mean one of the things we're looking at it's a project at smu in the quantum machine learning is how is quantum machine learning different from classical different data patterns different resilience to noise how can you know can it be used alongside classical to give another aspect another kind of uh accuracy to the data as you know that adds on to what's what is classically there already so there's a there's a few kind of yeah just trying to see what's different about quantum machine learning to classical machine learning okay thanks uh just to kind of switch into a different mode there's an interesting question here about uh out of all the industries do we think that finance will be the first to widely adopt quantum tech if not then which industries might do that especially interesting to hear from phil from the canadian government perspective so that kind of broader quantum tech which industry is going to jump on it first almost or yeah uh so the first part of the answer is that no i don't think that finance will be the first applications where we'll actually see quantum advantage [Music] um and again this is just my opinion um i think uh you know in nrc there's a lot of interest in uh at national research council there's a lot of interest in um the um simulation uh applications so simulating chemical reactions uh simulating chemical reactions is obviously very important when you're trying to increase yields for chemical reactions for for producing things like fertilizers or or other chemicals um and and uh so it's in a sense it's a simulation problem it's also sort of an optimization problem uh when you're trying to um design new materials that are very uh hard and important optimization problems that have to be solved and so some of the variational quantum algorithms that people are exploring now could have applications uh there and there's actually a company um a canadian uh company uh called oti lumionix that's that designs uh um uh special materials for um for uh displays and other things and they are looking at using quantum computers to uh to help augment some of their optimization techniques uh and then there's also more traditional optimization applications um uh you know around logistics uh you know uh allocating resources most efficiently and and the nice thing about optimization problems of course is that they also lend themselves to uh um to being tackled by computational annealers uh both classical and quantum and so that can in inspire and and um stimulate the the development of quantum of the same of of the quantum technology so the quantum technology that d-wave has to develop for its um computational annealers uh superconducting qubits uh you know is also very gate relevant quantum computers and i see another question and i just want to quickly add to it because it is sort of relevant to this uh there was a question about um uh using quantum computing to tackle large sets of data or tackle uh problems where you've got huge amounts of data and that at least to my knowledge is a mis is a misconception uh that's actually very challenging for a quantum computer because for a quantum computer to be able to to um compute on a huge data set requires what's called a quantum memory a quantum ram so not only do you have to be able to store a quantum state a register of quantum qubits for a long time but you have to be able to access those qubits uh you know on demand and that is a very very challenging um technology problem that that still has to be solved so i don't see those kind of big data processing applications being one of the first uh applications of quantum computers you know eventually it may come but um not one of the first i think optimization and simulation are the core ones maybe just oh sorry please go ahead i was just going to say maybe to add to what phil was saying uh i completely agree that again places where quantum simulation is going to be needed is really where we're going to see advantage first but we're definitely seeing financial institutions invest quite heavily uh in the early days of quantum and and i think it's driven by really two factors one um there's really a talent shortage in the space right now and it's going to continue to exist for quite some time so even if you believe that your application is going to show up in the next five or ten years you really start need to start building that expertise internally now in order to be able to leverage it when it does show up the other one is really driven by what i would say is probably the intellectual property uh environment uh in the united states and the fact that um financial institutions in the united states have started acting a lot more like tech companies in terms of building patent portfolios to go after their competitors and defend themselves and quantum is currently seen as a bit of a green field for this so there's a very aggressive i would say land grab that is happening in terms of trying to build out ip portfolios so you can operate when you do uh assume that there will be a quantum advantage by cross-licensing um and and final follow-up on uh phil's last point which is big data is not well suited towards quantum computing it i love it i think it's something we hear very often uh that people ask about that i don't really like to think about the problems that are best suited or problems th t it takes a very small amount of data to describe but have a very large space to explore for a possible solution so if you can think that within you know uh 10 or 15 numbers you can describe your problem but the possible number of solutions is in the trillions that's the type of problem you want to be looking at for quantum computing it might also just be worth noting that um in addition to the fact that i i think i agree with the consensus here that material simulation might be one of the low-hanging fruits or one of the first things that people toss us into it really kind of depends on your definition of first because one of the most challenging algorithms to implement shores algorithm was uh conceived and written down long before there were actually quantum computers and most people who are building computers or quantum computers now agree that peter shore's algorithm is completely um you know i mean it's sane it's sensible if we had the kind of computers that we're all trying to build then an algorithm that was written almost 20 years ago would be capable of running and so when we think about finance from that perspective it could be the case that the first industrially relevant application for quantum computers is something that's available significantly before the quantum computer that's able to run it and that's probably still very important because if we had the understanding that financial institutions would be able to radically reduce costs say by running a particular application then there'd be significantly more demand for these machines and so thinking about the software even in the abstract is very motivational and very helpful for being able to invigorate this industry another aspect to build on what andrew was saying um is that uh because we haven't had machines to play around with we haven't been able to develop until very recently we haven't been able to develop heuristic type of uh algorithms right uh we've only you know when i was in grad school i mean shore's algorithm was discovered actually 25 years ago and at that time uh yeah the only thing you could do was sit down with a pen and paper and draw out a quantum circuit and then prove some complexity bounds uh now we are actually able to sit down with hundred cubic well 50 cubic quantum computers and try things out that we can't prove any results about and see what happens and see what the behavior is like and that's i think a huge space of unexplored space that's going to be really exciting yeah it will be exciting i think there's a second question coming up in all of this though which is data loading like if you have a big data set or you have some which is very complex you have an efficient way of learning that in and describing that um so you might have a very complex system like a chemical system that you might have an efficient way of representing and learning that database uh you might have a very large matrix which you wish to look at the expectation average of how it behaves or its trace or maybe a partition function for a very large spin system but you have an efficient load and actually in terms of one and two cubic k decomposition of that matrix um in a similar way the gaussian process regression you don't touch qram you actually load it in the oracle and the matrix so actually as this is one of the this comes back to rappel's first point which is um if you use these things very naively uh and then the monte carlo um sampling case which is actually a good proven um well theoretically strong result you won't necessarily get your advantage uh it's not a hammer to break all problems it's something which you have to use relatively intelligently okay thanks so we're getting towards the end of our hour so i just want to ask the question of uh general question of how does someone who's interested in quantum computing start whether they go to what they do do they have to become a quantum physicist read quantum physics books before they start all you know uh who would like to take up the question well i can i can start by recommending a good textbook although it's 16 years old now so i wrote one of the seminal textbooks textbooks on quantum algorithms and we made a very concerted effort to write a book that would assume that you didn't know anything about quantum mechanics and in fact you you know one way or another when you learn quantum computing you are learning quantum mechanics you're learning the concepts of entanglement of superposition uh but it is very possible to do that from a strictly linear algebra perspective uh i guess i maybe want to you know while there are now toolkits and software stacks and things that are now enable somebody you know just to learn a few lines of python and apply quantum algorithms to understand and develop new quantum algorithms um you really need to um you know the barrier is is a fair bit higher uh the the mathematical tool that you need to understand these things is linear algebra you need to have a strong basis in linear algebra basis and pun intended um and that's um so if somebody really wants to understand what a quantum algorithm is and how it's working how quantum interference uh you know generates generates these unique capabilities uh go back and brush up on your linear algebra uh and then there's there's a number of textbooks now available um but that's you know i don't think you can there's no shortcut to uh doing the hard work thanks anybody any other avenues i know that uh ibm have have their ipmq network uh with three quantum computers available uh they have a development uh uh sort of a ide interface to it and you can sign up for free they've got two tools and stuff as well so that's another kind of a practical avenue any other suggestions i would offer the gateway drug of quantum machine learning through penny lane so penny lane is an open source package uh apache2 licensed community developed that that santa do kind of started off and still shepherds along the way but um full of great tutorials to to get you started again like phil said there's really no replacement for doing the hard work learning the math and understanding the algorithms but if you want to get a sense of what it is like to work with these devices uh what it's like to actually run on them um the really one of the easiest ways i found is through through penny lane and you can access the ibm machine or if you feel like it uh you know go through something like aws bracket or azure quantum to access a wider a range of uh devices and um yeah start on your favorite things that you can do on your term devices today great okay thanks so uh yeah it's the last couple of minutes now so you just want to go around everyone finally and just ask have we got some unfortunately we can't get to all the questions but uh we just asked us for the panelists what their final one thought for everyone to to leave leave this with what's the what's the key key thought to leave for everyone to leave with uh i know maybe raphael as i can see your face on the screen i i'm gonna say it it's cheap to get started today it's only gonna get more expensive as we move forward it's limited with the number of people uh the moment that quantum advantage lands the demand is going to go up so i'd say build within your institutions some initial expertise around quantum not so you can start leveraging today but so you can plan accordingly for the next decade as the developments come along okay thanks sir jane your final thought yeah i would say don't plan on the face of uncertainty as much as you can uh work out how you can get some advantage out of this if possible and you're not going to see benefits today on this computers for a hard problem which you can't tackle on your classical that's that's fair enough it's really about planning for the future yeah okay thanks uh phil uh sure i'll just uh i guess i'll i'll give my sort of crush to the old man sort of uh warning don't put the cart before the horse don't skip the fundamentals don't just go and and you know learn kiss kit and then call yourself a quantum programmer do the hard work learn the fundamentals it's that's the only way we're going to expand the field it's new people making new contributions and new algorithms and to do that you have to do the hard work great things and andrew yeah i think maybe just the the really important thing to understand is that quantum computers aren't magical computers they're not computers that are better than uh classical computers they're just something very different and so the thing to think about is problems which are already being solved by classical computers will probably continue to be solved by classical computers and so the real sweet spot the opportunity that exists right now for everyone who's watching this is recognizing ah if this is the strength of a quantum computer and understanding what this is requires the hard work that phil was just talking about but once you have that ability to understand i see how a crux of the thing that we would like to be able to do is empowered by the capabilities of that machine that's the real sort of aha moment that's going to lead to these next breakthroughs and so i think that it's very important not just to understand the quantum information theory component but also really to deeply understand the areas that you would like to be applying these machines because it's not the case that you know quantum computers plus netflix equals quantum netflix which is better than regular netflix it's going to be about finding real opportunities to harness the core strengths of these devices and that requires the intersection of two deep veins of knowledge and that's why it's so hard okay thank you very much well thanks for the panelists uh thanks to sg innovator bringing us all together thanks for the audience listening thanks for the questions and uh i'm certainly even more inspiring than i was before so uh hand over back to uh thank you jim and thank you to all our panel speakers and for the great sharing and thank you to everyone for your ongoing questions um it is always very heartening to see so much enthusiasm at our online sessions but real conversations often happen after so we do hope that all of you can keep connected the recording of this session will also be uploaded on sg novak's youtube channel so do head on there for review of today's event until our next online event bye everyone