all right great hello and welcome thank you so much for tuning in my name is grant sultan i'm a research scientist in the amazon quantum solutions lab and i have to say i feel very fortunate to have an opportunity to tell take this deep dive with you into the following question is now the right time to explore quantum computing i think for a number of you if you've already decided to tune into this talk the answer is probably yes but that may not be true for everybody and as well even if that is the answer you probably have a few different follow-up questions so by the end of this talk i hope i will be able to give you a few different answers or directly or indirectly answer the following questions starting with what is quantum computing and why is now a good time to get started where can we expect to see some benefit from this new technology how can you build expertise if you're just getting started what's the amazonian approach and who should get started today the kinds of things we're looking for here are answers to the questions how can you engage with quantum computing how can you choose between the different approaches that are available right now how can you find value in this new technology today things of that form so without further ado let's just jump in starting with what is quantum computing at its core quantum computing is a new paradigm for computation right it's a it's a whole new way of processing information information quantum information is encoded into physical systems using properties like superposition and entanglement that come from quantum physics and that information is processed according to the laws of quantum mechanics i'll use the phrase new paradigm a few times but what this means at its core is that we're solving problems in fundamentally new and different ways that the way you process information is distinct qualitatively distinct from anything we've done in the past and as you might expect this naturally lends us to the capability of solving a new class of problems putting this into slightly more formal language we actually have nowadays strong evidence that the so-called extended church turing thesis is false where here extended as a technical term roughly speaking what this means is that it appears as though there are and there do exist problems that can be solved efficiently using a quantum computer that we cannot expect to be able to solve efficiently using any classical computer meaning anything that could reduce down to a turing machine at its core seems as though there are problems for which quantum computers could do this efficiently while classical computers can't they might scale exponentially in some some parameter well you might imagine that if you're trying to build a whole new paradigm for computation one that is really using the laws of quantum physics that building these sensitive devices might actually turn out to be a really difficult scientific and engineering challenge and indeed that is the case but fortunately through a lot of hard work and effectively just sheer determination a number of these really powerful devices have been produced over the last several years by our community and in fact we've entered the so-called nisk era of quantum computing what this means is that just unpacking the definition nist stands for noisy intermediate scale quantum this means we have new devices which behave according to the laws of quantum mechanics they kind of look like quantum computers their intermediate scale they're not really at the level of being full fault tolerant devices that we can use for general purpose quantum computing but they're large enough that we actually can't simulate what they're doing using classical computers and they're noisy meaning they're subject to noise we don't yet have all the resources required to be able to apply quantum error correction so this is an interesting time to be in the field and again just the take home message here is that this is a new paradigm for computing it's a new paradigm for processing information it's just distinct from classical computing and it actually gives us a new way of solving problems which can do things pretty efficiently when you can't expect to do this in a classical computer so i ended this by saying that quantum computing is a little bit of a difficult nut to crack so why is now a good time to get started if it's such a difficult engineering challenge and the main reason is that there have been really rapid advancements happening in this field and will continue to happen that progress on the hardware front is really impressive there have been major developments in the quality and the quantity and the variety of different hardware offerings available today and these these technologies are really passing milestones at a a really good clip i mentioned a second ago that we recently entered the so-called nisk era of computing and while it's an open problem whether or not any nisk devices can offer quantum advantage in the near term or anything commercially viable we do have some algorithms which show pretty good promise towards this front and it's a question we hope to answer another reason that this is an interesting time to be part of the field is that major government initiatives have come into effect pretty recently and there's been a ton of activity a really flurry of activity in all sectors government sectors academic researchers industry startups sort of across the board and this this activity has made it a really fun and exciting time to be part of the industry but all right so i haven't i've told you that things are happening quickly and it's a great time but that doesn't really explain why now is a good time to jump in and i think the main reason that we're seeing from our customers is that well these advancements they're happening really quickly and they're going to happen whether or not we're ready for them and since quantum computing has such promise for long-term transformation it has really long-term transformative potential it's important to pay attention to it however it's difficult to predict which verticals and which use cases which industries are going to be the most transformed by this technology and so by starting now some of the more forward-thinking customers have the ability to understand the potential of the new technology and as well understand the risks associated with being late to the game there's some value in managing these risks and identifying the areas of interest for a given customer understanding where the strengths and weaknesses of the technology will lie and so in some sense there's there's motivation to hedge your bets and manage this risk another reason to pay attention right now is that quantum computing is a very nascent field it's very much early days for this technology and as you might expect with a whole new paradigm for computing it requires some re-tooling takes some some amount of time to learn how to engage with the field to learn the tools the techniques build the skills required to engage with researchers understand the results as they occur and while it takes time to get started there is a silver lining which is up by starting early you have the opportunity to develop in-house ip discover new algorithms really bring your own expertise from your industry and your background into this new field and in fact become a leader at the forefront of quantum computing become a new researcher be a thought leader as opposed to a passive observer so i hope i've established it now is the right time for a number of people to get started so when we do have these devices what do we actually hope to be able to do with them what what can they they do for us and on the theoretical front there's been some really tremendous advancements in algorithms that we have access to broadly speaking these fall into four categories i mean very roughly speaking so there are number theoretic algorithms like shorts factoring algorithm you may be familiar with there are optimization algorithms for solving difficult optimization problems more efficiently like constraint satisfaction problems or linear systems of equations there are oracular algorithms where the goal is to query some oracle the fewest number of times for example grover's search algorithm falls into this category where you want to search an algorithm as efficiently as possible and then another category is for simulation of quantum systems or approximation where you want to find an approximate solution to some optimization problem for example now what this means is that we have access to a set of algorithms that we can use in the long term when we do have finally a full fault tolerant universal quantum computer we have a set of algorithms we can use to solve difficult computational problems well at their core these algorithms are not tailored to a specific use case or a specific industry or a specific application instead what they are designed to do is solve a difficult computational problem and so you may have more than one use case that are coming from different different industries different verticals which at their core reduce down to the same difficult computational challenge and a quantum computer using maybe one of these algorithms can solve that problem the the computational problem of interest which then gives results in these broad different classes of applications across the board the issue with this is that right now we don't yet have access to a full fault tolerant universal quantum computer we hope to have one at some point in the free in the future it's not quite clear when exactly that'll happen so it's an open problem whether or not these devices are going to do anything in the near term we do have a few candidate algorithms typically they tend to fall in the the fourth category of simulation and approximation and those i'll touch on a little bit later okay so by now we've established that we that quantum computers have very large transformative potential we do have algorithms we can use in the long term so let's start to look at who can start to engage now and what they can expect to do quantum computing really has a wide sort of disruptiveness and we're finding customers coming to us from across many different verticals i've listed a few of them here but this doesn't really cover everybody they're coming really from many many different verticals and all these customers are coming with very different use cases and applications in mind so in order to try to figure out what is the most interesting thing that people are trying to do today or really what people are looking for we distilled out a few of the strongest customer signals generally speaking people come to us with uh with a number of different ideas in mind but overall we see a lot of interest in optimization problems and for good reason everybody has a difficult optimization problem they want to solve there's interest in computational chemistry material science and again for good reason i'll get into uh shortly hand in hand with that last one is hybrid quantum classical methods which again we'll also discuss and finally the strongest customer signal we see is in building expertise everybody wants to learn how to engage with this community start to understand what this new technology is how to apply it to their use cases of interest and start to make progress in the short term so let's double click on each of these a little bit starting with optimization it's really no wonder that optimization comes up a lot in the long term we do have algorithms that we can use for solving difficult optimization problems using quantum computers but in fact it turns out as well that even in the short term and in fact even today we have the ability to solve certain optimization problems using available quantum technologies so customers come to us with a variety of use cases there's a few examples here across many different verticals like options pricing portfolio optimization vehicle routing etc well while we have technologies and tools for solving optimization problems in the long term with a full quantum computer today what we can actually do as well is solve certain classes of optimization problems on available quantum hardware typically these optimization problems fall into a class known as cubo problems where cubo stands for quadratic unconstrained binary optimization so let's unpack that definition a little bit the goal here is to optimize a quadratic function of binary variables right so in this equation x is a binary variable taking values 0 and 1. y is a quadratic function of that so each x appears at most twice the variables themselves have no other constraints and the goal is just to optimize this function y in order to make headway here there are a couple of things we'll need to do you need to understand how to how to reduce a given optimization problem to a cubo you need to understand how to solve a cubo using quantum hardware or quantum technologies whatever you may do so a reason this is of so much interest in the short term is that in fact we can solve these types of problems using quantum hardware today that as i mentioned in order to solve a generic optimization problem if you're if you want to use this formalism you first reduce the problem into a quadratic unconstrained binary optimization problem and then in order to solve it using quantum computing or quantum computing hardware available today you have to embed that problem onto the topology of the device you want to use so a given keyword problem is defined by this matrix q and the variables can have arbitrary connectivity right they can they can all talk to one another defined by this matrix when you go to run this on a quantum computer the device may actually only have a set of qubits and not all qubits will talk to one another in order to use that device you have to first embed the problem of interest onto the topology of the qubits that you want to use such that the problem is well-defined this is a difficult problem but it's nevertheless one that can in fact be solved and is is done quite regularly and so today you can solve these optimization problems using quantum annealing devices from d-wave that are already available through aws well in addition to using quantum hardware right now there are other techniques for solving these problems many of these are called nature inspired techniques the idea here is that by studying processes that occur in nature for example quantum annealing using devices like t-wave and by adding heuristics and a few other really powerful classical methods some very powerful kubo solvers have been developed which can solve problems with say tens of thousands of variables even today an example of one of these solvers is available in the aws marketplace and that one as well can solve problems with tens of thousands of variables well while we don't expect that these techniques can scale the same way quantum computers will be able to scale for exactly the reason that we're interested in quantum computers in the long term in the very short term and perhaps even in the inter the medium or intermediate term these these methods these algorithms actually have some promise for solving potentially even commercially relevant sized problems immediately today using classical hardware and since you're solving you're approaching a problem from a different perspective than you may have been trying in the past since it's something coming from quantum technology there's hope that in fact you can solve these problems or find a solution in a more efficient or more practical way than you were doing in the past so these methods give some advantage today they are unlikely to continue to advance at a scale that will be able to compete with quantum computers as hardware matures but they are something that you can use to potentially even find commercial value right now something else i want to mention on this front of optimization is that these kubo problems can also be solved using another class of quantum devices namely gate-based quantum computers for example offerings from ionq or righty but i should point out that these are really proof of concept studies at this time since there's some overhead required in mapping a generic cubo problem onto these devices and the size of the computers to which we have access right now limits the size of the problems that you can actually solve in the long term this will hopefully be another way of solving these problems using a different class of quantum device right so let's move on to the next strong signal we're getting from customers which is in computational chemistry and again here it's no surprise that this is something of interest in general for a quantum chemistry problem the idea is to or oftentimes the idea boils down to needing to solve the schrodinger equation and since they are quantum quantum devices that behave according to quantum mechanics these machines inherently also solve the schrodinger equation so if you can map the quantum chemistry problem onto a onto the degrees of freedom in a quantum computer and if you can map the evolution of that system onto the evolution of a device somehow by applying the right gates for example there is good promise of simulating the dynamics of the system you're interested in and extracting the properties that you were hoping to get well that's something that we hope to do when we have access to fault tolerant quantum computers in the short term there is some hope that quantum simulation will be one of the first useful techniques to which we can apply quantum computers but actually because it's such a difficult challenge there have been really substantial progress made over the last several years in computational chemistry really powerful theoretical tools and techniques and some classical computational advancements and so this is an area where we see a lot of potential benefit from machine learning and high performance computing but if you really want to focus on a quantum computing approach to this problem the way we expect many of these these challenges to be overcome is by pre-processing the problem pre-processing the system of interest using these these techniques that have been developed finding the really difficult computational problem that lives at the core and solving that on the quantum computer passing the results back and using more of these techniques that have been developed for for computational chemistry in the past to post process the result and finally get the answer of interest so this naturally leads us into the next strong signal which is a hybrid method and this is the way we expect many of these computational chemistry problems to be solved in the short term hybrid methods seem to be a promising candidate for a nisk algorithm but i'll come back to this a little bit at the end of the slide one of the reasons there's so much additional interest is that actually these algorithms are indicative of the way we expect quantum computers to be used in the long run namely as a co-processor alongside a classical device so to help to help drive this point home i'll just illustrate this with a an explanation of how a variational algorithm will work and here this is just this is not a generic any uh not a generic sort of variational algorithm but let's just focus on a particular example so looking at the classical side of the problem the idea is to try to optimize some function maybe the problem could be say for example solving finding the ground state energy of some chemical system drawing a connection to the computational chemistry we talked about a second ago in order to do this we'll set up some quantum circuit shown on the right of that figure where let's even just say far you can say that the structure of the circuit is fixed that there are gates on specific qubits and they have a certain connectivity but this the way the circuit looks at every time is going to be fixed the gates themselves however can be parametrized by classical variables and so but the way the algorithm works is to first initialize those variables define for yourself a quantum circuit execute that on a quantum processing unit on some quantum device report back the measurement results to the classical computer and then using some classical back end some optimization algorithm whatever you prefer to use you update your variational parameters those parameters that define the circuits and the the gates in the circuit passing back and forth you re-run the circuit you get new measurement results and you iterate this process back and forth left and right between the classical and the quantum device the hope then is that as you vary these parameters you walk down this objective function for example find a minimum or an optimum perhaps it's the ground state energy while the circuit on the right prepares a state that's perhaps close to the ground state for example now these two tools the classical and the quantum devices they're well suited to different problems right a classical computer can solve a certain class problem very well a quantum computer can do different things very well and so they're two very powerful tools that turn out to work pretty well together the reason this looks like something that might be promising in the near term for disk devices is that these circuits since in this example i mentioned that they can for example be fixed if they are shorter depth or if they are able to execute an amount of time that or with a number of gates that is short enough such that noise does not yet corrupt the results then in fact we might have some hope of running these on a near-term device it's an open problem whether or not this is true and whether or not there is any commercial advantage or any quantum advantage in fact that we can get at this time from generic hybrid algorithms but there is promise for this all right so the next and one of the more important signals that we get from customers is in building expertise a lot of customers are coming to us this is a very strong signal everybody wants to engage build up an internal workforce understand this new technology and we think one of the best ways to do this is to learn by doing so to get started pick up a textbook take a course an online course sign up for a workshop perhaps a training enablement workshop that aws offers start to learn the definitions some of the basic mathematics and the language of this field from there you really need to solidify those topics in your mind and the best way to do that is to get your hands dirty and dive into some code and build a proof of concept a proof of concept project maybe reproduce some known results that you find interesting really understand all these basics just by building small scale proof of concepts and little projects the next step in this process is to map that proof of concept study onto a use case of interest to your your organization or whatever it is you're trying to to do build it out and add new features and then scale it up to the size that as large as you you in fact can or as large as you find interesting by doing this you start to figure out the limitations of available quantum hardware today but also the potential power as the as this hardware matures and as you start to reach the boundaries of what you could do with previous known classical methods the next step is to find a way to push this as far as you possibly can right now there are some of these for example nature inspired techniques that i talked about a minute ago where you can just swap in a different back end and hopefully be able to really push the boundaries and use some new state-of-the-art classical methods such that you're able to solve the problem at a scale that is potentially even commercially relevant today and as i mentioned before you're using new approaches to these problems that are distinct from something that may have been applied in the past and so there's a potential to have some advantage in your in your workflow as of right now the benefit of this approach is that you start with a small proof of concept to help you understand the basics you build it out and make it worthwhile for a particular use case you start to understand where the limitations of quantum computers are as well as where their potential power lies and how they can really make an impact inside the scale of a problem you then find some potential advantage using available techniques today and then as new hardware comes online you can you can just swap back in that new hardware really start to scale up your problems so that it grows as the community evolves and as new devices become available and the scale of problems that you can solve grows you're able to do those right away the next and one of the more important steps if you when you get to the stage is to iterate on the process go back to the beginning start with a new proof of concept and scale all the way up and start to really innovate bring in your domain specific expertise start to understand how you can discover new applications and build ip given what you know about the field and the use cases that you have in mind and really become a leader at the forefront of the technology so with this let me tell you a little bit about what aws is doing in this space broadly speaking we have three pillars there's the amazon bracket service which is a fully integrated aws service to design test and run quantum workflows there's the amazon quantum solutions lab which is a team of quantum computing experts with a really diverse set of backgrounds who work collaboratively with customers on research projects we have the aws center for quantum computing which is a new research institute in collaboration with the california institute of technology or caltech and there's a mandate to research quantum algorithms and quantum hardware so let's double click starting with amazon bracket this is a as i mentioned a fully integrated service in aws users can design and build quantum solutions using an environment of their choice they submit that into the cloud and they can have amazon bracket handle all of the cloud computing heavy lifting in the background they can spin up manage simulators that handle all of the required resources to test the solution that was built on state-of-the-art classical simulators for quantum computing when ready you just need to change a single line in your code and instead of running on a simulator you can instead send it to available quantum hardware already today and since this is part of aws with all of the other resources available this means that this approach is very well suited for hybrid quantum classical methods and as well the service integrates with all of the other services within aws and you immediately inherit the security and the privacy that you've come to know as of right now the hardware that we can use through amazon bracket is in three offerings we hope to have more as they come online there is there are quantum annealers from d-wave there are ion trap quantum devices from ion q and there are superconducting quantum chips from righetti a question we get quite often is one i think is important to really understand right now is why why is it worth having access to so many devices why should we care that we have multiple different platforms to play with and i think one of the the reasons is that these different hardware providers have different strengths and so there's some value in having access to multiple different assets the use cases of interest are going to change over time the quality and the variety of the hardware is also going to change over time and so it's really important in this early stage of quantum computing to be able to pivot and be flexible as these event these advances happen right now with with aws you have access to many different devices and this flexibility is something that we offer right so the next pillar that i really want to click on was the amazon quantum solutions lab and as i mentioned this is a team of quantum computing experts who take a unique approach to working collaboratively with customers specifically we work backwards from a target use case right we work with a customer to identify a use case of interest and then working from that we start to develop and build new prototypes new algorithms and any sort of new approaches to solving this problem using quantum computing but we're fully plugged into the rest of aws and so we have access to really powerful machine learning and hpc support and so in parallel to this track of quantum computing we also design and build some of the best state-of-the-art classical approaches so that once we have a solution that leverages quantum computing we can also benchmark that against the best known classical approaches today a benefit this helps us understand the capabilities and the limitations of each of the different approaches where we can hope to find some advantage and how these things can work together in the near term and something i want to emphasize about the quantum solutions lab is that you don't need to have any sort of quantum computing expertise to engage with us just bring us any problems that you have even if the problem that you have in mind turns out not to be one that's well suited for quantum computing we also have access to these other tracks that we work in parallel and so we can hope to to solve that using any method that seems to be applicable right so let's start to to wrap things up here one of the the last things we want to talk about is what you can do today versus tomorrow using quantum computing and with this try to figure out who should look to engage right now so as of today things that you can do with quantum computers you can design and test solutions that includes hybrid quantum classical algorithms and workflows that i mentioned we can build small scale proof of concept studies and projects and try to use this to learn and build some new expertise we can potentially even push the boundary of what you can do with classical methods and nature-inspired technologies maybe even finding some commercial improvements as of today and we can join the the research frontier and try to discover new applications and new algorithms that are potentially viable in the this nisk era so if you're looking to do any of those things then absolutely you should get started today in the longer term we hope to have some you know first applications of devices some some of the first demonstrations of commercially relevant quantum advantage and eventually we'll have these fault-tolerant universal quantum devices if you're only interested in getting started when we have access to those larger scale machines or further down the road then it's not necessary to jump in today with the caveat that i did mention there is some risk associated with being late to this game right so i just wanted to share this quote from jeff bezos which is that every new thing creates two new questions and two new opportunities and in quantum computing this is definitely the case because this field is very still very research heavy and it's full of open questions but it's such an exciting time to be in the industry because we're discovering new opportunities everywhere we look and so with this let's return to the question we started with is now the right time to explore quantum computing i think the answer is indeed yes absolutely and with aws you can do this in a very flexible way that gives you access to many different backends you can work with experts in the field and you can do this on all on a platform that's very well suited to these hybrid quantum classical workflows so with that thank you very much for your attention
2021-02-11