 Welcome everyone. This is our interactive panel with Dr. Marcus Walden, our distinguished guest and leading colleagues from Purdue College of Engineering and College of Science. I'm Dimitri Peroulis, the head of ECE and it's my pleasure to open this panel. This panel will be organized and monitored by Professor Anand Raghunathan. Professor Raghunathan is a colleague in ECE. He's the Silicon Valley Professor in ECE. He serves as the associate director of the SRC DARPA funded center on Brain and Spark Computing and he's also the co-director of our newly launched center for a secured microelectronics ecosystem. Professor Raghunathan's interests are in the areas of Brain and Spark Computing, AI, computing with beyond CMOS devices and hardware-enabled security. He's a well-recognized person in the field. He's a fellow fighter plea. He has received multiple awards including nine best paper awards, Qualcomm IBM faculty awards and our very own Purdue College of Engineering Faculty Excellence Award. With that my personal thanks to Anand for doing this and Anand the floor is yours. Thank you very much for the kind introduction Dimitri. Can you confirm that you can hear me? Yes. Wonderful and I will start shedding my screen. I'd appreciate it if you could confirm once more that you are able to see my screen as well. Yes. Okay wonderful. First of all I'll start by thanking the panelists. We have a really eminent set of panelists here today. Hopefully we can have a fun discussion. I also would like to acknowledge Professor Sumit Gupta who will be joining us soon for putting together this panel, you know, inviting the panelists and working with me to think about the focus of the panel. So what he came up with was the future of AI and quantum computing. Will they synergize? So I will, you know, present just a couple of slides that I put together to motivate the panel and kind of set the context and then introduce the panelists and turn it over to them and again looking forward to the audience participation. Some logistics I'd appreciate it if everybody could keep themselves other than the panelists could keep themselves muted and ask your questions through the chat. Remember to type out the questions either to me or to everyone so that I get to see them and I will read them out to the panelists. Okay with that let's get started. If you look at the top technologies for the next decade and beyond and, you know, there's many such lists you can find out there but most of them will have two common technologies that you will see and those are AI and quantum, right? Here's just a couple of examples and I interestingly came up with this new acronym, the Dark Age and I thought okay, you know, that's interesting, we're going to go back to the Dark Age and D, it's actually a acronym standing for, you know, D for distributed ledgers like blockchain and so on. AI, R is extended reality, augmented virtual extended reality and Q of course, the fourth horseman that is quantum computing. So we'll focus on the A and the Q, AI and quantum. Again, these, you know, very few dispute that these are very, very promising technologies have already started to make their mark. The question here in the panel, you know, each of these is probably deserving of, you know, multiple panels on its own but we are focusing on the intersection of these two areas, these two technology trends. Will they form a virtuous cycle? You know, what will the relationship between, you know, AI and quantum computing and quantum information systems at large be? So this is the topic that we will try to debate and discuss. I've started with a few seed questions to the panelists that will hopefully warm up the discussion but the most important bullet on the slide is the last one for the, intended for the audience, your questions go here. So I do encourage the audience to jump in and ask questions. I know that there's several faculty and students at Purdue who are, you know, experts in these two areas. So we look forward to an engaging discussion. So the questions, I'll just quickly read them out and I'll return to the slide when the panelists actually start addressing them. What are the top two challenges in AI and quantum computing and feel free to pick two in each that the next decade of research should aim to address? And then we will focus, you know, on that on that relationship between quantum and AI and the two facets of it, right? How can AI help drive progress in quantum computing and quantum technologies as well as how can quantum computing help drive progress in AI? And then, you know, we have a few other questions but I'd really love to see questions coming from the audience beyond that point. Okay. So with that, I will hand it over to the panelists and introduce each of them one by one and maybe have them give a brief, you know, three to four minute kind of opening statement. And then we will go through, you know, each of those questions. So let me just stop sharing the screen here. We will start with Marcus, our invited presenter from whom we just heard a really wonderful talk on innovation. So for those of you who were not at the seminar, Marcus is responsible for, Dr. Weldon is responsible for coordinating the technical strategy across Nokia. And previously, he served as president of Bell Labs where he was responsible for defining and creating next generation disruptive innovations and the research that will form the foundation of the future ICT industry. What I found really remarkable about his background was that he holds a PhD in physical chemistry, but is, you know, really insightful and knowledgeable about a wide, wide range of topics. And, you know, he's one numerous technical, scientific and engineering society awards for both his technical work, but also his vision and leadership. And I think we saw an excellent example just, you know, from the talk he gave of his ability to distill, you know, very complex, multi-dimensional problems spanning, you know, the business aspects, you know, commercial aspects and technological aspects into the essence of the right questions and maybe the solutions that need to be pursued. So why don't we start with Dr. Weldon and if you could just give us your initial opening thoughts. My opening thoughts, one is despite supporting quantum computing at Bell Labs, I'm a quantum skeptic, but I believe if I change AI to augmented intelligence meaning, and that also explained my attitude on quantum computing. Fundamentally, I think AI needs to be coupled to human needs slash physical world problems and not just do pattern recognition, which I think is sort of, you know, the first phase of AI is simple pattern recognition, whether it's spoken word or image, I think, but our brains do much more than recognize patterns. We reason about those two sensory inputs, you know, auditory or visual. And I think we have to create a reasoning AI system. And the reason I'm skeptical on the quantum topic for this panel is I'm not sure a quantum system will help us reason at a humanistic level help us reason about quantum phenomena, actually fully support that. But if I am a focused or logically think the innovation of the next few decades is about teaching machines to think collaboratively and assisting humans. I think the spot the part that is quantum theory related is a small part of that space, the part that is physical world macro phenomena that typically are not quantum phenomena is a much larger space. So for the purposes of my opening statement, I will say I think I'm a believer in very why I'm a believer in need for verifiable AI systems that verifiable and a couple of ways that the answer is has not been perverted or distorted. And secondly, that it represents a reasonable view of the physical world. And I think quantum systems can't really help with that for the macro problems I'm interested in. So that's my intro. Okay. Thank you very much. And that's great to have that view because I think, you know, I hope that that will engender some healthy discussions. We'll move on to Jennifer Neville. So Professor Neville is the Miller family chair professor of computer science and statistics at Purdue. She is an elected member of the triple AI executive council and chaired prominent conferences in the field of machine learning and data mining. She received the NSF career award was chosen by IEEE as one of AI's 10 to watch and was also selected as a member of the DARPA computer science study group. So Professor Neville is an expert in the area of machine learning and data mining. So if you could share your opening thoughts with us, please. Sure. Thanks, Anand. So you can hear me. Okay. I made sure to unmute. Yes. Great. Yes. So I thought as Marcus started talking that he was starting to say the sentence, I was going to say which is that I'm not an expert in quantum. And so, you know, I work on machine learning and AI methods, particularly focusing on developing methods that will help do prediction and make decisions in complex networks, complex systems. And so I feel like I hardly know enough to say anything about quantum, although I have been on a dissertation committee of a student who was applying quantum methods to optimization and machine learning. So I know that much about it. But in general, I guess I share Marcus's view that the sort of direction that AI needs to go is to move towards more decision making over larger sets of agents than what's currently focused on now. So either we focus on pattern recognition, prediction over large size problems with many instances, but make one prediction. Or we focus on reinforcement learning methods that are helping agents make sequential decisions in a continuous environment. But there's like one agent or two agents. And so I think sort of marrying those two or bridging the gap between them and having systems that have multiple agents making decisions at different time horizons, which requires planning and reasoning at level of complexity that we're not doing right now. I think that's really the future of AI. And I guess my initial thoughts on quantum of how that's going to help is that in general, everything that we tackle that's of any interest in AI is computationally intractable. And so having methods from quantum that would help us estimate more efficiently to learn those models and reason inside them will help us push forward AI. So those are my initial thoughts. Great. Wonderful. We'll come back to that. So the next speaker I'd like to introduce is Professor Alexandra Baltaseva. Professor Baltaseva is the Ron and Dottie Garvin-Tonges Professor of Electrical and Computer Engineering at Purdue. Her research focuses on nanophotonics, plasmonics, optical metamaterials, and nanofabrication. She's a fellow of the National Academy of Inventors and the Materials Research Society and has received numerous awards for her research, including the MIT Technology Review PR35 Award. So Sasha, if you could share your thoughts with us. Yes, of course. Thank you very much, Anna. Well, first of all, I am extremely honored to be part of this panel. And our photonics team at Purdue is entering this to emerging or concurrently happening revolutions in quantum and AI from the perspective of optics at large and optical technologies. I spent many years working on optical materials, and this gives me yet another great pleasure to be speaking today because Bell Labs is known to be the idea factory that actually fueled a range of what we called quieter revolution in materials that actually led to absolutely disruptive technologies. So what I believe is needed in the field of quantum and obviously I'm not a skeptic, I'm a true believer and I consider the area of quantum much broader than just quantum computing because we are witnessing a rise of both just interest but also huge investment and tremendous progress in the areas of both quantum computing but also quantum communication systems that will give us the ultimate security and quantum sensors. And I would say that those areas of quantum, quantum communication systems and quantum sensors would be the lower hanging fruit for the whole quantum technology and quantum information science and technology revolution. So what the challenges and what we have to do is in fact to build on what we know in the areas of devices and I'm speaking about broadly about electronic and photonics and magnetic devices. And now how to integrate this knowledge with existing machine learning approaches to enable the next generation of quantum practical quantum on chip devices. We have databases of optical properties. We have a tons of physical concepts that enable one or the other phenomenon that we would like to utilize. We know different architectures for different devices. Everything including semiconductor industry, quantum photonics and other areas going and entering the area of heterogeneous integration and multiple functionalities. All this require a collaboration as Dr. Walden just mentioned of a human brain and a machine that will be able to process all the data that we have generated in the area of materials, device design and approaches in inverse design together to actually push the frontiers of this field for the end to enable practical devices. Thank you. Thank you, Sasha. Appreciate that. Last but not the least we have Professor Kaushik Roy. I'm very pleased to introduce my colleague. He is the Edward G. Tiedemann Jr. Distinguished Professor of Electrical and Computer Engineering at Purdue. He has been at Purdue since 1993 and before that spent three years at Texas Instruments. He has received a whole range of awards to numerous to list out here, but notably the SRC Technical Excellence Award from the Semiconductor Research Corporation, the SRC Inventors Award, the SRC Corporation Aristotle Award, the DOD Vannevar Bush Faculty Fellowship, the Faculty Excellence Award from Purdue College of Engineering, the Humboldt Award, IEEE Circuits and Systems Society's Technical Excellence Award. Kaushik is also the director of the Center for Brain Inspired Computing, a large multi-university effort at Purdue focusing on cognitive computing all the way from algorithms to hardware. Kaushik, if you could give us your initial thoughts. All right, thanks Ananda and thanks Sumit for including me in the panel. So let me start by saying that I also don't know much about quantum computing. In fact, I took and I audited of course this semester that was offered from ECE and I didn't even do the homework. So the students in the class probably know more than I do. So what I'm going to do is I'm going to get started with AI which I know a little bit more and so there are several interesting problems to be tackled in AI and then AI and then there are several issues that needs to be considered. To start with I feel that explainability and understanding AI is still a black box. So there's a need to really come up with techniques to have better explainability, better reasoning and to be really able to say why we are doing what we're doing and why we're not able to get answers in certain cases and to be able to explain that properly. So that's one thing that's going to be extremely important from the AI point of view. And some of the other areas and that's from the algorithm and learning side of things, but if you are to really think about the hardware aspects of it, it turns out it's going to be energy consumption, energy consumption, energy consumption and that's huge and if you are to really have AI or machine learning integrated into the IOTs and on these edge devices, there's a need for really thinking about devices, architectures and to be able to co-design so that we can get a huge amount of energy improvement and to be able to run things with the battery. Now, not knowing much about quantum computing, certainly I agree with the fact that AI can certainly help in coming up with the right kind of materials, right kind of devices to explore that space for quantum computing. On the other side of it, if you are to really think about AI algorithms, is it possible that I can potentially somehow be able to use AI algorithms in the quantum domain? That's another area that one can potentially think of. I don't have any magic bullet or answers for that, but those are possible areas where quantum computing and AI can potentially collaborate. Okay, thank you all for your very insightful initial thoughts and so I guess maybe I will go down the list of questions, but since we had at least a couple of, if I should say quantum skeptics, maybe I will try to explore that aspect a little more and at least from my reading and what I understand of what people are making of this, for example, this area of quantum AI, maybe explore and ask a few questions down the road, but let's start with your top two and this could be a very brief answer. What are the top two challenges you think need to be addressed in AI? I know many of you have already talked about things like explainability and opening up the black box, but if you could just very briefly state in your view, what are the top two challenges for AI and to the extent that you are comfortable for quantum computing as well? I will start with Marcus maybe. Okay, well I've given my challenges in AI. In quantum computing, I think there's still only Shor's algorithm and Grover's algorithm, both pioneered Bell Labs, that are provably mathematically superior on a quantum system and therefore can achieve quantum supremacy if you want that on a quantum system they will outperform a classical system. What there needs to be on the quantum side is a set of algorithms that go beyond factorization and unstructured sort that can be used generally. Obviously there are quantum mechanical calculations one can do that of course work well on quantum systems, but most of the interesting quantum systems actually end up being analog systems in my view, so they're actually modeling physical phenomena through capital oscillators and icing model oscillators and so I would say, and that's because the physical world is more interesting sort of space to explore with efficient computing than the quantum space simply because there are no set of quantum algorithms that are provably better than a classical computer other than those two, so that's the dearth of stuff that you can do on quantum computer that that's it and I don't allow for quantum chemical calculations which of course should be superior to decompose all the states there because they're inherently quantum systems so I think it's human world problems that can be solved by a quantum system more efficiently than a classical system I think that's a gap and so that would be my big one then of course most quantum computers are massively supercooled liquid helium cooled systems that are superconducting and or have you know fraction quantum ball states so they're big monoliths of computing resource I wonder in the new edge cloud paradigm with low latency systems where the quantum systems will always be running in the back end somewhere but they won't actually ever be able to scale to the edge given the nature of the materials involved in the cooling systems etc therefore they're a background back end system they're not a foreground real-time processing system which again but sort of marginalizes them so by the way I do agree with the design of materials but I put that in the quantum chemical slash quantum system design so that's what I would say okay maybe I can continue and I would like to definitely build on that and I'm not an expert in quantum algorithms either but what I would like to point out is that indeed these are very complex system and we need to develop really robust platforms to actually demonstrate scalable quantum computing what's important to realize though in the whole area of quantum is that there will not be a single platform for quantum technologies and I think that this is something that is like not clear and people are sitting and waiting and and and waiting to hear okay so the Wiener is you know superconducting qubit or the Wiener is this for enabling the quantum technology whereas in reality and that's what we see already in emerging markets of like quantum sensors and other quantum-based devices that we will have to branch out into different platforms and for example talking about quantum computing quantum photonic computing might not be the way like let's say straightforward way to universal quantum computing but it might be enabling technology for solving a set of problems and also for demonstrating supremacy to start with so I do believe that we will have to pursue different directions and and have this different platforms investigated in order to enable different applications on another point which is different from the you know quantum aspect I would like to go back to classical machine learning algorithms and I think that what we have been missing in the whole area if you wish of engineering and and technologies is just to integrate machine learning approaches with our development and optimization of devices physics informed machine learning algorithms is something that we are missing well at least in the fields that our groups are working in and and that's where we have to go how do we couple our knowledge about physics and phenomena involved in a specific system whether it's a dark matter detector or a single photon source and a train or develop a machine learning algorithms that will help us to design the system to speed up the readings and to achieve greater performance and greater sensitivity for example in case of sensors Jen or Kaushik do you want to add down to your initial comments in regards to the top two challenges yeah I mean you know when I when I initially talked about the fact that explainability is really one of the biggest problems and opening the black box of AI you know I can add to that there are other issues that one has to really think about I mean for better learning you know one of the things that there's a need to focus on from the from AI point of view can we learn with less data can we have better generalization can we have for example you know better I mean it is also related to explainability in a lot of ways you know adversarial inputs that some of these systems the AI systems get fooled very easily we don't know if the brain gets fooled or not but certainly some of these systems do get fooled the brains do get fooled in different ways but can we also learn from you know neuroscience and if we take some cues from neuroscience and try to really build systems taking again biological inspirations that there's a good possibility that we can actually implement not only better learning algorithms but possibly in fact even better hardware and on the other side of it if I were to really go into thinking about the hardware and the devices you know traditionally we have been all always using the standard CMOS circuits and the CMOS transistor certainly are excellent on off switches though somebody might complain that the scale scaling of technology there may not be that good switch anymore but anyway they're certainly good switches but they may not be really a good you know they may not be able to mimic a neuron or a synaptic dynamics really well so to that effect there's certainly a need for we thinking about new materials new devices that can potentially mimic the neuron neuronal and the synaptic dynamics in a more efficient way to be able to build you know a hardware that can implement some of these algorithms or brain-inspired algorithms more effectively and efficiently so that's certainly coming more from the AI side of things algorithms and the hardware and I guess we'll be talking more about the quantum part of it later on yeah so I guess I don't feel comfortable identifying challenges in quantum but I can say that in the AI space for sort of several of the issues that were already mentioned like the physics-based learning or the explainability or the sort of neurosymbolic stuff that I was alluding to those the the sort of difficulty of those problems is that really we're trying to learn and optimize our models over very very large search spaces and even finding explanations means searching over the space and to see how why did you narrow it down to the decision that you made and so I think a key challenge in AI and more generally just in optimization is is how to become more effective at doing combinatorial optimization over those large spaces and that could be a single model where it's parameterized in a combinatorial fashion or over a space of possible models where you're searching over the model structure and then it becomes combinatorial at the model structure level and that's really the thing that's hampering progress in AI right now because all the successes that we have with deep learning and neural networks are really with differentiable functions that make the optimization that much easier to do not that it's easy but it's easier than the combinatorial optimization and so if quantum computing was able to help us do that combinatorial optimization much more effectively then I think there would be sort of we would make progress in leaps and bounds in the AI space and a lot of the things that are difficult right now to have full-scale you know sort of human level AI and reasoning and planning would start to become much easier so I think that's really where we need to to move towards and in sort of on the flip side I guess I've heard some things about how even to understand what the quantum sensors or systems are doing we might collect large amounts of data to then try to figure out what's going on and similar types of sort of learning needs to happen in large search spaces if you're trying to prove the safety of certain AI systems and so interestingly we can potentially use machine learning algorithms to look for patterns over those large spaces to try to figure out how to search over them more effectively so even though we need quantum to do search maybe machine learning can help learn how to do that search more effectively both for quantum and then to funnel it back to the AI methods so I'll build on what Jennifer just said a space where I'm not a quantum skeptic is where if you need to train a quantum system then I absolutely agree that if there were an AI system surrounding a quantum system where quantum system computes something from some inputs and then decomposes the states or collapses the states to an answer someone has to judge the goodness of that answer it may be quantum correct but is it correct or is it the desired answer for the problem in question right so unless it's a closed form problem where you know the answer space when in each case you just acknowledge the quantum system behaved as as expected but if it's a large space there's a lot of expectation values you can imagine having a AI system supervise the learning of a quantum system to discern which of the outputs are the desirable outputs think molecular design for example if there were structures computed by a quantum system that were possible drug designs right previously unknown and the way we do that today is either make it or have an AI system with guests right and it guesses based on its trained models but if you've had to quantum system a quantum system could calculate molecular designs that would be optimal for drugs or whatever and an AI system monitors that with some set of rules that say yeah that that has good let's say marketability or manufacturer ability which a quantum system will never know about right you could imagine an AI surround on a quantum system would actually be a very interesting culprit no that's that's fascinating I you know I think I like to focus a little more on on on what both Jen and Marcus just said which is if you know looking to the next stage of AI right you know going beyond perception to you know reasoning decision making you know whether you call it causal AI and Euro symbolic AI broad AI and so on I guess one of the challenges is it's not from a computational perspective it's not just more linear algebra and more matrix math necessarily that's going to be needed you know today's hardware like you know the GPUs or the GPUs and and so on are really good at that stuff and you know again personally I don't think a quantum computing system is ever going to be competitive at just you know matrix multiplication that's not the right problem but if in the AI you know path if there is a need to search over large combinatorial spaces right and you know integrate sort of those search engines you know with the more traditional statistical AI systems is quantum computing the way there because certainly even from the neural perspective the demand for computers growing way faster than traditional hardware can keep up right I mean open AI estimates it's doubling every three and a half months that is you know way way super Moore's law or anything that you know current hardware roadmaps can can even think of providing so do you think that's an angle where maybe perhaps in a limited sense with quantum computing can contribute yeah but I then I get back to my algorithm problem and I need if I need to operate on algorithms sort of provably better on a quantum system otherwise and the quantum system is not going to be lower power if I think about if I include the cooling system right a quantum system is a is a energy hog the actual qubit structure etc may actually be relatively low energy but the surrounder is massively power consuming so I think it has to be the set of algorithms or problem spaces which has maybe molecular quantum system design and the two known good quantum algorithms then absolutely you're right if that can be generalized to a lot of problems then quantum is huge and the AI surround helping that quantum system learn or achieve a human valuable outcome because as I said if I think about quantum systems don't really understand anything about what we might consider economically viable so if I'm trying to answer a problem quantum system will simply collapse to whatever is the quantum minimum right given but but it may not be economically viable that answer so I think there's some sort of the AI system in some ways to encapsulate all knowledge we have and the quantum system could be used for the part of the problem that was inherently quantum good does that make sense so from your perspective of you know sort of the investments going into quantum computing right the the Google's of the world the Microsoft IBM's is that you you think the premise or is it different because clearly you know it's not clear that just factorization can drive that kind of investment most of those I think and then they're actually annealing systems and quantum and they sort of quantum annealing systems or analog annealing systems that actually set something up in a state and then anneal it to an answer but they're trying to compute things that are actually more coupled states but they represent what I would call macroscopic physical phenomena because they're describable by say icing mathematics or icing model so those systems all are good at solving icing problems but they're not truly quantum problems right they may set up the states in a way that looks so semi quantum and they're entangled but in the end I think most of the problems they're solving I would all argue icing type problems which you know are not fully quantum they're just coupled but I think we need to assume in coupled systems and truly quantum systems and I honestly I sponsored one in Bell apps and we couldn't ever figure out what we would do with it it just seems super cool and so that's not a good justification for research we actually had one that was based on a myorama fermion a boson interaction based on a fraction of quantum more effective phenomenally interesting thing but in the end it wasn't clear what it would be good for honestly so we were going for the sake of going because the physics was super interesting but I'm still yet to see something that says and I think NIST has a list of quantum good problems but whenever I looked at it I was never inspired so it's more it really in the algorithms for a broader class of problems you know in the quantum space that's the bottleneck okay so maybe we can take one or two from the audience because I I'm starting to see a few questions flowing in and this can be any of the panelists feel free to address what do you believe is the future of a silicon qubit based hardware implementation of a quantum computer you know furthermore do you believe that there would not be a singular physical implementation serving as the zeitgeist such as today's computers yeah I think Jen answered it very well oh actually yeah sorry uh alessandra uh there won't be because all the technologies are good for something in quantum space um and the extent of the qubit range you can get versus the stability versus the error correction so quantum systems essentially have all these error correcting qubits uh and different physical materials have different requirements of that because they have stability that's different right and so therefore the gate structures you can build with each one or the extent of the quantum system is different for each one so I think there will be a set not one that it won't be a von Neumann machine x86 or ARM architecture processor there'll be a set and there'll be a pool of resources that are available centralized that you can submit your tasks to and get an answer back I think I think that's how I see it being and it'll be a whole host of different material systems so it won't be on your desktop it won't be in the edge cloud it'll be a pool somewhere for the set of problems it's good at solving yes I also I just want to add a couple of things from the technology driven platforms like many people believe that it really won't be practical if we won't base it on existing like semiconductor production lines and one of the approaches for example in quantum photonic computing is actually to utilize the infrastructure which is already available but the truth is that we would need to have this smaller fab that would be working with all those other exotic materials and architectures that are hard to realize but that would be specifically tailored for maybe addressing a specific need of a problem depending yeah of what people have in mind sounds good thank you there's a couple of questions that are I would say more related to the talk Marcus's talk from before and I will since you know we have you here maybe post them towards the end so just this is for the people asking the questions I'll focus initially on the questions related to the panel and then and then we'll maybe you know ask a couple of more general questions and anybody can feel free to respond the next one I'd like to pose is should the problems I'm assuming that this is for you know quantum computing be formulated in terms of solutions to stochastic differential equations would that help in some sense and and perhaps it's related more to the quantum annealing but I'll put it to the panelists to see what you think can we make more progress towards you know bringing more applications to quantum computers by thinking of more problems as solutions to stochastic differential equations probably not being true expert I would say probably that's that's an interesting space again to frame problems that way and have them sort of coupled solutions what you're looking for is a space with a coupled solutions right that you can then simultaneously create all the states all the solutions and then based on the applied input achieved from many solutions the one outcome from a complex problem space okay anybody else has a take on that there's one other question quantum computing aside what would be the role of quantum communications in the future do you envision a quantum network with fundamentally secure communications that may be integrated with classical networks for example to increase the security of the network yeah quantum key distribution is already used over optical networks and Sasha can answer that but yeah they're already deployed so that you can determine whether or not the information transferred over that one is it's it's a key is distributed that the key then can be used in a classical way right but you want to make sure the key distribution is is was not was not intercepted and and indeed there's some sense that you could then look at the information transfer and look for perturbation of the information transfer as well in a photonic communication network I think maybe for the most ultra secure networks but there are other security approaches of course that I think it's nice to have quantum key distribution but equally there are pretty effective methods for distributing keys that work well today of course a quantum computer could break some of those RSA methods so there's a bit of a chicken and egg there that maybe we need better key distribution if we have a quantum computer that can actually crack the cipher codes of current cipher models or security models but Sasha what do you think well I definitely think that it's coming up and we already seen the first demonstration using the satellites in China and actually this country was falling a little bit behind on both investment and and investigation of quantum communication systems now a department of energy has made really tremendous investment in the area both quantum computing and quantum communication systems so we have a chicago exchange link and I mean that's that's it's just a question of time when these systems would be would become practical and here again we are looking into systems that would be integrated with existing platforms so with ontopotonic circuitry for example with optical fibers and yeah and that's where we're going okay so I'd say there's this is more of a comment but I'll put it to the panel in case anybody wants to react it's interesting that intel the specialist in silicon is building their architecture I'm assuming this is quantum computing architecture on superconducting bits not semiconductor quantum dots so does that mean silicon is I guess the implication there is silicon maybe doesn't really is not the way to go for quantum maybe we'll start with you Sasha do you have any thoughts or Marcus anybody yeah I mean the Bell Labs one was based on three five materials indium gathering last night quantum two-dimensional quantum structures that and that's typically the case there's a nano material approaches those graphene based approaches I think silicon is the platform but probably non-silicon based devices is certainly the case but so that's where the photonics and the quantum come together because they tend to use the same similar materials so that's that's true I think all the quantum phenomena I've seen generally are wells or layer structures that are confinement structures that achieve some sort of quantum state that is not based on silicon alone it may have some silicon elements to it but it uses tries to use CMOS processing to produce those but in the end it's it's a lot of you know vapor deposition and multi-layer structures and then CMOS circuitry around it to to program it yeah again it's just the the question of what problems and what are our aims if we're different platforms and if we are talking about silicon and conventional semiconductor industry we have to mention one more time that the the future of the whole industry is heterogeneous integration so whether we are talking quantum or not essentially every device is heading towards ultimately integrating dissimilar systems interfaces 2D materials together with established silicon and many more speaking specifically about one of the most promising platforms for quantum photonic systems and on-tip circuitry silicon nitride for example is one of the promising candidates and again this is something where you can build on what's available and explore both the CMOS compatible materials like silicon based but also three five there is a lot of research on integrating these two platforms together as well and in future we will be integrating these different materials and architectures together because we are heading towards multifunctionality and more complex devices I have a question that seems right up Jen's alley oracle based algorithms like Grover's search algorithms seem fundamentally similar to the k-armed bandit problem since Grover's algorithm has been proven to be superior over you know classical search algorithms how feasible would it be to extend this similarity or superiority to multi-armed bandits so I guess the implication is you know is there a connection there that would that you know would make it helpful at least in the rl context for quantum computing to help yeah so I am not familiar with that result but I saw the question and then just search for it last question was being answered there's already two papers on people trying to solve multi-armed bandit optimization using this kind of idea so I guess the answer would be yes so this seems to me exactly the kind of thing that you know I'm not familiar with because I'm not working in the quantum machine learning area but any of the advances in the algorithms the computation algorithms that are made if there can be shown to be a relation to the types of optimization problems that are needed in reinforcement learning or other areas that we're trying to push forward then absolutely this is the kind of success that we would be able to build on but I don't know so if you if you literally just search for quantum multi-armed bandits you'll find the two papers that that I just found so I think that would be a great thing to look into and I will definitely look into it more after yeah and these are the this is the algorithmic expansion I talked about the trick is nearly every time the reason why there are only two preferably good quantum algorithms is every time one of these is proposed you can decompose it into a classical equivalent that performs as well or better so that's always the trick it's not that a quantum system can't do it it's can it do it better than a classical system and that's that's the hard part so I know a little bit about k-means testing and k-base problems but I you'd have to see if you could actually decompose it into something that was classically more efficient to compute than a quantum system and that's always the test and that's where most of them fail is everyone says oh look I got excited I did it on quantum computer but then you say yeah but you know you could have actually done a classical system with greater efficiency by any computing metric right of number of computational cycles and so I loved about the algorithmic space is is expanding but my only hesitation is it's been expanding and contracting for years in quantum computing yeah and your your observation is exactly the experience I had when I was on this dissertation committee where they were trying to do optimization with the d-wave chip and and specifically they were showing they could do it but in the end they weren't actually faster than just directly optimizing with our standard methods yeah exactly but I'd love to be so I'm a skeptic but uh an optimist maybe I'd love to see that that algorithmic space expand and more for new computing technologies of every type a related question I think you know and and this maybe to a certain extent you know is is you know sort of a terminology or scope issue right are we are we too hung up on quantum supremacy uh you know if there's a benefit practical benefit to these quantum annealing type systems I mean I understand the terminology discussion aside you know should we hold our feet or the the quantum community's feet to the fire on quantum supremacy or if there is a you know should the focus really be on these you know pseudo quantum if you want to call them pseudo quantum systems yes short answer yeah but I suspect it's the quantum community that's actually holding that that's actually promoting quantum supremacy because it's headline grabbing if if you wrote the headline we will be useful in some part of parameter space that doesn't get the same headlines as we've achieved quantum supremacy okay sounds good a couple of other comments questions how about there was a great discussion on exploring new quantum technologies driven by AI I guess you know Sasha and Kaushik briefly mentioned it how about new adaptive AI algorithms driven by the physics of quantum technologies or let's maybe expand that to you know quantum annealing type technologies can we should we try to rethink the AI algorithms to really take advantage of those hardware substrates or is it is it stochastic gradient descent or you know whatever else in the hardware space I guess I think in the AI space we are happy to try to read redesign things that are going in multiple ways that are going to help us learn better and so if the way of thinking about how the quantum systems are working can transform how we frame problems in terms of search or optimization I think that definitely will happen I guess the way I have looked at things is to be more inspired by biology and how our biological systems learn but I think similar things could come from other types of other types of systems so basically I feel like in the space of machine learning and AI we're really just trying to abstract out these sort of problem formulations of how to do this search and learning and pattern recognition given sort of observed data inputs and our ability to interact with the environment and so if there are new substrates that would allow us to do that more effectively I think that would definitely be interesting to explore okay great one more from the audience is it possible that AI may actually hit a wall beyond the current success of ML models or do you see a path forward in harder learning slash optimization type problems I think we've sort of covered it a little bit but maybe if you could address and maybe start with Kaushik and anybody you know anything is possible but then again you know I mean I see the possibilities of moving forward with new algorithms is enormous I mean I think the reason why sort of AI made such a huge progress starting in you know 2010 or 2012 is because the hardware was available and more recently what you also see is that some of these you know progress that has been made in natural language processing and so on and that requires a good amount of computing right and so I believe that once we have better hardware and we're able to explore these new algorithms more effectively I I see you know interesting things happening I guess I can follow up on that I think that I totally agree with what Kaushik was saying that the success currently of AI is due to the availability of the hardware and the data and the number of people that are working on solving these problems altogether that makes for a massive effort to sort of push forward this success in AI. AI typically has over promised what we're able to do and so I like as the cynic in me I say we're absolutely going to hit a wall where we're going to come up against harder problems that people expected to be solved more quickly but at the same time I think the amount of interest and effort that's being put forward from these all these different dimensions means that we won't hit quite the same wall that we did back in the 80s when we had the first AI winter where people sort of gave up and said it's it's way too hard so I think we have we're on a good trajectory and I think as long as we keep identifying the sort of local successes things will keep moving forward. So I'll go with go ahead Kaushik. No I was just to add to what Jennifer said you know there's a lot of interesting developments are also happening in the neuroscience domain so again I believe that there are new learning possibilities you know taking cues from what we learn from neuroscience and that is also possibly going to push us forward into building interesting and more you know complex dynamical systems. So I'll build on what you said Kaushik, build on what Jen said. I think analog coupled neuromorphic things might be better chips for new AI than quantum chips because there's a basis in and again the problems we want to solve are sort of human problems not quantum problems we exist in a neurological analogy world it makes sense that if we had a computing device there was a highly coupled neuromorphic analogy thing it would solve those problems as efficiently as we do for physical world problems but it would explore the space much more richly than we do because we live a linear existence right and then we compare notes with other humans and say oh that must be the answer but of course you'd like to do a much larger search and analysis like AI systems can do so call it analog search space or neurological search space I think would be really interesting and probably more interesting than quantum overall in terms of human impact. A colleague of mine actually this is more of a comment says and I'm not sure if this is specifically directed to Jen and I think that this could trigger a whole different panel so I'm just gonna you know so we end on maybe a bit of a provocative note with a preview for maybe a future panel I'll say you know this is in the chat you know from Professor Jay Kumar wait a minute do you agree with the hardware guy that ML success is due to hardware? Yeah because the algorithms are 20 years old right their CNNs come from Yanlacoon and Co in the 90s doing you know image recognition or figure recognition the change was the hardware and then novel versions of CNN like GANs etc. Not just the hardware the availability of data right so the hardware the hardware helps us be able to harness that data and then I also should give a shout out to the thousands of grad students that do the tuning of these algorithms either on internships or while they're working on their PhDs because that's also a huge component of this as well. Maybe by May I would like to jump in Jennifer you just mentioned graduate students and I would like to bring up another challenge which is the most quantum eye and that's training the future workforce now I'm putting the head of a workforce lead for the quantum science center and that's where we are lacking so this panel is organized by ECE and we are responsible for training the next generation of scientists and engineers that will be pushing these two fields further and they have to understand both of them. That's a great point I mean you know we have just you know getting to the end of our time but if anybody wants to share their thoughts on that angle on you know education and workforce development either an AI or quantum you know any thoughts we can close on that note. I guess I would say that that also points to the complexity of what we need from the students right now because we talked about you know having this stack that goes all the way from you know chips to the algorithms to the data and the mathematics that lie on top of that and the higher level reasoning that we need to put into all the system to have it you know do the kind of AI things that we want really we're expecting students to be sort of familiar with things all across that stack which is asking quite a lot of people and maybe looking forward students could think of sort of where they sit in that stack and and think about expanding their view to you know maybe one layer above or one layer below wherever they're sitting that would be really useful to them sort of moving forward in the works. Which is a pitch for the multidisciplinary nature of Purdue and the collaborative nature so I think it's a good thing I'll end only an end of my favorite quantum quote when you think you understand it you only reveal that you don't so that makes it that makes it very hard to for students but I'm highly simple an AI is somewhat similar in a mysterious way although it's mathematically understandable its reasoning is is hard to found them so it's two areas where the understanding is mystical slash mythical and they've come together to make a real odyssey for students but if you can operate in that space as Jen said it's you have a great career ahead of you wonderful okay so we will close on that positive note and you know I'd like to thank all the panelists for sharing their insights their time and in particular Dr. Weldon for agreeing to do this at the end of of a fairly long day thank you very much I believe this is being recorded and will be made available for those who I saw a number of people joined you know at various points through during the panel so if you're you know you're interested in catching the parts that you missed the recording should be should be available and I'd also like to acknowledge Professor Sumit Gupta who is the organizer of this panel so that again I'll you know thank everybody involved and you know call it I guess call it call an end to the panel thank you all and my last picture and is the questions we didn't get to on my talk send me a note on LinkedIn and I'll answer for you thank thank you very much and I do apologize to those whose questions I couldn't I think I asked asked most of the questions if not all that are relevant to the panel there were a few good questions maybe more specific to the talk perfect so that's for anybody who asked a question that I ended up not asking please post them directly to Dr. Weldon on his LinkedIn thank you thanks everyone thank you