ML4AU Meeting (August 2026): Quantum Technology and Its Impact on AI and Machine Learning

Moji will then MoJi will then do a more of a deep dive uh and then of course we'll open the floor. I mean this is a community of practice. So this is not about necessarily us getting on our high horses and talking about things. This is a you know engagement for the community as a community of practice. Uh uh it'll be really good to have some some discussions as well. Uh next slide please. Uh so just um before we get to maybe I mixed up the slides I might do acknowledgement of country. Um I would like to acknowledge the traditional owners on the lands that uh I'm on. I mean I'm actually in Melbourne so it's um not the my home community is the um is different but here in Melbourne I'm in the suburb of called Warren and I acknowledge the Banarang and the Wanjuri people and of the cooland nation and pay my respects to elders past present and emerging and certainly regardless of what technology we're talking about uh obviously myself and many of you in the room are I'm sure passionate advocates for having a material impact in closing the gap for our first nations people. So just really quickly for those those of you don't know um CUF is a not for-p profofit organization and we are obviously looking at cutting edge and data and digital capabilities. We do a lot of training and we're hoping to build you know contribute to the next um you know the emerging future researchers and innovators in Australia and beyond and also we play a role in in catalyzing a lot of partnerships both within Australia and with international peers. The reality is QF could never do any of this without many many partners and collaborators and and that is reflected in some of the work that we'll talk about today. um especially what MoJi is going to talk about. We do work with hundreds of projects and and that's super exciting and and really rewarding u but obviously like many of us we are we would love to do more but constrained by Australia's funding and research and development environment. Next slide please. Yeah. Thanks. So just a little bit more of a dive. So these are the kind of different areas that QF is involved in. Um and you see that the quantum and AI we've actually deliberately lumped it together and that's because most of this kind of area is looked after by MoJi with the help of many of colleagues both within and outside of CUSF. Um but it is certainly an area that we're um working in both with university academic partners and hopefully also moving into um some of our partnerships with for example with Scantum. Uh next slide please Moy. Uh here are some of the partners and you know a quick slide just to say that we don't do this alone. Um we work with many of the partners including of course ARDC uh and many of the anchors capabilities as well. Next slide. All right. So first of all I think it's really important to understand when we talk about quantum you know from a government policy point of view this is not about necessarily just about quantum computing. what's really talking about well what we're really talking about here is building a new industry for Australia and you know uh people have talked about they're quite a heavy investment especially considering our constrained funding environment that for the sai quantum to get them onshore in in Queensland and Brisbane specifically it's heading into $1 billion right to get them here uh uh to set up shop manufacturing essentially within within Queensland. Moreover, the Queensland government has contributed to that and also most recently I think a couple well actually on Wednesday uh the Queensland government's investment uh with the University of Queensland and many partners is kind of opening up the very first kind of open access quantum test bed for example. Um so the huge amount of investment but why why the why Australia and from a commonwealth and state level investment why we're doing this so the reality is Australia as a economy we are a one trick pony you know we are essentially a resources sectorbased uh economy uh there are some exceptions for example international students and what education that we export thought, but predominantly we are a onetrick pony. Like we put all our eggs in one basket if you like. Um things that we dig out of the ground, some would argue and for those who pay attention to politics will know that you know we even within that sector are probably not deriving the full value of what the country could garner. I mean for those of you followed Dave PCO will know about conversations around taxing gas for example the export of gas and so on but more broadly now when you think about quantum what we're really talking about is bringing a new industry where Australia can immediately derive some dividends by being first in a global market context. Of course we'll always compete with the other more um you know other developed countries in Europe and in in the US but in saying that we have punched above our weight when it comes to quantum technologies and quantum computing. So the idea really here is if cy quantum is considered sort of the anchor you want to build a whole community of smallmedium enterprises around that quantum industry but it's not enough to think about what we where we've come from which is really been driven by passionate advocates like Kathy Foley the former chief scientists around quantum opportunities but when you're thinking about an industry and well beyond research that it can't just be about quantum physicists that that in order to create an industry you need lots of skilled and varied highly skilled workforce uh to be actually you know to build an industry around this and this is again it's it's it's not just about the research and academia the PhD graduates this is also about all the other ancillary workforce skilled workforce you need to create industry. And this could include things like, you know, people who look after cryogenics in a very cryogenics in a quantum computing context is very different to people going and servicing domestic or commercial air conditioning. But do we have that workforce to do that? uh more applicable is even think well not sorry not more applicable but similarly you know how do we you know quantum computer that has sufficient cqits might be a little bit you know further away than we would like but at the same time if you can't code in a particular way and think about some of the applications and doing some prototyping u around what quantum could do into the future we unless we train our new graduates about coding in sort of in quantum environments whether even if using kind of quantum emulators then who cares that we have quantum computers if there's no applications right so I think we got to start thinking about those applications now uh rather than waiting for the the the you know thousand cqit or beyond uh computing capabilities I think I won't um talk too much about the the AI and quantum convergence. I'll leave that to Moji. But that is also a immediate opportunity right now that we can explore uh and are exploring to to make quantum real. Next slide please Margie. So just focusing on the skilled workforce. Um we do have a a project with the Queensland government that um Danny Melanchello is leading from Cusif. uh it's about what we we've called it kind of catalyzing vocational and higher education skills continuum for quantum career pathways or in another words this is actually what it's saying is that can we think about a workforce for quantum that is more than uh just about higher education and universities but it's a continuum of also in and inclusive of also the TA uh students as well and their skill sets um so this project was funded by the Queensland government. Uh yeah, next slide, Moji. Thanks. Um and the project's objectives is really we want look you think it's difficult to change courses and course content in universities is even more difficult in the T sector in the vocational sector. So what we try to do is in this project, it's a three-year project funded by the department of environment, let's see, science, tourism, and innovation or any combination of those words. Um, and what we're trying to do is to bring and expose uh TA Queensland students from different programs into quantum opportunities. And you got to remember a lot of the quantum opportunities are not just about quantum computing. Quantum technologies include sensors. um uh the um sensors, optics, many other facets of quantum technologies are actually uh are actually happening right now and a lot of the research infrastructure facilities uh like the Australian nanop fabrication facilities in the encro are large part of that that equation. So these are sort of our objectives but at the same time we were kind of learning on the go because we didn't quite understand how the quaint the vocational sector works. Uh so there was a lot of learnings through uh through this project and we're only really into year uh the second year within a three-year kind of program. Next slide please Margie. So one of the things pilot programs we did was a very limited number of students but we wanted to understand what would it take to take uh take TA Queensland students through a really intensive program uh in in if you like in the the flavor of quantum technology. So the natural partner here was the um Australian nanop fabrication facilities specifically the Queensland node and we did sort of a six week program where they were thrown into the mix of some of the quantum technology. Don't forget also in most cases the TA student cohorts are u you know many of them are 15 16 year olds. So this is really early exposure to to quantum technologies. [snorts] Um we went through this program and now we learned a lot from this this initial kind of pilot program. Next slide please Moji. So our next um approach to this is a tiered model and this is what we're working through and in some cases implementing um because to do a completely immersive uh option is really uh costly first of all not in terms of necessarily dollars per se but time and effort it takes to do it and there are also many kind of um uh challenges that have to be faced things like work cover uh things that you don't really think about but are real uh considerations. Work cover, insurance, liability, all these kind of things became uh things that we needed to solve through that project as we were delivering it. So we come up with a multi-tered approach which uh we're in the middle of deploying which is everything from kind of low intensity light touch supervision and and and kind of handholded demonstrations and exposure to shadowing to all the way to tier three like what we did with ANF but we wanted to have a more tiered approach so that uh we could cover a larger cohort of TA Queensland students uh to give them experience in quantum technologies. And the reason why we're doing it, if I go back to the project exposure, is that we want TA Queensland and other vocational provisors to think about that future. Um because again I I go to the point what we don't want to do as a country is invest a billion dollars in bringing an anchor tenant and the anchor tenant saying in 5 years time we're ready for our computer but you haven't provide us with the workforce to keep an industry going. We're moving offshore. That is the risk, the real risk that we want to try to mitigate now and build a pipeline of talent now so that when the computers are ready that we actually have a a a workforce that can meet the demand both technically and in terms of capacity of enough people being able to deliver on the quantum promise. Next slide please Maji. Uh a little bit more details about the program itself. So we've got kind of two pathways. Uh, one area we're looking at is more general lab skills and lab techniques, certificate three and four, but also we have a more specific targeted uh, uh, quantum area around cryptography that MoJi will touch on. Um, and he'll he'll highlight why that's a priority area u, especially in the fintech and the financial sector. So, it's a more targeted approach. So it's very much in the data and digital space and cryptography and uh that we're we're exploring as well. Next slide. And of course finally to round things off this is all all has to be connected back to national policy. Uh for those of you I'm sure all of you if you if you're not aware should know about the ambitious Australia uh report and and and it's now implementation. Obviously the the whole idea around vocational to industry workforce model is captured within ambitious Australia. So very much our work here is very much aligned with that. Uh of course um you know the ambitious Australia is only just starting to be implemented the recommendations that are coming out of it. But of course it's always nice to have programs running that are aligned with with national policy. I'm we'll pause there. So this is a really high level kind of snapshot about the macro view of quantum technologies and and uh where the state of affairs if you like at at a very high level. I will pause there and open the floor up if people have questions or comments for that matter because people have different opinions of about the prospects of quantum and quantum technologies or quantum computing specifically. So, I'll pause there and open the floor. If there are no questions or comments, of course, we can move on. Sorry, I can't see everybody. Is that >> Hi, Sach Alan here from ARDC. Um, so great to see to hear about this project. Um [clears throat] the I wondered Sach if you could comment on uh quantum literacy starting in primary school and [laughter] uh and and whether QIF has been doing any work in that space or are you aware of um other partners um uh I'm very familiar with um uh you know when Queensland's quantum strategy was released you know seeing the the workforce capability element embedded in that was really exciting. Um yeah any >> yeah look I think the the the Queensland government's program is actually there's a the broad program name is called the quantum academy uh and I think within the academy there is certainly school secondary schoolbased um uh uh if you like again expo early exposure to quantum technology so it is in fact going right into secondary school at least I'm not sure about primary school but secondary school certainly so I think the government certainly the Queensland government's of the view that again I think they are acutely cognizant of the risk that if they don't have this workforce ready you know side quantum and for that matter any other um supportingmemes around that anchor tenant could disappear if the workforce isn't there because that's a really easy excuse for a big multinational company to say you don't have the workforce We're off. >> Yeah. And and such are there any additional learnings uh that we could take from you know the success of the biotechnology um industry and investment that we've had in Queensland and I guess any learnings that we could transfer. Yeah, look, it's interesting, isn't it? Because Queensland made huge uh step changes with all the capital investments that were made through the smart state program. And I think a lot of the biotechnology industry was built on the back of that heavy investment. And I guess Queensland was also um strategic and you know little bit of luck as well in that Chuck Feny came to the party with Atlantic philanthropy and donated huge amounts of money to the Queensland effort. So it was it was a little bit of good strategic insights and and clever investments. But you know I can only speak to even like when I worked for Qout if you think about Qout's journey with the Chuck Feny investment you know it went from a university that just became a university from a technical college uh with a very modest research um profile to now you know $110 million heric return for example but that kind of jumping from like 3040 million million to 100 plus wouldn't have happened without that investment and that goes beyond biotechnology of course but it's just case in point about you can build it they will come but you have to also then wrap around that support structures and some of the support structures in that particular example was investing in research infrastructure actually uh and that research infrastructure included things like you know the equipment and the but also the people. It was a very deliberate strategy to recruit research infrastructure people as well you know people who are experts in electron microscopy people who are experts in you know and you know genomics and omix and for example so obviously things like omix and omix capabilities were instrumental for the biotechnology industry and some of the successes of microbia and and others in Queensland. Yeah. Yeah, and it's exciting. Final comment, exciting to see where the intersection between uh our burgeoning quantum ecosystem and the biotech uh industry is going to converge. Yeah. Anyway, enough from me. [laughter] >> Who is next? Sorry, there's a few hands up. >> We've got um Jack for you. >> Go ahead, Jack. >> Hi there. Thank you very much for that. Very exciting and and very interesting. Um I guess my um question is um similar to um Illan's which is um uh in addition to sort of training up the workforce um from a certain level say uh let's say TA um uh uh now um what are thoughts on people transitioning from other industries? So I Australia has a very educated population in all kinds of things but not necessar I would say quantum physics is you know one of the rarer things out there. Um is there some sort of thought on people retraining into it if they're already let's say masters in something else postgraduate in something else? >> Yeah interesting question. Um I think part of the challenge of course is that you know if you think about sort of quantum theory and quantum physics so this is more direct roles in the quantum technology industry I mean Australia has to be a little bit careful because there's so many people moving away from fundamental mathematics for example right and and we have to deal with that at school age u so in some ways is the retraining highly educated and retraining is feasible but it needs to be underpinned by kind of a good understanding of the mathematics the fundamentals. Uh on the other side of course the what I call the kind of incillary support so people who do really sophisticated welding and people who do cryogenics people who do uh you know when I say plumbing I'm talking about different types of plumbing. So if you think about the traditional trady path in TA they can certainly be retrained to deal with high pressure vacuum and more you know to support an industry. They are definitely ready to be retrained or given the opportunity to be specialist train you know be specialists in in kind of quantum related technologies because you know most of the HVAC TA courses are all geared towards domestic and commercial markets but if we have an industry in quantum they could specialize in dealing with you know hydrogen and and so you know helium cooling and blah blah blah. So there is definitely potential for retraining there. But I think in terms of the you know quantum physics kind of related jobs, we probably do have to deal with some of the skills gaps in mathematics that the country seems to be facing at the moment. >> Yeah, super interesting. Thank you. >> All right. Thanks Nana. I think it was Nana next was it? >> Yeah, it's Nana. >> Sure. >> Then we might wrap up and um continue on with the next part of the presentation after that. Nana. >> Sure. So I would uh I mean this is an exciting topic and I would like your take on um uh say uh I mean uh the uncertainty surrounding this that is um while quantum offers long-term promises but at the same time it is both timing and and the scope uh remain uncertain and how should you approach this and um so I mean as such we know um excuse me for saying this Australia has always been uh risk averse in its investment technological investments and that too if it gets burnt a bit early then it shies away for for too too long and um so is there a measured approach to this >> look no I think I Again, digress a little bit. We do have problems with our political systems, the Westminister system of two parties, threeear cycles and you know quite fickle in their >> policym that's let's put that to the side and I did notice that Kathy Foley did um publish something recently in I think innovation Australia about saying you know hold tight just be patient around quantum. So I think the fact that she had to say something like that tells me that you know there are naysayers and people who might think this is a waste of time. Let's be honest there are there is certain parts of the >> um certain parts of the uh ecosystem who have doubts about quantum's utility moving forward. But I would also say that people can learn from history, right? I mean AI has been around for many many years, >> decades and decades. And it didn't necessarily deliver on its promise then. There was a bit of a pause and maybe a few backward steps, but when the hardware caught up, all of a sudden it all accelerated really rapidly. So we shouldn't lose sight of history lessons from history that sometimes there are pauses there are you know you might have to wait but doesn't mean all the work that was done was wasted it just came to fruition later so we have to be careful that um I think the only thing we can do now to answer your question is we should remind people of that historical context that there are industries that are booming now that had little bit of hiccups along the Right. >> Yes. >> Thank you. >> Yeah. So, yeah. Thank Thank you. I mean, yeah, that is keeping that in mind. It's a it's a wise advice. Yeah. Thank you. >> Um, okay. I think I'll hand over to Moji now. Thanks, Mi. [clears throat] >> Sorry. Thanks, Ash for the great overview. I might have to share again because it's for me it came out of full screen which is a bit distracting. Quickly try to do that. Sorry. All right. Yep. So, um Sash had for us um a great overview of what's happening in Australia, what we need to support that industry. I go a little bit more into the details uh with technical outcome reporting, not uh technical method uh reporting uh to show you like where there are promises and where this going. get to my own slide. A tiny bit about myself. Um I did my PhD at Griffith University was um experimental quantum computing with Trapion. We did it with Georgia Tech in the US and Franhoffer in in Germany. Um then um I also worked as a senior data science/posttock at uh UKQ applying AI to quantum machines and applying AI to health u data sets um a bit of detour in Queensland Treasury emission modeling and then it's been a great year at QF as a head of AI and quantum algorithm and um yeah we are trying to see where it really can both AI and quantum can help researchers and create impact. I also run this meetup uh quantum computing Australia meetup. Uh it's a meetup of website and um QF is now a sponsor and we are doing joint meetup with quantum Australia. It is live streams. We will we have a few meetings a year and then we summarize where the quantum industry is now and where are the applications. So it's for non basic mainly for non-quantum experts um and focusing on application. So if you like to you can stay in conversation you that's one way of joining. [clears throat] Okay. Um this is how I arrange my talk. You start from where quantum computing is today. Uh and then uh we focus a bit on quantum machine learning. Um where is evidence where is promise um and a bit better focus on where the signal is real means that you we are seeing a difference actually rather than pipe. Uh then we go to quantum security um postquantum cryptography to be exact. Um and a bit of a detour to a relevant project that we are doing federated learning and and I tell you how it is relevant to postquantum cryptography um and then in general where Australia is heading. Okay, where are we today? So I don't have much time. I usually start with what is quantum computing. Let me give you an introduction 2 minutes before I jump into this. So you have normal computers zero and one you do like adding multiplying and differencing on top of on a CPU. On top of that you can build very complicated algorithms. Um and you can do great things with quantum computing. You are actually saving that data on quantum objects. This can be a photon, this can be an atom, something like that. And when you do that, laws of quantum mechanics lets you have them in in between zero and one and then match them, pair them with other bits which gives you it is called entanglement. And these two capabilities can give you a huge space, a lot of numbers with like very complicated ways of manipulating them. And it can be harnessed to do things that no supercomputers can do. But what do we have now? Let's say for example IBM has like 400 cubits also 1,00 cubit um superconducting quantum computer. This 400 cubit is working better than 1,000 because the noise goes so high that um the number of faults goes up and then it cannot do better than the 400 one. So that's the problem we need. We we have a lot of noise. We need to fix that. That's mainly a technical challenge rather than basic scientific challenge but it's a very hard technical challenge. Um and when we reduce that noise like noise meaning you try to save some number or do some kind of operation it doesn't do what you want. Uh if we reduce that to like 0001 let's say then we can uh add to that fault tolerance which is you use a lot more number of cubits and then if some of them fail you can compensate for that with the extra ones that you have. Meaning from the outside you have you can simulate perfect quantum operations. We don't have that yet. Scantum is heading for that road their road map 5 years I would say like seven 8 n years based on my own experience but we get there. It's again it's a how much money you put on it how many people you put on it. It's a technical challenge rather than a basic uh theoretical hurdle that you can never pass. And another problem data loading uh you have some numbers on a normal computers you want to put it just put it on a quantum computer to start operation. It can be really slow. um people working on that as well and for now advantage is domain specific not even domain specific when I say domain I don't mean like broader domain it is being triled in transport but in many cases it still cannot do anything in some very narrow problems specific problems people has shown that um this this can shows results on smaller data sets that kind of proves that if we can apply it to the larger data set, it will do much better than the uh supercomputers. What is why can't apply apply it to bigger data sets? Because you need a bigger quantum computers. You need like a 1 million cubit let's say something like that that again quantum is aiming for directly. But at this point numbers are low, noise is high. So a lot of work is like simulation and that basically with that you can go like 30 cubits means that if you're doing buses scheduling you can do 30 buses not 150 that you really need for Brisbane let's say uh yeah so uh where's promise and uh yeah where is the actual outcome >> [gasps] >> I need to make this clear that a few first few slides that you see might come across as very negative. What they're doing is dehyping what we have now. So it's not at this point we don't have uh algorithms that can do so good in very practical applications that beat supercomputers. That's the only thing that these slides are uh conveying but it doesn't mean that we cannot do that in 5 years or 10 years or something like that. It it is waiting for the hardware mainly and also algorithms people are hardly working on making new algorithm making the current algorithms better. So the theory for many many applications is already there we just need better devices and some others we need to create better algorithms as well. Now um one result for example showed uh a peer-reviewed paper show that sorry a review paper showed that in many of peer reviewed papers on quantum algorithms a simple logistic regression which is basically sort of the simplest model that that we have classically uh three out of five they can easily beat beat a quantum uh machine learning model. Uh and then if you basically sit on it and use much more complex one, you can beat every in every case you can beat a quantum algorithm except for for some narrow cases. Now it will be a lot better in the future but still now also I will show you some result where quantum computer is beating classical. It's narrow but it's we have it today as well. Uh so there are things that that are working today very good and practically I wanted to mention that in terms of like having a better assessment of fairer assessment of where quantum technology is. First of all quantum technology is not all quantum computing. Everything that use laws of quantum physics is a quantum technology. For example, quantum s sensors made by cyro. It's like at least for a decade even more that are being using by m mining companies. That's the quantum technology mature being sold uh solving a lot of problems. Um for example and uh using machine learning on top of quantum simulation to extract information for things like drug discovery. it is being explored today and showing some results there. Uh some very specific mathematical problems that has some this is like a mathematical term group symmetries but what it says that very specific problems that can be helpful for some let's say optimization tasks in those specific one quantum computers um are showing better result than than the classical one. What is there are some things that are not really happening soon by soon I mean next year but it might happen like five or 10 years quantum language models this LLM AI that we have we can't still put them on a quantum computer again it's small we don't have exactly figured out how to harness um its full power and things like that so still we better having it on normal computer molecular and material properties is very close very good algorithms there but still needs a bigger devices to do on do it on big enough molecules to be impactful. Uh for example beating gradient boost boosted trees on tabular data is something that they haven't done yet but it is very likely that they can do it. Uh and yeah broad quantum advantage claims we don't have it yet. Um it's it's a very powerful machine. You have 100 cubits. You have two to the power of 100 numbers to save in there. That's huge. No, you cannot build a RAM on a normal computers that can save that. But also harnessing that is a huge challenge. So for sure it breaks many paradigms that we have here but there are challenges in harnessing that today but yeah very hopeful that we do it. So um I share a peerreview paper this was done by my previous group at UKQ. Professor Sally Shrapnel was my supervisor there. She's a She has a PhD in quantum computing also a medical doctor. So best combination of skills one should have for applications of quantum in in digital health. Um so what they did they looked at 169 paper 82% use ideal uh situations to evaluate their result. only 16 um looked at noisy conditions. Noisy conditions means the current device that we have. Now it doesn't mean that the ones that looked at the ideal ones aren't useful because when we have fall tolerance we have ideal but it's not here today. So it's two kind of work. One is you are simulating it uh and showing that if I have ideal machine I get these kind of results. Now we need to wait for the actual machine to do practical one. The other one trying to harness something from what we have today. And between those uh none of them really definitively showed uh better result than the classical one. when you factor in every realities that that you have like for example uh limitations number of cubits and the noise and everything and there are some other technical problems that I want to go through but these things all need like some classical optimization behind it and unlike neural networks that we have like kind of back propagation for those of you who are technical in machine learning we don't have a back propagation for um quantum algorithms. So it makes it really hard to tune those things to do a great job but people are working on it and I'm very hopeful that very soon we can solve that. one result that shows promise today that's one um quantum enhance stratification of breast cancers. This was a paper in quantum machine intelligence on this data set uh metabric which has 1,980 samples of multiomics breast cancer patients and then they did clustering like dividing that into different types of patients that you have there. So what they showed that is that if you want to match the expert clustering like dividing those patients into groups that expert recognize as uh separate specific group groups uh if you try classic classical kernels RBF kernels we need 500 uh samples to reach 0.9 matching um expert clustering but quantum kernel only needed 100. So this is something that we are focusing on now as well and when you don't have many data points but you have a highdimensional data this is a territory that almost every classical uh machine learning method including neural nets um and also the more classical mathematical statistical based one um hit hit some problems and yeah quantum algorithm are showing some promise there but some this is like five time less data points so there are yeah many experiments in OMIX that you only have limited number of patients if you want to learn from that m neural net needs like millions of data points at least 10,000 I don't know 100,000 preferably million but you might only have like 100 or 200 or 500. So usually we fall back from those two more statistical ones but now there is a second fallback that is better than even the statistical classical ones and that's quantum kernels. So we are exploring that area. Uh it is in the initial steps where we are doing there but with University of Melbourne we are exploring where we can apply this uh quantum machine learning methods uh into OMIX data and get some result there. uh uh not exactly machine learning but uh but a relevant applications that we are working on with Qout funded by Queensland government. We are working on quantum optimization for traffic. We specifically this is for uh Queensland 2032 games. Um and we are working on currently bus electric bus scheduling uh for the games. So realistic routes that we have in Brisbane and then finding [snorts] the best way of sending these buses to different routes uh while considering for example the charging constraints and everything and there is a huge big machinery of optimization multi-steps iterations and we are working on replacing one component with quantum um optimization algorithm and keep the rest normal and overall get get uh some some better results there. Um so this is yeah one of other applications of quantum computers is optimization and a lot of financial companies and also transport companies are playing around with this. Again we are limited to the number of cubits or a specific algorithm uh needs the same number of cubits for the same number of buses and we are now simulating 30 to 40 cubits that's the number of buses that we can do ideally we want 150 ideal cubits to do 150 buses and then when you get to those numbers 200 300 buses then no supercomputer can beat you because this is exponentially what they need to compete with a quantum computer in that arena. But at this point with that 40, we can still somehow do it do it on a supercomput. >> How much time? Yeah, still have a little bit of time. [gasps] Uh quantum security. So I need to clarify this quantum security one quantum security is that you for example send data using like photons in a quantum state to make sure that it's secure. This is not where what I am going to focus on now in this this one. This is what uh focusing on what quantum computers break if you have big big ones. and then what do we need to do about it? So one one of the first very uh efficient uh quantum algorithm is called shore algorithm. What it can do is if you multiply two very big odd numbers, it can tell you what were those numbers in an efficient way. Normally a normal computer needs to do a lot of trials and with the 400 um digit numbers that we use it it will take them a millennia to tell you what those numbers were and if they tell you then yes you can break um security on the internet means like encryption when you encode a message send it over internet to make sure that nobody can reads it when they look at it it's just like a jumble of random numbers for them unless you have a key to decrypt it back. But what they can we don't have big enough quantum computers to break it today. But what can be done is that people can save those communications and then when they have a big quantum computer in five or 10 years start decryting it. So it it is a it becomes a reality in a way in five or 10 years but also it's a reality today because if you have a sensitive data is health data, financial sector data some of them you don't want to be uh exposed even 10 years from now. So you need to switch to something that quantum computer cannot break and those exist specifically some standard uh ones by NIST in the US that are most people are focusing on to switch to and what they do is they use different kinds of encryptions that even quantum computers cannot break and uh for sensitive areas we need to start like switching to those which is a huge project very costly it is projected that US federal ederal agencies will be spending 7 billion to move from current encryptions to yeah postquantum cryptography. Um so uh like sash mentioned this is another area that we are working with with TA um and uh some other national and state or organization to um create awareness and upskill people so that they are um they have the skills to help this transition. Uh one other project that we are working is a side project um to federated learning projects. Federated learning is when you have sensitive data let's say at hospitals you don't want to collect them all in one place because it's um because of the privacy concerns. Um what you can do to train a model on them is that you send a copy of your model to all of them where they are in the hospital. Let's say they learn a little bit from them. They come back. You do some kind of averaging, copy again, send back and do it 100 times. And then you learn from all of them without moving that data or without the researchers who built the model u seeing the data that they are training on. And what happens here is that the patient data even the like encrypted one is not moving across nodes. So we might think that if you encrypt it and send it over it's safe. Nobody can see it. Uh only the researcher is the is the threat here. But with this post uh with this um quantum era threat even if you encrypt it and send it is still vulnerable in the future might be saved and used. So this is one way that federated learning adds to the security of patient data and not the only way. It doesn't also expose it expose it to to the researcher and it's not collected in one place so that one hacker can hack one computer and get access to patient data from 10 hospitals let's say and yeah many other ways that it makes it a lot better for sensitive data. Yeah, these are the things that we are working on. Quantum transport, uh workforce readiness and national investment in accelerating to. So most of you know about SA quantum but more recently uh Silicon Quantum computing received like 60 million um to expand expand their work. They're in Sydney. So it is a area that is growing and it needs a little bit of patience like AI. I mean neural net we had it like a few decades ago but for it to become what's today even five years ago people were skeptical if they can change anything in practice. there were like trials on some um u data let's say medical data which was promising and things like that but this huge paradigm change by LLMs when we had like big enough machines to run big enough models uh suddenly changed everything is creating something that be became so much efficient assistance that we cannot ignore anymore more so yeah it's quantum technology like specifically quantum computing not technology the technology in general it is already we have mature examples but quantum computing um still is in like 1970 80 era of AI um but with so much money that's governments and um private private companies are pouring into this. It will evolve a lot faster than um than than a few decades. It will be in five or 10 years that we will see some big changes coming from this. But that said, it's not as widespread the effect you basically it it is good as some algorithm but not all. So it will work alongside normal computers. There are things that still normal computers do do better. most of the things and then if there's a specific problem like optimization or drug simulation or something like that you send it over to your quantum processor get the result and continue with your normal yeah sort of covered this before getting into this uh yeah that was a snapshot of what CUsef is focused on and in more generally what [sighs] where the sector is and where Australia is heading but now I guess it's a good time for questions Nana. >> Yeah. So um yeah thank you. Um a couple of questions I mean the first one is around the reproducibility. Of course, there is reproducibility crisis everywhere in science and I mean let's but my question to you is is there something inherently limiting about quantum level reproducibility of computations and um I uncertaintity principle or whatever that might be playing a role in it. >> Uh yes and no. uh by that what I mean is there is this thing that when you encode a cubit somewhere between 0ero and one and then you sum operation it ends up somewhere else each one of them end up in somewhere else in between zero and one but when you actually look at them each one of them either they tell you I'm one or zero if it's like 20% state is 20% one and 80% zero and you look at there is a 80% chance chance that it tells you I am zero and and becomes zero and stays zero um comes out of that superposition and 20% chance that it tells you I'm one um so if you're if you're measuring at the end each time you get you might get a different answer and for a lot of algorithms um the solution is you need to run it multiple times like let's say 1,000 times and then look at statistic to see exactly each c where each cubit is and then the more number of runs that you do you yeah it gets better. So that final statistical answer is fixed and reproducible only if you want if if you want more certainty you need to run it more but >> sure >> the intermediate results are a bit random. >> So I mean the the distribution is stable but >> distribution is stable >> not a not a sample. >> Um yeah so it certainly certainly makes sense. Um yeah. So um so we you any algorithm you would um deise should not be dependent on the intermediate samples or distributions. And >> no yeah yeah what what you do is um you you start from your like for example let's say quantum ML you input with a usually a fixed encoding uh translate that uh number zero and one and something into something quantum and then you have u a kind of algorithms that is tunable you need to train it and tune it but after training and finding the good parameters you keep the parameter fixed and each time you send in the same data you get the same kind of distribution at the end. So uh yeah it is in that sense it is reproducible. >> Sure. Sure. And um and the other one uh of course is related. You already said once you observe it it crystallizes into uh some some state >> and um so is there an application for quantum memory? Of course it is. It will be unstable per se but um on the is there a way is it being investigated? Yes, it is being investigated. Uh and it it will be very useful because sometime you want to for example delay your actual quantum bit that is doing the computing might not be that stable. It might can keep there for 1 second and sometimes you want to be able to keep it without looking at it. keep the same superposition uh for let's say uh five minutes until another calculation is done and then you come back to this and and use it. uh one thing that is directly people are doing is quantum encryption when you uh send data uh on using photons in quantum state for example um because you lose photon in in a fiber after um I don't know 60 100 kilometers you need to you have a repeater and quantum transmission cannot actually have a repeater so you need you need to generate them and entangle them and in there people are working on like quantum memories to enhance that the distance that we can do with like quantum repeaters that's one area but in general yeah people working on quantum memory more stable yeah quantum memories that you can save a quantum state in it and use it later >> thank you thank you and s is a great Um, great topic and you you you made it very accessible. Thank you. >> Thanks, Nana. There's a couple of questions in the chat. Moji. >> Okay, let me have a look. [clears throat] What's the optimal approach for integrating classical machine learning and neural networks with quantum computing considering um data set size, computational complexity and model performance and available quantum resources. Yeah, great points Saha. Um we can't we can mix them. One one way we can do this is for example let's say I have a small quantum machine it can only work on uh let's say 50 features uh but my actual data set has like 10,000 features. One thing you can do is use your neural net as a feature translation from a high dimensional to a lower dimensional, then feed that into a quantum kernel and do for example clustering with that. Whether or not it's better than just leaving the clustering to to uh neural net, it's kind of an open question. um we haven't proved that this combination can do better. Uh it's it's a relatively new things. Quantum machine learning even compared to quantum optimization or cryptography is a lot older in the ' 90s and things like that. But people are working on it. But yeah, that's one way we can mix and match neural net with like quantum kernels. Um I think there's another question from Muhammad as well. Muhammad did you want to ask this one? >> Yeah, thanks Moi and Sasha. I think this is a really interesting topic and quite complex in my understanding the quantum um theory. So I think um as you mentioned MoJi in one of your slides about the test bed and roll out to the operational environment. So it's it's just like a kind of um my impression that the the roll out of the test beds will be quite crucial to have that lens of how these concepts will look like in the early stages and um giving that community more confidence on how we can roll out these early stage and early phase ideas into a more operational environment. So I think my my point was more about um the test beds that we have like we are aiming to have in Australia and one of the test bed that we have in EU I think in previous weeks um their role would be quite um critical I guess. >> Yes. Yes they are uh it is very important those kind of infrastructure because you can directly work with the real system. It is very limited the first version only five cubits but they made it the whole infrastructure that builds the around it those cooling and then translating into like web applications is easily transferable to more cubits and they have a rapid plan of making it more and more but they decided to even with with five they decided to make it available let the researchers play around with it. Five means that you can try a lot less number of samples or smaller mathematical problem. Uh still it might be good for people who wants to look at the role of noise how they can limit that and even for a small one and then expand that to more number of cubits. But yeah, these kind of infrastructure are essential um for content researchers in Australia also those who are in industry and working on like startups to be able to have something real to play with. >> Next question. >> Um there was another question um in the chat as well from Paul. um want to jump in feel free to. Um and he said I would like to ask what specific quantum frameworks and or hybrid algorithms should ECR in biomedical field prioritize learning today to prepare for the quantum era. >> Yep. Yeah. Um so there are two two big like applications optimization and machine learning and for biomedical I think more promising side is is the machine learning side rather than optimization uh for transport and logistic and financial sector it's it's the other one mainly um but where what's quantum algorithm so in general quantum kernel are very promising. And then we have like things that we call quantum neural nets which there is a paper that says that these two are the same quantum kernel and you just approach it in two different ways. So uh underlying thing is the same but if I want to point to one algorithm quantum kernels to learn from data with low low number of data points. Um so you you need to look for where in the in the field there is for example this expensive sequencing happening and then only you could do it on do it and collect it from 100 patients rather than 1 million and then you want to see how we can cluster it efficiently then yeah we can trial things like quantum kernels to see to see how it does But again it's it's a young young field and if you if you like to you can reach out to me through QF website there's a like a contact us and then we can just explore what was possible for your field >> just a followup question >> if you could comment on the vanishing gradient problem with this neural net that Yeah, thanks. >> So each machine learning models there are numbers in there that needs to be tuned to match it to your data. Initially you start with some random numbers for a for let's say for example logistic regression. You have 10 columns in your data. You have 10 numbers and you need to tune those numbers to something that matches your data and it can do classification for you on on your data. Uh no, it's the same in that sense. It's not different. It might have million or billions but still there are some random numbers that you change it and it can do for example cat versus dog classification on on images for that it randomly tells you this is cat this is dog um nothing better than picking random but after you tune those numbers and tuning those numbers lots of numbers if you don't have a good way of knowing which way to change them uh to get that result. It's it's an impossible combinatorial search um for for good numbers and continuous number too. But what what's great about most of these ones including neural network logistic regression and everything else is that you have gradient. It shows you for each specific number that you can tune. It tells you in which direction and with which power you move a little bit to make the result a bit better. And you do it again and again. 100 times, 200, 1,000, all of them will converge on some good numbers that can do cats versus dog for you. Unfortunately, quantum computers because we can't look at the numbers that they currently saved in quantum state in there, we cannot have a gradient for the classical numbers that we have in there that we tune and we control that whole machinery. What we need to do is we need to regenerate that that gradient with like getting two points and get get getting a bit mathematical but getting two points and then look how much it changes between those two points. But it means that if you have 1 million for example parameters, you need to do two mill two million operations to just figure out that gradient to do one step of 1,000 and it's still not as good as analytical gradient. Back propagation is the name of algorithm that does this in neural net. So we don't have that in quantum machine learning. And when the numbers get big, gradients become small, very close to zero. even simulating that becomes a huge challenge. So this is one of the big challenges in both quantum optimization and quantum machine learning that needs to be fixed. But a lot of people are working on it. I'm very hopeful that we can um tame that problem but at this point is one of the huge challenges. Yes. >> Thank you Moji. Um, there was a couple more questions in the chat, but I know that we are at time and I'm not sure if people need to go or if um people want to hang on for a moment or two more. Um, MoJi Sach, do you have a couple more minutes or do you have to rush off at all? >> I'm okay. More than likely, I can't answer anything, but that's okay. Moji can stick around. >> Can Ingred, you had a great question. Are you still online? Yes, you are. Um, did you want to ask that one because there's a little bit of information there, but you can see. >> Sure, sure, sure, sure. Um, it just so happens uh that I've just reviewed, um, an open data journal submission and it touched on this topic. Um, so happy timing. Um, but what I found really curious, um, was about the really small scale of this data set. Um, and I just wanted to understand a little bit more about why this choice. And I think you touched on it previously about um, honing in on some particular features. So I wondered if you could elaborate on that because it was uh the use of kernels and yes um uh that was why I was quite surprised because it was so small and I understand that might be um an element of not wasting compute but I just wondered if you could elaborate a little on that. It is a small because small is where the the great neural networks that we love fail to give us results and normally yeah because they have so many parameters to tune um having so many like low number of data points uh is a challenge for them to deal with. And before quantum computers, we used to use classic machine learning algorithms like kernels, like logistic regression because for example logistic regression there's assumption that the data is linear and therefore it's easier to to tune it and get some result. It won't be as good as if the relationship is not linear as so complicated. It won't be as good as a neural network could what a neural network could reveal. But we can't use neural network anyway because we don't have that many data points. So it struggles to learn. So this is the best that we can get classical limited assumptions. Um and kernels, RBF kernels, classical kernels are one way of not putting too much assumption in there. Some kind of auto automatic translation between between the initial space of u let's say numbers that you have uh for for these patients into something huge dimensional. Goshian makes it huge dimensional and in that huge dimensional even putting a linear divide between them can reveal uh very good patterns. Uh so a competitor now is quantum kernel. You can do it with a lot less cubits uh very low number of cubits and apparently with result of that paper uh you have you need five times less number of data points to get the same kind of clustering accuracy compared to our RBF kernels. So it is the matter of showing that yeah these are data efficient that's why it a low number of data points becomes relevant and that yeah that's why we that paper focus on that specific one and yeah near term we really need to look for those kind of applications to show something but in the future when we have enough number of cubits we can try these is against neural nets on big big problems as well. Lots of columns, lots of data points at this point. They are not big enough so that we can translate a huge data set into something that or a small quantum computers or the simulated ones on on supercomputers can handle. >> That's great because I think there's a clear rule of thumb that people are working with there. Um and it it also I think happily for those who are new to this actually helps people think about breaking down the problem like where your most um pertinent challenges are and this paper certainly touched on that. So I think it was there was a lot that was implied um in that um and unpacking that for people who are trying to understand choices for example I think is really really powerful. So thank you for the answer. That was great. >> Sure. Sure. And in case in the future something come up that might need like collaboration with QF uh I just posted or the link to our website and just if you get in contact we we love to have a chat about these kind of projects.