Quantum Error Mitigation using Mitiq with Misty Wahl
Download MP3Kevin Rowney 0:30
Welcome back, today we have on the podcast Misty Wahl, who is a member of the technical staff at the Unitary Fund. Before that, she worked in the rather significant topic of extreme ultraviolet lithography at ASML. She is a primary contributor on the Mitiq project up on GitHub, which is a package for error mitigation for quantum computers. Really interesting project.
Sebastian Hassinger 0:55
Yeah, it's fantastic. I'm looking forward to talking to Misty, I've been tracking the Mitiq project from the very beginning. In 20, I think 2019 It was brought to us at IBM and I happen to be working for IBM Quantum at the time, by Will Zeng and Nathan Shammah, who were the founders of the unitary fund. We thought it was a really interesting project at the time, and it's now quite, it's matured quite a bit and made a lot of progress. They've been releasing papers, research papers on the Arxiv, based on using Mitiq for error mitigation, and getting some really interesting results. And, and more importantly, you know, I think as a as a phenomenon Mitiq, and Unitary Fund represents sort of maturation of quantum computing as a field. And, you know, we've talked many times about how quantum computing is coming out of the physical sciences, quantum mechanics, quantum physics, quantum information theory. And it still is, you know, has very rooted in fundamental science and basic research, but is increasingly moving towards being a technology that requires engineering discipline, and, and sort of the kind of measured progress that you see in emerging technology and other fields of classical technologies. And I think Mitiq and the Unitary Fund, and the work that Misty is doing, are really good examples to see exactly how that dynamic is playing out today, and what the advantages are of that kind of approach?
Welcome back, we're joined today by a very special guest, we're excited to have her with us Misty Wahl. She's a member of the technical staff at Unitary Fund, which is an open source quantum software foundation, they do micro grants. And also, they are leading the development of a project called Mitiq, which is a framework for error mitigation of quantum calculation, quantum computation. And Misty, very good to have you with us. Thank you for joining us. And welcome.
Misty Wahl 3:31
Thanks for having me.
Kevin Rowney 3:32
Misty. Yeah. Good to have you on.
Sebastian Hassinger 3:34
So if you could start off Misty with just some background as to you know, what, what is a member of technical staff at Unitary Fund? What is your role? And how did you get to that role? I think that, you know, we've said it many, many times before, everybody has a very unique path to into the quantum computing field. We always love to hear this sort of personal stories.
Misty Wahl 3:57
Sure. So yeah, well, I in particular, do a combination of software development, quantum software development, and also research in quantum error mitigation. But yeah, we do a lot of different things. Some of our staff members emphasize other aspects of research and quantum computing. But the common thread is open quantum software, open source, and also community building and education initiatives in quantum,
Sebastian Hassinger 4:31
awesome. So what was your background Misty? Did you study quantum physics you come more through the computer science direction or
Misty Wahl 4:40
non traditional background? So my degrees are in mechanical engineering. I started out as a mechanical engineer, I worked for a semiconductor equipment company called ASML. And I was there for nine years I started out as an engineer, but then I moved into project management that I kind of from hardware, I went into software project management. And I was kind of at a crossroads in my career. And figuring out what I wanted to do next, I knew I wanted to become more of an individual contributor. And I was trying to figure out what I wanted to emphasize what I really wanted to study. And I heard a talk from IBM on quantum machine learning. And I realized, okay, that's the grand challenge that I want to pursue. And next, and I wasn't really sure how to get in. So I was doing a lot of like self study doing different courses online, I went through the MITxPro certificate program that quite a few people who have kind of followed this non traditional path field have done. And that was like, during the pandemic, early 2021. And after that, I wanted to continue my learning journey. And so I heard about Unitary Hack. And I signed up for that. And I ended up contributing to Mitiq through that, and learning about quantum error mitigation, and learning about quantum software development through that experience. And I would say most importantly, connecting with the team and getting that mentoring, mentorship, being part of the community. And it just kind of happened really organically building up those contributions and strengthening those connections. And then, in March of last year, I became a full time member of the technical staff working on quantum error mitigation and Mitiq.
Sebastian Hassinger 6:41
Awesome. And just to rewind for a second. So ASML, famously, is the tool supplier for -- that uses extreme ultraviolet, right lithography for two nanometer program. So that was, that was a solved problem, you decided you want to go find something else that was more challenging.
Misty Wahl 7:03
It was an amazing experience. And I definitely learned a lot. And you know, definitely, I can see parallels between, you know, doing something that's never been done before. A lot of people didn't believe that it could be done, or at least that it could be industrialized the way it ended up being. And I contributed to that program. And then, yeah, as it started to mature, I was starting to think about, okay, what's the next thing? And I mean, not to say that quantum computing is the only next thing, but really looking for that? Okay, what is it the going to be at the cutting edge? And it just really clicked for me when I
Kevin Rowney 7:47
used to be have a taste for a stern and formidable subjects, right. Mech-E is a no nonsense degree, right ASML man, they're doing it right at the limit. And then quantum computing, not not a walk in the park.
Misty Wahl 8:02
I suppose it's a theme.
Sebastian Hassinger 8:05
That's very cool. And Unitary Hack. It's really a unique event that Unitary Fund holds. I think it's in early summer, right? May or June, something like that. And essentially, that anybody participating in the company or entity participant can tag issues in their GitHub repos with Unitary Hack, and then participants can just grab those issues and try to solve them. So was your entry point more sort of, you know, classical software development, you were sort of squashing bugs that were more classical in nature? Or were you already sort of addressing the quantumness of the Mitiq framework?
Misty Wahl 8:46
I already went and just dove in to the quantum part as well, I was actually pretty new to software development at that point. And so it was I was learning everything at once -- I certainly had done programming before but not not like that not you know, for, for other people to use, it was kind of my own scripting. So
Sebastian Hassinger 9:12
tool, right, like your tools versus something, somebody else is gonna use definitely a different challenge. So that's, that's many learning curves at once. It's pretty impressive. So like, what was it that drew you to the Mitiq framework?
Misty Wahl 9:29
Well, part of it really was, I mean, at that time, I didn't really understand much about quantum error mitigation. So I was really looking for something that would that really did address the quantum aspect, as well as the software aspect. And so I saw that I actually joined partway through the hackathon. So it was also kind of like, okay, what are the remaining quantum related issues here, and this was one where it was actually applying Mitiq ZNE to a VQE problem and kind of showing that To help converge faster, it turned out that this was actually more of a research project. Because no one had really no one had really tried that exactly with Mitiq before. And so similar things, but not that exact problem. And so it actually took some time beyond beyond the hackathon to really get that.
Kevin Rowney 10:22
For the benefit of our audience you mentioned VQE, variational quantum eigensolver is right. It's one of the major algorithms, but yeah, they must be very sensitive to error propagation. So you can be Yeah, so it kind of
Misty Wahl 10:35
convergence. Yeah, cool. Yeah.
Sebastian Hassinger 10:39
And so Mitiq is a framework that allows for multiple approaches, or bring your own approach, really, you can build your own approach for error mitigation, the one that you first started working with zero noise extrapolation, can you describe how that works in at a high level?
Misty Wahl 10:55
Right? So zero noise extrapolation is kind of like it works in kind of two stages. So what you're doing is you're changing the noise level of the computation by manipulating the circuit in some way. And you do that a number of times at different noise scale factors. And then you extrapolate from evaluating those different noise scale factors, you get different expectation values from your computation. And then you extrapolate back to the zero noise limit. So you say what would be the expectation value if there were zero noise. And so there are different ways of applying ZNE. So the first way was done was through something called pulse stretching. And so kind of like increasing the duration of the pulse, you get more noise in the computation that way, and then the Unitary Fund, did something called Digital zero noise extrapolation. And so there, you are actually inserting extra operations in the circuit, we call that unitary folding. And those operations, ideally, they compile to identity. But in reality, if you implement them on the device, they're going to increase the noise level as well. And so if you do that many times that you get a lot more noise. And so then you're kind of further out on your curve. And so you get this nice curve, that you can extrapolate that. Well, that's fascinating.
Kevin Rowney 12:24
So what you're saying is that by introducing noise on purpose, we're able to then create a curve from which you can extrapolate down to the optimal value. Yes, super cool. That's cool.
Misty Wahl 12:35
And you can actually, go ahead, sorry, you can actually do it in the other direction, too, through something called probabilistic reduction. And so that's some framework that some folks that unitary fund also came up with the noise, extended probabilistic error cancellation, and virtual ZNE. So you can actually use that to instead reduce your noise level. And then you have these, also noise scaled values, but they're lower than, then you're kind of baseline computation.
Sebastian Hassinger 13:07
That's interesting. So I was gonna ask that you talked about either stretching the pulse or introducing additional pulses. Is that related in any way to something I've heard of called dynamical decoupling, which is introducing a random pauses in between gates in a way that sort of counteracts the noise that builds up when there's no operational and qubit.
Misty Wahl 13:31
Well, that actually works a bit differently. So that's kind of Yeah, so that's inserting pulses in idle windows, I mean, it's still still all these operations, in the ideal case, a compile time identity. So it's kind of the same idea of like, you kind of have an equivalent circuit in the ideal case, but in the noisy case, you don't, and it should help your your result in the end. But for dynamical decoupling, it's more about keeping your qubits busy, so they don't kind of wobble around an idle window.
Sebastian Hassinger 14:04
And it for for zero noise extrapolation or probabilistic error reduction. How much work does Mitiq save someone starting from scratch to implement that or some other mitigation strategy? I mean, it's a it's a reusable framework, right? So it's,
Misty Wahl 14:23
it is yeah. I mean, it's, it's designed to just, you can apply it in a few lines of code. Of course, if you're working with hardware, you may need to tune the parameters to actually, you know, see the improvement that you want, from Mitiq. So, we are actually working on a calibration module, and we've released the first version of it. And so it's kind of automating some of that parameter selection, where we have some predefined circuits. And then the user can run the calibrator and that will kind of give them some intelligence on what are the best parameters to select? Interesting.
Kevin Rowney 15:08
Kind of sounds like, you know, TensorFlow or pytorch. I mean, it can save you some time, but you probably should know what you're doing first.
Misty Wahl 15:17
We have a lot of great examples in our documentation. So I would really encourage people to check those out.
Kevin Rowney 15:24
That does remind me of what one of the questions I just thought of as me, if somebody you know, is listening, I'm certainly warming up to this idea of myself of wanting to get into Mitiq, and do some just fun goofing around with a what would be a good pet project for them to toast or to take on as a first sort of like hello world in in error mitigation?
Misty Wahl 15:45
I would definitely try reproducing some of the examples in our documentation, we have some really simple ones for ZNE. I think zero noise extrapolation is the easiest to understand conceptually, it's also the easiest to set up. So I would say start with those. We have a number of kind of, like variational algorithms, and you can, you know, play with different parameters, change the noise levels, see how, you know, try different noise scaling techniques, and start to get familiar with what are the options and how does it respond? And then I think you can build up from there.
Kevin Rowney 16:21
Sounds cool. And it's m-i-t-i-q. It's up on up on GitHub, right? Yes. So
Sebastian Hassinger 16:27
right, or Unitary dot Fund?
Misty Wahl 16:31
You can also find it through unitary dot fund. Yeah. And actually, we have the unitary fund organization on GitHub, and then Mitiq is under that umbrella. It's one of the repos in our organization. That's cool.
Sebastian Hassinger 16:45
Is it? Do you find, I mean, you're coming to this from a non traditional starting point. So it may be hard for you to to, to reflect on this. But it occurs to me that like most researchers are not building reusable tools while they're doing their research. And I'm just like, most researchers are doing something often it's whiteboard based, maybe they write some circuits that they run on simulators, or maybe actual hardware. But it seems like you're, you know, the Mitiq project is doing research, while developing, as you said, project or product code for other people to use. I'm just curious if you have any sense, what that does to the research process or the research, quote, you know, the quality of the deliverable? I mean, it feels like your deliverable is reusable code immediately, which is not usually the case with academic research. Yeah,
Misty Wahl 17:40
that's true. Yeah. Although I think we are starting to see more labs open sourcing their code, which is really encouraging. And even the things that we don't release as part of Mitiq, or at least not right away, are open source on our research repository. So our research repository is public, as all of our code and data, they're mostly in the form of notebooks. But anyone can download those and read them, and, you know, even kind of contribute to those if they want. So, yeah, this is definitely it's definitely different, I think, yeah, it does encourage, you know, a better standard of code. And I think it does. You know, we do have to think rigorously about what we're implementing and why. And I think that's really helpful. And research, I think it also helps in the writing process of explaining what we did, why we did it, why it's important. So yeah, I think it's, it's a great tool. We're also seeing, you know, a number of applications, like quantum simulation papers coming out that are using Mitiq. Not so much I read mitigation research, which is interesting. I think, I think our mitigation researchers, they have their own code at this point. And they're, they're using that. But I only want to
Kevin Rowney 19:10
Yeah, I'm sorry, interrupt that. I was just wondering, I mean, one of the big themes we keep up hearing on the podcast across multiple episodes is how some of these new frameworks in quantum computing could be either applied to to actually running, you know, experiments, essentially, about physics and the way the universe works. And or, you know, the same frameworks are excellent at getting new insight on the actual physics of an individual machine, a specific quantum computing architecture. I mean, I wonder to what extent some of these zero noise extrapolation algorithms might inform either of those two threads of of investigation. What do you think?
Misty Wahl 19:49
Well, I think it can be, I think it can be both actually. I mean, it was ZNE, it's kind of something where, you know, we're kind of applying it after the fact. So I think we're Getting limited information out in terms of the noise characterization there. So I want to be careful in terms of the performance of the device. But what we can see is that if we think about something like quantum volume, like you can actually enhance quantum volume experiments with zero noise extrapolation. And we've actually done that, for up on the archive about that,
Kevin Rowney 20:28
I'm just imagining that that could be different curves of ZNE response, depending upon the class of hardware that you're dealing with. And so I don't know, there could be cool comparative measures you can make there with respect to how different architectures have strengths and weaknesses.
Sebastian Hassinger 20:44
Yeah, actually. And to that point is, is there does Zanni or probabilistic error reduction do either than we're better on superconducting versus trapped ion versus another modality of qubits? Or is it? Is it pretty much neutral to the underlying implementation of the hardware?
Misty Wahl 21:02
We have seen different results on different modalities? And yeah, different devices? I don't remember off the top of my head. But we do have a pic, we do actually have a paper on that as well, where we test a number of different Yeah,
Sebastian Hassinger 21:19
it was gonna say, do Yeah, that basic question, actually. So what backends are supported through medica? What hardware? Can you run Mitiq on?
Misty Wahl 21:30
Anything, pretty much anything, it's just, it's about, you know, can you then convert, you know, the quantum program, to whatever the back end will will accept. But we can wrap pretty much anything as long as it's in a supported front end. When we support the major front end, so like qiskit, Penny, Lane, Braket, pyquil?
Sebastian Hassinger 21:56
Cool. Cool. And,
Misty Wahl 21:59
of course, we're written in Cirq.
Sebastian Hassinger 22:03
And I guess, I mean, it sounds like though pulse control would be required for the back end, you'd have to have access to pulses in order to properly implant is that right?
Misty Wahl 22:12
No, actually, so I want to clarify Mitiqis gate base. So it's gate level that is, so you don't need pulse control. Of course, that means that sometimes you're your error mitigation may get compiled out. So there are certain things where you have to specify that the optimization should, you know level should be zero, or maybe, you know, if you have some control over what kinds of optimization are done, you need to do that.
Sebastian Hassinger 22:42
Right. But you said pulse stretching was one of the the ingredients in zero noise extrapolation, I think, right,
Misty Wahl 22:50
but originally not.
Sebastian Hassinger 22:53
Right? You said it's moved to a digital
Misty Wahl 22:57
abstraction. Got it.
Sebastian Hassinger 22:59
Interesting. That's very cool. So I mean, I guess, you know, there's, there's when you think of the path towards fault tolerance, there's sort of like, basic qubit quality. And then so the one gate and one qubit, and two qubit, gate Fidelity's. But then there's sort of the hardware level, you know, fault tolerance, error codes, surface codes, and LDPC, and those types of things. And then and then there's sort of a level of abstraction, that's where error mitigation or other potential strategies would come in, do you see that sort of continuing to have a role over time? Or do eventually you get to the sort of this idealized state of the hardware, and you don't need to worry about mitigation anymore? at the software level,
Misty Wahl 23:46
I think you will, you'll always have errors, you'll always have quantum errors. And so you will always want to try and enhance your results. So there will always be a place for error mitigations as far as we can see. And in fact, I think error mitigation is going to be very important in the transition to fault tolerance, because it's not looking like one day, we're just going to flip a switch and quantum computers, fault tolerant. But, but you know, as we make these incremental gains over time, you know, we can use our mitigation at every step to enhance the results, right. And I actually worked on a technique combining quantum error mitigation and quantum error correction. And so that's scaling the distance of the surface code or of the error correcting code in general. And then using that as the noise scaling method for CNE.
Kevin Rowney 24:47
Isn't that your latest paper that was just
Misty Wahl 24:49
my paper that I recently presented at IEEE Quantum Week
Sebastian Hassinger 24:54
Very cool. So okay, so say more about that. So the distance is important. You In error correcting codes, because that has to do with the degrees of that you can prove three degrees of error that you can correct. Is that right? Yes,
Misty Wahl 25:07
that's correct. Yeah. And so if you have like a low distance, the idea is, if you're below the threshold, that's the caveat here. But if you're below the threshold, then you can, then you know, your low distance code, you're gonna have like, more or higher distance, you're gonna have less air in general. And so you can actually use that as the scaling knob. So you can deliberately, say, I'm going to encode at a lower distance, I'm going to have more error in my calculation, then I'm going to encode it at higher and higher distances until I'm at the capacity of the device. And then I have these noise scaled expectation values from each of those computations, I can extrapolate that back to the zero noise limit. And now I have a more accurate expectation value than I would have had had I just encoded at the highest distance I could. And then that's it. That's the error rate. So it can get this lower effective logical error rate by using this technique. Interesting. That you can do
Sebastian Hassinger 26:21
that, does that help with the efficiency of the error correction scheme as well? In other words, I mean, there's an overhead of, you know, a ratio of x physical qubits to why logical qubits, and that ratio can be very, very. So does this help you sort of achieve better results with a smaller number of physical qubits per logical qubit?
Misty Wahl 26:42
Yes, because one of the tricks that you can play is you can actually run certain parts of your competition in parallel. So you can run like different circuit instances. I mean, this, of course, if you have like modular circuits, you have the right size device, you can run your computation in parallel, you can take additional samples at these lower code distances, because you've freed up the capacity of the device at the lower code distances. And then as you go up, you can take fewer samples, but you have less error. And so you can also use that parallelisation As an additional knob. So you can kind of smooth out some of Oh, interesting some of the noise so that way, as well. And kind of adjust to like where the points are on the curve, and optimize your extrapolation in that way.
Sebastian Hassinger 27:34
That's a really good example of how error correction and error mitigation are going to need to sort of be used in collaboration with one another to get get better results
Kevin Rowney 27:46
together with big volume. Yeah, exactly.
Sebastian Hassinger 27:49
And something else occurred to me, so are we talking about all errors? Or is this phase flips or bit flips? Like, is it? Is it more suited for one type of error and qubits than the other? Or is it just errors generically?
Misty Wahl 28:06
Just generically. Yeah, I mean, we use a pretty simple error model. For for the demonstration. But certainly extensions of a work could be like, you could actually simulate, you know, like, surface codes or stem or something like that, and and try the same thing and see how that performs. Or assume that you have a different kind of error correcting code, you could assume different distances throughout the circuit. We didn't, we didn't turn all of those knobs we're just proving, yeah,
Sebastian Hassinger 28:38
that'd be interesting. Because you know, certain physical qubit implementations, like cat qubits, for example, are more resistant to one type of error versus another. So be interesting to see. If if this hybrid approach to mitigation plus correction is more effective in one scenario or another.
Kevin Rowney 28:56
I'm wondering if I could go out on a limb here and perhaps encourage you to speculate a little bit. I mean, it's, I don't want to put you on the spot too much. But yeah, this is one of the things that I do is that, you know, the threshold theorem, naively stated, right, is that there's a certain threshold of error correction, error mitigation above which we can get to essentially fault tolerant computing. And it's been a big couple of years right in, in those algorithms advances among them this the work that you're doing. I mean, what do you think? I mean, is it within sight? Some, some outcome where quantum computing could get to a crossing that threshold and be ready for primetime? Or are there still many breakthroughs left, left to hurtle over before we're in the end zone, so to speak?
Misty Wahl 29:43
I think we still need many more qubits and much better qubits. But I think, you know, that's, that's a big part of why we're doing what we're doing is that we need all of these ingredients like we need better error correcting codes. We need our mitigation you need that. or mitigation techniques. And, you know, we need this to be as open and as successful as possible to really make those kinds of breakthroughs. Yes.
Kevin Rowney 30:09
Yeah. And here it is, again. I mean, if you could speculate, do you try to put a timeframe on that one? I mean, I know our audience is curious. I mean, what do you think it's a few years, five years 10? Or, you know, maybe my kids lifetime?
Misty Wahl 30:25
I mean, I've been hearing 10 or more.
Kevin Rowney 30:28
Yeah. Still, it's exciting to hear that it's within, it feels like it's within grasp, in some, you know, reasonable amount of time. It's not just some science fiction remote.
Sebastian Hassinger 30:41
Really interesting. Grand Challenges are usually measured in the on the order of a decade.
Misty Wahl 30:48
And for any shorter you.
Sebastian Hassinger 30:57
You talked about Unitary Fund being about community building as well, what is it? You know, how does mythic itself get used to sort of build open source quantum community?
Misty Wahl 31:13
Well, I think a big way that we do that is actually through our community calls on Discord. So anyone can join those calls, and just interact with the team, audio, video, you know, whatever they're comfortable with. We also have forums on Discord, we have discussion forums on GitHub. And so we have a lot of different communication channels. And anyone can join, we welcome people to join beginners all levels, all backgrounds. And, yeah, I've definitely seen a lot of people get connected with us and I got connected with Unitary Fund through doing that. Also Unitary Hack, like having events like that, where, you know, there's a lot of buzz and you know, a lot of people around and kind of trying to figure out what's going on. And, you know, another way inroad for people to see, like what we're doing, and it's all it's all open, it's all you know, very accessible, so people can start to get a sense of that, and see where they would like to contribute.
Kevin Rowney 32:22
So cool.And Unitary Fund is a 501(c)3, yes? All right. Yes, it is. Impressive. And so how do you how do the economics work? I mean, there's there's a core team of like, competent, ambitious people, there's some funds allocated towards I gather, you're promising researchers? How did the how does the basic economics of the of the Unitary Fund work?
Misty Wahl 32:48
Sure. So we're, well actually, we do a few different things. So there's our internal research and internally driven projects like Mitiq and Metriq. And so that's funded through government grants. And then there's the micro grant program. And that's funded through corporate sponsors and individual donors.
Kevin Rowney 33:09
Fantastic.
Sebastian Hassinger 33:10
Mitiq actually was partly funded by a corporate sponsor, it was me. It was largely a government grant. But there was some additional funding that was required, I convinced my employer at the time IBM to foot the bill for it, because we, we really liked the project when will brought to our attention. So I was quite happy to be able to, to to help get that off the ground, because I just think it's it's amazing. Is there I guess, you know, there's the open source projects within Unitary Fund have I think, all been at least the ones that are micro grant, or internal projects have all been software, sort of on the on the the application side, right. It's circuits or compilation or error mitigation? I'm wondering, I mean, I know that there's been discussions about doing more open source hardware. And I just wonder, like, what would the impact or what would Mitiq be able to do if there was an open source, a control platform for qubits? That was also under the open source community? Umbrella?
Misty Wahl 34:26
Yeah, well, actually, um, recently, we put out a paper on open hardware as well. So kind of a review of all of the efforts that are ongoing, or many of the efforts that are ongoing in the field. So I'd encourage people to check that out. But yeah, I think, definitely, you know, we could have lower level control that would, you know, give us a lot of flexibility in terms of the techniques that we could apply and also our understanding of, you know, how well those techniques are working at the hardware level. And also, you know, noise, more noise aware techniques. Right. So getting more of that feedback kind of lower in the stack would be Yeah, that would be fantastic.
Kevin Rowney 35:12
That open hardware phenomena. That's, that's so cool, too. I a little bit of a, you know, DIY crafter guy and the, you know, the onset of microcontrollers that were that were consumer affordable. And it's I mean, it's just that the whole maker scene exploded and was just energize so immensely by that. I mean, again, this is no speculative thread here. But I mean, to what extent do you think it's possible that there could be at least some element of home DIY, Kevin wants a cube? Exactly quantum computing rigs, right? And I know lots of the gear so insanely expensive, there's no way you would want a dilution fridge in your basement. Unless you're, you're, you're crazy. Well,
Sebastian Hassinger 35:53
Can you spell Raspberry Pi with a "Q?"
Kevin Rowney 35:57
it feels like it feels like there's stuff happening out there we have you. Have you seen anything along those lines that might summon optimism for the ambitious maker?
Misty Wahl 36:07
Yeah, I mean, I think you already said it, the infrastructure around, you know, quantum computers, pretty expensive, to say the least. But, you know, but I think maybe some of the kind of more like the control electronics or, you know, other, you know, supporting components, that could be an opportunity for people to ellos and develop those. It's all important,
Sebastian Hassinger 36:36
right? I mean, I'm aware of QICK and Qubic, I think QICK has a channel on the discord now. And they're both using Xilinx boards with FPGAs to control superconducting devices, I think, or maybe even more than one type of connecting. And those are both ones. QICK is from Fermilab and Qubic is from the Lawrence Berkeley National Lab. So both really interesting open source projects. And that's exactly what I kind of had in mind is, I'm imagining what the Mitiq framework would be able to do if it had integrations into an open source control platform. And there was, you know, the ability to, to, you know, some future, you know, iteration of ZNE or a probabilistic error reduction, or some other approach that was actually reaching down into the control platform and using the native capabilities of that piece of hardware. As part of the the error mitigation, it seems like, you know, that opens a whole bunch of other doors for future directions.
Misty Wahl 37:42
Definitely, I think one important prerequisite for that, though, is that we really need to integrate with something like QIR or MLIR, which is definitely on our roadmap. And it's going to be a big effort, but definitely worthwhile one. So
Sebastian Hassinger 37:59
intermediate representations. Can you explain why that would be important?
Misty Wahl 38:04
Yeah, so I mean, well, part of it is that you want to be able to have greater interoperability in your quantum programs. So you want to use these intermediate representations. Also, including more classical programming elements. Right now, you know, most quantum front ends are pretty limited in terms of what they can do, when it comes to classical programming. And so having that all together in one framework, would be very helpful. And then that would also allow us to talk to, you know, the control hardware a bit a bit more easily.
Sebastian Hassinger 38:43
So in a way, like the the gate based circuit that you would produce in Mitiq, would be then translated to this intermediate representation, so that it could be executed in, in whatever way is appropriate for the hardware rights is literally a go between the intermediate between the high level circuit and the low level execution.
Misty Wahl 39:05
Well, that would be the that would be the prototype, I would say. But ultimately, we would probably want to look into applying error mitigation at a lower level, like in a car.
Sebastian Hassinger 39:18
Gotcha. Gotcha. Okay. So that would if you if you supported something like, like you use an MLIR or QIR, then you'd be able to incorporate those lower level controls in your error mitigation techniques, closer to got it. And would that be just another evolution of ZNEor probabilistic error reduction? Or is that would that be new approaches to error mitigation? Or both?
Misty Wahl 39:44
I would say both. Yeah, definitely. We would want to start with what we know and what we can test easily. But I would hope that that would open up a lot of opportunities to apply different error mitigation techniques or even development who are negation techniques. I think right now, a lot of the reason why medic is designed the way it is, is because of some of the limitations that exist in quantum programming right now. And what we're able to do with with the front ends, we have their limited interoperability, and we're already able to do quite a bit and that it can convert between different front ends that can accept pretty much any back end. But there's a lot more that we could do if we were able to operate at a lower level.
Sebastian Hassinger 40:32
Interesting, are there theoretical error mitigation strategies out there that you are thinking might be next on the Mitiq roadmap?
Misty Wahl 40:44
I think, a really interesting, one of the really interesting directions is some of the machine learning, quantum error mitigation techniques that are been coming up recently. So I think you'll be seeing more of those no
Kevin Rowney 40:56
doubt, that same phenomenon appeared several times in the podcast, that's huge frontier really fast into Yeah, yeah.
Sebastian Hassinger 41:04
That's cool. So with that, can you describe more? How ml might be used in in their mitigation?
Misty Wahl 41:12
Well, I mean, there's a couple of different ways. I mean, one is kind of, you know, learning your, or mitigation parameters, you could also think about, you know, noise aware techniques, you could think about different ways of compiling circuits, to kind of optimize for noise. So it's kind of all part of the same thing. But like looking at the, you know, slightly different parameters, whether you're talking about like, circuit compilation in general, or you're talking specifically about inserting gates in certain areas, or, you know, scaling up noise or scaling it down. And, yeah, there's a lot of different possibilities.
Kevin Rowney 41:51
One might speculate that the ZNE technique could be more complex curves of response that would perhaps elicit the need for more advanced algorithms beyond regression to find a zero point, who knows?
Misty Wahl 42:07
Definitely, but I think there's also, you know, many more areas to explore in terms of QEC inspired quantum error correction techniques. And those could also benefit from, you know, ml. So, the, the positive, there are a lot of possibilities.
Sebastian Hassinger 42:27
If you had pulse control level control over a qubit system, would there be potentially like ways to train ml models to tune the pulse schedule for particular gates or, or even sort of, you know, have undirected impulse control over the qubit and just tell it what, what result you want, and it sort of hacks at it until it comes up with just guesses sometimes until it comes up with the right way to get the logical result.
Misty Wahl 42:57
That's a, you know, it would be great. If we could do something like that. I think something, you know, it's, there are probably, you know, quite a few, quite a few steps are needed to actually get to something like that. But that's, you know, that's that's a great future vision for our mitigation, I would say. Very cool.
Kevin Rowney 43:18
This has been really fun. We really appreciate your time. Thank you.
Sebastian Hassinger 43:21
Yeah.
Misty Wahl 43:21
Thank you for having me.
Sebastian Hassinger 43:23
Our pleasure. I mean, it's it's a great project and unitary Fund is a terrific organization. And I mean, I think Mytek is such a great tentpole in the in the open source community that's emerging in quantum I think it's it's fantastic work. So actually work Yes. Thank
Kevin Rowney 43:38
you so much. Thank you.
That was That was great. I mean, Misty, she's really an impressive person. I mean, to start with a Mech-E degree and that's a no mess around your Basecamp to start with, and then she jumps into UV lithography. And then wow, I guess she self taught herself on software engineering, while simultaneously learning quantum mechanics. Yeah, not bad and
Sebastian Hassinger 44:48
extreme ultraviolet.
Kevin Rowney 44:51
extreme UV. She's doing research on quantum error. mitigations it's pretty cool.
Sebastian Hassinger 44:56
It's very cool. Yeah, no, that was great. And I really liked Getting more in depth sort of background on zero noise extrapolation, probabilistic error reduction, and especially that latest work that sort of combines error correcting codes at the hardware level with error mitigation strategies. So that really suggests a really interesting future possibilities for the for the medic framework and for the the error mitigation field itself, no
Kevin Rowney 45:24
No doubt, no doubt. And it was fun. Just I don't know, messing around a little bit on the topic of my dream of someday being, you know, home DIY quantum computing, open tech, you know, someday
Sebastian Hassinger 45:35
You'll get your little tiny dil fridge or your lasers, your optic table in your kitchen. Absolutely. All right. Thank you so much for joining us. Until next time. Thanks.
Kevin Rowney 45:52
Okay. That's it for this episode of The New quantum era, a podcast by Sebastian Hassinger. And Kevin Roni, our cool theme music was composed and played by Omar Costa Hamido. production work is done by our wonderful team over at PodFly. If you're at all like us and enjoy this rich, deep and interesting topic, please subscribe to our podcast on whichever platform you may stream from. And even consider if you'd like what you've heard today, reviewing us on iTunes and or mentioning us on your preferred social media platforms. Were just trying to get the word out on this fascinating topic and would really appreciate your help spreading the word and building community. Thank you so much for your time.