WEBVTT 00:00.000 --> 00:11.440 All right, yeah, the Mexicans are traveling in June, they're going to talk about the 00:11.440 --> 00:17.440 Apache project related to Quadum. 00:17.440 --> 00:18.440 Thanks for joining. 00:18.440 --> 00:19.440 Good to see everybody. 00:19.440 --> 00:21.680 Hope you're having a good fun as them. 00:21.680 --> 00:25.080 Hope you have a good trip home, wherever you came from. 00:25.080 --> 00:27.480 We came from Chicago and Portland, Oregon. 00:27.480 --> 00:29.120 I had only 17 hours of flight. 00:29.120 --> 00:31.880 So grateful to be here. 00:31.880 --> 00:37.280 Just wanted to give everybody a heads up on a new feature in Mahoot, which has allowed 00:37.280 --> 00:41.600 us to put Quadum compute in our back end. 00:41.600 --> 00:46.880 We're just going to talk about who we are, where the project has been over the years. 00:46.880 --> 00:54.960 The rationale for putting Quadum into Mahoot, quick demo or just screenshots of live notebooks. 00:54.960 --> 00:59.120 And then a way calls to action for how you guys can get involved. 00:59.120 --> 01:03.000 Who am I? 01:03.000 --> 01:07.720 I came from Portland, Oregon, like I said, I've been working on Mahoot for about 13 years. 01:07.720 --> 01:13.720 I got into that when I found a bug in one of the similarity computations on a project 01:13.720 --> 01:14.960 for a bank. 01:14.960 --> 01:19.400 So we were making recommendation in June and we were dropping all our results on the floor, 01:19.400 --> 01:22.720 got on the user list or the dev-lift at least and asked how I could help. 01:22.720 --> 01:28.800 And so I fixed a bug and got the bug for doing matrix math being able to see where 01:28.800 --> 01:34.080 all the arithmetic happens and it's been a fun ride, Metraver there. 01:34.080 --> 01:36.440 Have a lot of hobbies. 01:36.440 --> 01:42.880 Head of technology at a speed chain, which is a B2B startup and then Trevor, who are you? 01:42.880 --> 01:46.560 Hi, I'm Trevor Grant. 01:46.560 --> 01:51.040 I also am a PMC, which is a project lead on the Mahoot project. 01:51.040 --> 01:56.240 I wrote a book in some papers, but nothing on Quadum. 01:56.240 --> 02:01.520 I'm not a Quadum expert and I aside from giving a talk at Fas Dem, don't really pretend 02:01.520 --> 02:03.320 to be one, at least not to be people actually. 02:03.320 --> 02:04.320 No Quadum. 02:04.320 --> 02:06.520 I've got two boys. 02:06.520 --> 02:07.520 I've been kind of out of space. 02:07.520 --> 02:08.520 I used to speak a lot. 02:08.520 --> 02:09.880 I haven't for the last couple of years. 02:09.880 --> 02:10.880 I had a couple of boys. 02:10.880 --> 02:11.880 They're really great. 02:11.880 --> 02:14.720 But I've been card-free since 23. 02:14.720 --> 02:17.720 I've actually been card-free since 22. 02:17.720 --> 02:21.480 A lot of people in here from Europe probably doesn't sound like it's really impressive 02:21.480 --> 02:22.480 in America. 02:22.480 --> 02:23.480 It's pretty hard. 02:23.480 --> 02:25.280 I'm very proud of myself for that. 02:25.280 --> 02:32.000 As always, I have a special thank you to IBM for patroning my artistic endeavors, such 02:32.000 --> 02:37.960 as my vacation to Brussels. 02:37.960 --> 02:42.280 A new wee art, as I was alluding to, we are not Quadum researchers. 02:42.280 --> 02:45.640 You can, I think, awesome phrase I heard. 02:45.640 --> 02:49.360 You can nerd snipers, or try to show that we don't know what we're talking about. 02:49.360 --> 02:50.360 You're going to win. 02:50.360 --> 02:52.000 I'm just going to give that to you right now. 02:52.000 --> 02:56.320 If you're a freshman in a Quadum program, you are going to blow us out of the water with 02:56.320 --> 02:59.800 some questions, some got you question on Quadum computing. 02:59.800 --> 03:02.960 We're not, you're not in a PhD either. 03:02.960 --> 03:06.040 No, we're not a PhD at all. 03:06.040 --> 03:08.880 I've got a master's degree. 03:08.880 --> 03:12.200 Again, we write some stuff, mainly blog posts. 03:12.200 --> 03:17.840 But who we are, we are pretty confident software engineers, who has a background that we're 03:17.840 --> 03:24.520 going to get into of abstracting over back ends and why that's important. 03:24.520 --> 03:29.240 We have a history of doing that, and when we saw this quantum space emerging, it seemed 03:29.240 --> 03:33.720 like a good fit for a number of reasons, so with that, I'm going to turn back to Andrew. 03:33.720 --> 03:39.000 Yeah, so this project has evolved over the years, and basically it's been a small club 03:39.080 --> 03:47.880 of people interested in machine learning, data science, analytics, natural language processing. 03:47.880 --> 03:55.000 We've evolved from initially a sub project of Lucine, became a top level project as 03:55.000 --> 04:00.200 a recommender engine, which Sean Owen donated, I think it was called Taste. 04:00.200 --> 04:06.360 It's evolved over the years from a Hadoop-based MapReduce system, which turned out to be 04:06.360 --> 04:11.640 a bad architectural choice because you're doing full table scans every time you do an iteration 04:11.640 --> 04:19.360 on your machine learning algorithm, so we re-tooled and worked on a spark back end. 04:19.360 --> 04:24.000 Then the direction of the project took a turn where we realized that there were so many 04:24.000 --> 04:29.320 back ends popping up, there was flink, there was H2O, and so we decided that the right 04:29.320 --> 04:35.080 thing to do was to provide a uniform interface where you could write code one time and 04:35.080 --> 04:38.440 then run it on any back end that you chose. 04:38.440 --> 04:44.120 That basically fast forwarding was where the idea for quantum came, because as you'll see 04:44.120 --> 04:48.680 the several back ends in the cloud providers, we thought we had an opportunity to make 04:48.680 --> 04:53.480 something that would benefit the community so that you didn't have vendor lock in, you 04:53.480 --> 04:58.280 didn't have to decide for the next 10 years what platform you'd use, so that's the motivation 04:58.280 --> 05:05.720 for where we are with this. 05:05.720 --> 05:14.040 Yeah, and so coming off of that, we put a Q-mat into Mahood as a way to be a layer in front 05:14.040 --> 05:17.960 of any of these back end systems. 05:17.960 --> 05:24.360 The core capabilities of the project are matrix arithmetic, so matrix multiplication, 05:24.440 --> 05:30.440 factorization, any kind of dot product, similarity calculation, things like that, and what 05:30.440 --> 05:37.000 we looked at when Trevor decided that this might be a good place to work was that the 05:37.000 --> 05:44.200 math behind quantum looked a lot like linear algebra, so we figured it was a good fit. 05:44.200 --> 05:51.080 So everybody here who's a researcher, I'm sure this is grade school level, but this came from 05:51.160 --> 05:56.440 another talk that we did too, but this is just explaining that when we looked at this, we just saw 05:56.440 --> 06:05.240 unit vectors, you know, basis vectors, and so it made a lot of sense for us, and so matrix multiplication 06:05.240 --> 06:11.480 obviously is just what we do all day and in distributed level, and so making it a new layer made 06:11.560 --> 06:14.760 a lot of sense for us, and you're going to do that. 06:18.360 --> 06:22.520 So as I said in the abstract, if you may or may not have, what you may or may not have read, 06:22.520 --> 06:29.000 there's approximately as many quantum programming languages as there are quantum computer vendors 06:29.000 --> 06:40.840 at this point, so you've got a very diversified field, and KISKID is an other slide on this, 06:40.840 --> 06:46.360 but KISKID is emerging as a standard, they also realize like, hey, we don't, 06:46.360 --> 06:50.760 nobody wants to pick one because they're afraid, I will do all this research, that quantum computer 06:50.760 --> 06:55.000 will go away and we'll have to start over from scratch, and so KISKID is kind of making all these 06:55.000 --> 07:00.360 different backends, and I do work for IBM, and I want to, in fact, at least say this is not a 07:00.360 --> 07:07.640 pitch for IBM KISKID, again, upcoming slide will show me turning on them soon, and anybody 07:07.720 --> 07:16.920 in here working for IBM, before I say, I guess it's a livestream, oh yeah, it's a livestream, 07:16.920 --> 07:22.200 I won't say anything too egregiously mean to them, I already said thank you for putting my vacation, 07:22.200 --> 07:28.200 so, yes, you have KISKID, you've got bracket, you've got circuit, and the list goes on and on, 07:29.080 --> 07:34.200 we started with the KISKID bracket in circuit because they are arguably the bigger frameworks, 07:35.160 --> 07:42.280 and we will expand out from there, we have different vendor platforms running on, 07:44.200 --> 07:51.240 very, very similar operations, and I've heard rumors that bracket, for instance, will have 07:51.240 --> 07:57.720 optimizations written into bracket that makes quantum stuff run faster on, they will optimize for 07:57.720 --> 08:01.400 when you're running on AWS, and they're, we're not going to have that, so if you are 08:01.400 --> 08:07.400 performance constrained, which I don't think quantum computing as a disciplines really there yet, 08:07.400 --> 08:13.160 where you're trying to milk like super low latency or things like that out of it, but that day will 08:13.160 --> 08:17.960 come, and when that day happens, you're going to, at least, you're probably used KISKID, 08:17.960 --> 08:24.120 or you use KU-MAT to prototype what you're going to build, go get your BC funding, and then pick 08:24.120 --> 08:28.520 your vendor, and then you can optimize and you can really squeeze all those last little bits out of it. 08:29.320 --> 08:33.560 But our background and who, again, being able to abstract over multiple back ends was a 08:33.560 --> 08:41.640 core value we already had, we understood why it was important, so that is that. So yeah, actually, 08:41.640 --> 08:45.960 and normally if there was another IBM or in here, there would be, and I'd be like, you know, 08:45.960 --> 08:52.280 KISKID does all this, it does, they also won't, they will give you the optimal version for 08:52.280 --> 08:58.600 IBM quantum computers. I feel like at Bob's demo, we're all really big and open source, 08:58.600 --> 09:03.400 and we understand why of vendor having the abstraction layer for all the other vendors is probably 09:03.400 --> 09:08.280 a bad idea. If anybody really needs me to go into that, we need to go into that, we can save 09:08.280 --> 09:15.720 a per Q&A later, save a per Q&A later, when next slide. So current states, 09:16.280 --> 09:26.120 we, at the moment, have basic gates. We plan on doing some machine learning stuff and more, 09:26.120 --> 09:34.760 but basic gates, I mean, this is really not particularly high octane stuff and three back ends. 09:34.760 --> 09:40.360 However, the same circuits, it's translated by once for one, you can just switch one line of code, 09:40.360 --> 09:45.000 and you can run Q-mat, or you can run KISKID, you can run bracket, whatever you want in one line. 09:46.680 --> 09:53.960 We also have some, Tom's side card methods like being able to do drawing, being able to 09:53.960 --> 09:57.320 parameterize circuits, which is starting to lay the groundwork for quantum machine learning. 09:58.520 --> 10:02.520 We have a getting started in quantum guide, which is really fun too. 10:03.800 --> 10:08.200 And as I said, we're not quantum experts, so if you want to make some pull requests, 10:08.600 --> 10:12.840 come and read our stuff, I'm sure there's errors in there, and that's a good on-ramping for 10:12.920 --> 10:18.520 other people getting into quantum computing, which is nice. And we have a co-labrunable notebook, 10:18.520 --> 10:24.520 which we will talk about soon. But coming soon, again, we're eventually working toward doing quantum 10:24.520 --> 10:29.720 machine learning stuff. That is a goal that we have in mind. And then the other beauty of that, 10:29.720 --> 10:34.360 too, is you can lay out a quantum machine learning circuit, and then swap one line, and now you can 10:34.360 --> 10:40.760 run it on, whatever you want to run it on. More algorithms coming back to falls down because it's 10:40.760 --> 10:45.080 really fun coming to Brussels, some day getting some VC monies, so we don't have to work for IBM 10:45.080 --> 10:49.960 any more, we can just buy our own plane tickets, et cetera. So I will hand it back over to Andrew for the demo. 10:51.240 --> 10:55.480 Oh, so yeah, there's a problem with the crackling in the browser. You want, I think it is. 10:56.120 --> 11:00.280 The next slide just shows the notebook and the screenshot. Yeah, yeah. 11:02.280 --> 11:07.640 So we've used co-lab. This interface is really what would GitHub, just as far as developers 11:07.640 --> 11:12.680 tooling. When you have a notebook in GitHub, you can just pop, there's a button for popping up a 11:12.680 --> 11:19.000 co-lab notebook. We do have that button on this exact file. I happen to not point to the right 11:19.000 --> 11:28.280 notebook. So here are some screenshots. This is all run-able. We do have a simplified install, 11:28.360 --> 11:38.040 so you can just run these cells. This is all in github.com slash Apache slash Apache slash Mahoot slash 11:38.040 --> 11:44.280 cumat slash examples. And we have some examples of the circuit being drawn, so you can visualize the 11:44.280 --> 11:49.800 matrices you're multiplying together. We had some other stuff that I think is a work in progress. 11:49.800 --> 11:55.080 There was some bloch spheres. So this all is online. You can run this stuff, bless you. 11:55.400 --> 12:00.120 You can run stuff, you can make pull requests, you can tell us where we're wrong, 12:00.120 --> 12:08.440 you can tell us where we can improve. And then, like Trevor has said, we are not experts. 12:08.440 --> 12:13.240 And we are software engineers who have specialized in machine learning and matrix math. 12:13.240 --> 12:19.960 And we are dipping our toes into quantum computing. We actually looked at this morning at 12:19.960 --> 12:25.080 breakfast. I was looking up how to package this into pipi and I found out that user 12:26.200 --> 12:31.800 Zonky had already done it for us. So somehow there's an unsanctioned Python module out there 12:31.800 --> 12:35.880 that you can install with pip. So we're going to unfortunately Zonky didn't give us any 12:36.680 --> 12:42.360 way to get in touch with them. So we are going to go through the press. If anybody's gone, 12:42.360 --> 12:47.160 so Zonky uses a hook here on my file. If you have, please let them know. 12:48.120 --> 12:53.160 Please play with the notebooks, join the mailing list, and buy us beers. I had my first 12:53.160 --> 12:58.920 non-alcoholic Belgian last night. It was actually very delicious. And I think that's a question time. 13:09.480 --> 13:10.120 Any questions? 13:10.760 --> 13:16.760 Yes, Nate. 13:22.760 --> 13:24.760 Question. 13:29.960 --> 13:34.600 Yeah, the question is what do people use in the food for prior and would that also translate 13:34.600 --> 13:38.760 over to the quantum backend? People have done a lot of recommended systems, a lot of 13:38.760 --> 13:44.200 similarity calculations, a lot of analytics. I think that the quantum stuff, the 13:44.200 --> 13:48.920 the end goal, will be to be able to run workloads in the quantum stuff. I'm really, 13:48.920 --> 13:51.560 do you know more about the use cases for quantum that would be different? 13:53.640 --> 13:58.520 Yeah, I mean that to be fair goes to why do use a classical computer versus why use a quantum 13:58.520 --> 14:03.480 computer? And as I understand it, the answer is their different use cases. So 14:04.360 --> 14:10.040 we're probably not, until we get to some sort of singularity, and I don't know how far 14:10.040 --> 14:15.720 different use of this in the future, where those kind of problems start happening, or you're 14:15.720 --> 14:17.080 getting more crossover. 14:26.760 --> 14:29.320 That is a great idea, and we could probably do that, yes. 14:30.280 --> 14:36.840 More distributed. There was actually a workflow prior where it did an analysis of your job, 14:36.840 --> 14:41.640 and it did, it was able to decide what solvers to use and things like that. So yeah, that would be a 14:41.640 --> 14:45.000 great, a great picture. Yeah, thanks. 14:48.840 --> 14:57.160 Question? Commons? What would we get wrong? Thanks a lot. Thank you.