QHack Recap — PennyLane, Amazon Braket and Beyond 🚀

Eventually, though, you’ll want to get your hands dirty. As the late Asher Peres once wrote “quantum phenomena do not occur in a Hilbert space, they occur in a laboratory” (Peres, 1995). The good news is that the advent of quantum cloud services like IBM Quantum Experience, Amazon Braket or Azure Quantum (in preview), and quantum programming languages and frameworks has made it a lot easier to run quantum “experiments” without an actual lab — whether it’s using a good ol’ simulator or one of those Devs-like quantum “chandeliers” we keep seeing on quantum computing articles.

💡 Tip: A great way to get started is to check the code samples from IBM, AWS and Microsoft, and the awesome quantum computing and quantum software repos available on GitHub.

From finance to healthcare and everything in between, quantum computing, and QML in particular, has a huge potential to disrupt and revolutionize strategic key areas, and there are already plenty of quantum applications and use-cases out there. Some, nevertheless, are still far, far away… 🔭

In the age of noisy intermediate-scale quantum (NISQ) devices, emphasis on noisy and intermediate-scale, it’s often hard to distinguish what pure quantum and quantum-enhanced (hybrid) devices can and cannot do (or, as Scott Aaronson puts it, read the fine print), and steer clear of false advertising. Ignoring these limitations will lead us into the perfect storm of hype, hubris and hogwash.

QML is a vibrant, new field with great potential, but a very young one still 👶 It’ll be interesting to see what the top 20 teams come up with during the Open hackathon. Who knows, maybe the next quantum killer app is already afoot…

💡 Tip: As of February 2020, the best way to learn QML is to attend Peter Wittek’s QML course on EdX — the lecture videos are available on YouTube. Unfortunately, Peter disappeared in September 2019 during a hiking expedition to the Himalayas. His legacy, however, is stronger than ever. The Unitary Fund (UF) and the Quantum Open Source Foundation (QOSF) have recently announced the creation of the Wittek Quantum Prize for Open Source Software.

After 3 full days of lectures and amazing guest speakers, here are some of the highlights — all videos are available on Twitch:

Day 1

We kicked things off strong with a high-level introduction to quantum differentiable programming (Q∂P) by Maria Schuld (Xanadu + University of KwaZulu-Natal), which paired up nicely with Josh Izaac’s (Xanadu) talk on Q∂P with PennyLane.

😋Tongue-in-Cheek Moment: The 1st half of this talk can be summarized by a few quotes from Karpathy’s book of aphorisms (unofficial title):

The PennyLane website contains a page dedicated to quantum differentiable programming, and you should definitely check the original PennyLane paper (Bergholm et al., 2018) for additional details.

💡 Tip: You’ll find some great introductions to differentiable programming (read Q∂P minus the Q) here and here — this last one uses Julia. If you want to know what powers the ∂P engine viz. automatic differentiation and how it’s used in ML, check out (Baydin et al., 2018).

Patrick Coles (Los Alamos National Laboratory) offered some great advice on how to design and build quantum variational algorithms (pictured below)

In the NISQ era, variational algorithms running on hybrid devices are pretty much the workhorse of QML applications. Virtually every problem in the QML challenge involved some sort of variational circuit.

💡 Tip: For more information, check out this great (and freshly out of the print) review article on variational quantum algorithms by Cerezo et al. (2020)

Day 2

Sukin “Hannah” Sim (Harvard + Zapata Computing) gave an update from the trenches of parametrized quantum circuits (PQC) optimization with some interesting work on parameter-efficient circuit training (PECT).

During the afternoon session, Guillaume Verdon (Sandbox@Alphabet) announced Floq 🦜— a quantum SaaS API built on top of TPUs — because, apparently, quantum ❤️ TPUs

Having placed in the top 50, I got the chance to play around with an alpha version of the Floq simulator — keep in mind, it’s still highly experimental (!). It will be interesting to see how the service evolves and incorporates the feedback from QHack’s early batch of users.

Day3

In a throwback to my early days as a particle physicist, Sofia Vallecorsa (CERN) showcased some great quantum technology initiatives in high energy physics (HEP) — from quantum SVMs for Higgs boson classification (Wu et al., 2020) and particle tracking with quantum GNN (Tüysüz et al., 2020) to quantum GANs for detector simulation (Chang et al., 2021) and beyond.

💡 Tip: For an introduction to QML in HEP, check out Schuld et al. (2020)

Finally, Murphy Niu (Google AI Quantum) gave us a peek behind the curtain of what it means to do ML at Google Quantum. A special mention to the work on quantum sample learning (Niu et al., 2020) and the final observations on where Google is heading QML-wise.

1. If you want to win, team up—time is your worst enemy during an hackathon and working alone can only get you so far up the scoreboard; assemble a team (the hardest part!), define some goals (e.g. reaching top 50), identify possible constraints (different timezones, day job from 9 to 6, exam season around the corner, &c.), create a battle plan (tackle “low hanging fruit” problems first, go for a certain category, &c.), distribute the challenges among your teammates according to their expertise and start climbing that hackathon hill ⛰️
2. Know thy primitives — make sure you have a solid knowledge of Pennylane’s quantum operations and templates, it will save you a lot of time
3. Read, don’t F-read— read the problem statement carefully from top to bottom and don’t shy away from lengthy descriptions
4. Don’t cower before (seemingly) hard problems — this happened to me with the natural gradient problem and it ended up being one of the easiest to implement
5. Don’t stay too long on the same problem — either go outside and breathe some fresh air or move on to the next problem
6. K.I.S.S. and follow your instincts— don’t overcomplicate (elegant solutions usually involve just a few lines of code) and trust your intuition (if you “know your stuff”, it’s usually right)
7. Think outside the box but stay in the box— using drag-n-drop simulators like Quirk is a great way to gain some intuition on what an Ansatz is actually doing but not much else; run experiments while coding and keep a log of all the changes you’ve made and what their impact was