Building projects in quantum machine learning (QML) lit a solid desire to see the frontier of research in QML up close. How do QML researchers traverse the black-box ambiguity of quantum algorithms? Where does an idea for something like recursive QAOA or applying effective dimension to QNNs come from? Those were the questions that led me to spend my summer as a research intern in the Quantum AI team at Zapata Computing.

While tweaking, refining, and fleshing out a research idea, I discovered that the scientific process is much more structured than I had assumed! From an outsider's perspective, it's easy to look onto the multitude of polished publications and conclude that scientists ideate something out of thin air, test it, then proceed to put pen to paper. How else do these papers seem to come out at such a clip?

The reality is muddier; I've realized from conversations with the 20+ post-doc researchers at the company and through direct experience. Arriving at the core research idea itself can take months, if not years, of refining. Then developing that initial hypothesis (which usually turns out to be partly flawed) into a valuable insight is less of something to stumble upon and more of a grinding, collaborative process which slowly inches the collective forwards. Research is hard! 

At Zapata, my research primarily focused on investigating quantum generative models. Although we made significant progress on the initial idea, the time constraints of the 9-week internship meant that I had to leave behind a few open questions. Nonetheless, the rest of the team and I were pleased with the preliminary results I reported. I can't wait to see where my learnings take me next within quantum computing (see below for info on my next research term!) 

I'd also like to note that the people of Zapata are some of the most unexpectedly lively quantum researchers you'll meet. It's a refreshing community to be around :)

In early August, 20 youth from across Canada & the US were flown down to California for a week-long seminar series covering topics around rationality, machine learning, statistics, and how one can go about life doing the most good.

I'm still unsure of what exactly made the experience so worthwhile, but I'm confident it's some linear combination of the vibrant, intellectual community and that environment driving meaningful ideas. Those are not orthogonal bases, but there was an excess of exciting ideas put forth, whether at 10 pm in a hot tub or 7 am on a calm stroll through the woods. 

One of which was the thinking behind how one should go about developing the most precise map of reality. Put in simpler terms, how do you ensure you're right with the plethora of beliefs you hold? 

Briefly put, leave your beliefs open to being updated, and update them slowly as you gather data points from the world. We hate ambiguity, so we often relax into having binary beliefs. Either X is 100% true, or it's false. 

But we often forget, we're not infinitely wise. Our map of the world is flawed. So assigning 100% or 0% probabilities to any belief outside of propositional logic is likely foolish. Instead, approach the world with humility and welcome the discomfort of holding probabilistic beliefs. Even superforecasters start with largely indeterminate probability distributions, which slowly evolve given more data points to a stance like, "I'm 80% confidence that X is true given supporting arguments/confidence in Y + Z both being true." 

To find out how well-calibrated your hypothesis engine is, do this 10-mins assessment.

Qiskit hosts a summer school featuring quantum computing lectures and labs curated by researchers at IBM Quantum each summer. This year, the focus was on quantum machine learning, and the community's appetite for it was apparent by the 60 minutes it took for all 5000 spots to be filled.

I enjoyed helping out many of the students that took part through my capacity as a mentor. The lecture material and lab content were top-notch, and they just posted all of it online a few days ago. I've linked it below. 

The organizing team also ran an early access QGSS version for mentors and internal IBMers, where I learned a ton. The much smaller student size of 40 meant more hands-on help directly from the researchers, which was phenomenal.

Aside from a boatload of technical items, what I took away most came from how the opportunity came to be. Amira Abbas was one of the principal lecturers and organizers of the event. She's an inspiring figure in the field, and I had spoken with her before. So I sent an email inquiring how I could participate in QGSS, and she signed me up to join the early access QGSS! 

Irrespective of the industry, it can be a huge leg up if you reach out to the people you admire in your domain. They're usually amiable people, and getting to know them only brings you closer to where you'd also like to be!