Hi there!
My name is Joe Slote, and I'm a little hard to place. Something like a mathematician-scientist-engineer-designer.
I hold an MSc in Mathematics and Foundations of Computer Science (MFOCS) from the University of Oxford and a Bachelors in Mathematics from Carleton College.
I am the Director of Science and Technology at Dascena, Inc., a digital health company in the Bay Area. We use machine learning to provide early warnings for elusive diseases like sepsis in hospital patients. Since my joining in October '17, I have refined our machine learning algorithms' loss functions, re-architected our product from the ground up, and doubled our data collection performance.
In general I try to direct my time and energy towards improving the health of people and the world. I find effective altruism and humane technology particularly compelling expressions of this goal.
As a researcher I am primarily interested in supporting foundational developments in science and technologyâ€”developments which have the potential to elevate human experience and empower those engineers and designers who help to shape that experience. If I were alive in a previous era I would be working on computability and complexity, (classical) information theory, or quantum mechanics. But I am alive in this era, where my attention is captured by statistical learning theory, quantum computing, and the mathematics that support them.
- My masters thesis represents a small step in the quest to establish a theoretical basis for the surprising power of deep artificial neural networks. In particular, I investigate the simplest question you can ask about artificial neural networks: can a given network architecture even well-approximate your target function? This and related questions are surprisingly difficult to answer, and the source of this difficulty lies all the way back in our shallow understanding of function composition.
- I created a knot invariant, the fold number, which uses properties of a certain class of embeddings of the plane into R^{3} to characterize the complexity of knots. The invariant is best-described through origami: imagine drawing a loop on a piece of paper, folding up the paper such that it self-intersects (this is "ghost paper," by the way), and then cutting away all but the loop. It turns out all tame knot types may be produced in this manner, and the minimum number of origami folds required may say interesting things about the knot itself. I'll be presenting on the topic at 7OSME at the University of Oxford in September.
- I've studied the scaling behavior of error in the boson sampler, a simple optical device conceived to demonstrate the supremacy of quantum computation over classical computers. Together with Ivan Deutsch and Bob Keating at the University of New Mexico, I discovered this scaling was actually quite favorable, further validating the experimental viability of the device. V. S. Shchesnovich beat us to the presses that summer.
- I keep the open problem of meander enumeration in my back pocket for dull moments. So far I've found operadic methods to be a promising avenue to structuring meanders in a way that can be easily counted.
I write software and make art in my free time. Past projects have included a clearinghouse for face-to-face textbook trades and a whole host of sculpture.