Random Projections Personal blog Sven Schmit

CODE @ MIT 2022 round up

After two long and rewarding days, CODE 2022 is a wrap. Here’s a biased look at some of my personal highlights; Keep in mind that this is far from a complete list: the conference was filled with quality content and due to the parallel nature of the talks, I missed more than half of them to begin with. Note: the plenary talks should be posted on... Read more

Beware of the Bayesian imposter

Sometimes, statistical guarantees are not what they seem. Here, we discuss the implications of a classic work that demonstrates a paradox with the Bayesian approach to experiment analysis: when not careful, the experimenter runs the risk of running a frequentist analysis without realizing it. This can have important implications: when combined w... Read more

Language modeling with Jax and RNNs

Jax is a relatively new Python library aimed as a drop in replacement for Numpy for machine learning research. It sets itself apart due to its functional approach, which I find really enjoyable. Recently I have been playing around with implementing a simple RNN using Flax to get beyond the basics of Jax, but without adding all the bells and whis... Read more

Experimentation with resource constraints

Based on work at Stitch Fix around experimentation with resource constraints and our introduction of the “virtual warehouse”, Greg Novak, Dave Spiegel and I wrote a post on the Multithreaded blog. We introduce the post with the following thought experiment: Suppose a group of squirrels is considering two possible strategies to survive a har... Read more

Large scale experimentation

Ramesh Johari, Virag Shah, and myself wrote an academic paper on “large scale experimentation” that introduces a framework on how to think about optimal testing when there are a lot of possible experiments to run. To accompany the more technical academic paper, I wrote a blog post with a more intuitive exposition of the main ideas for the Multi... Read more