The recording of this week's talk, by Palak Jain (@thepalakjain.bsky.social), is now available online as well along with the slides:

"Enforcing Demographic Coherence: A Harms-Aware Framework for Reasoning about Private Data Release"
www.tcsplus.org/welcome/past...

extra cute that I'm in this exact same situation 🫣

I use 'Souper Cubes' to freeze food 🥘

Dal and cooked rice also freeze well and are great for when you want something simple to eat.

Aside from the stuff that's already been mentioned, I find freezing a variety of curries is a great way to have a customisable meal for later, especially for difficult weeks. (Just warm up a curry of your choice and add veggies/paneer/fish that you have lying around)

I think you're exactly right. I can tell the poems apart precisely because the LLM poems read like they were written by a middle-schooler 😂

(8/8) Finally, I'm looking for a research internship this coming year and would love to set up a time to talk!

(7/8) We prove that this is the best achievable error bound that depends only on w, for a large range of values of w. When w is small, the error of our mechanism is similar to the polylogarithmic in T error in the insertion-only setting, bypassing the hardness in the turnstile model.

(6/8) We present an item-level differentially private mechanism that, for all turnstile streams with maximum flippancy w, continually outputs the number of distinct elements with an O(√w · poly log T ) additive error, without requiring prior knowledge of w.

(5/8) Specifically, the maximum flippancy is the largest number of times that the contribution of a single item to the distinct elements count changes over the course of the stream.

(4/8) We show that in the worst-case, every DP mechanism has additive error at least T^1/4! However, we can do much better in certain settings. -- We identify a parameter, maximum flippancy, that is low for natural data streams and for which we give tight parameterized error guarantees.

(3/8) With only insertions, existing algorithms have additive error just polylogarithmic in the length of the stream T. We uncover a much richer landscape in the turnstile model, even without considering memory restrictions.

(2/8) We consider the achievable error for differentially private continual release of a basic statistic—the number of distinct items—in a stream where items may be both inserted and deleted (the turnstile setting.)