Professor at Penn, Amazon Scholar at AWS. Interested in machine learning, uncertainty quantification, game theory, privacy, fairness, and most of the intersections thereinhttps://bsky.app/profile/aaroth.bsky.socialhttps://bsky.app/profile/aaroth.bsky.social/post/3mki7bzozqs23We updated our paper --- and solved the open problem highlighted in the old version. Now our lower bound construction has only polylog(1/eps) many groups instead of poly(1/eps) many groups. The construction is also simplified. [contains quote post or other embedded content]27 Apr 2026 13:44 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mki7bzozqs23https://bsky.app/profile/aaroth.bsky.social/post/3mkak6omyt225How many samples do you need from an unknown distribution in order to train a model with multicalibration error at most epsilon? Answer: 1/epsilon^3 samples is both necessary and sufficient.24 Apr 2026 12:38 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mkak6omyt225https://bsky.app/profile/aaroth.bsky.social/post/3mk22hl567k26Say hi to @marcelhussing.bsky.social at ICLR [contains quote post or other embedded content]21 Apr 2026 22:40 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mk22hl567k26https://bsky.app/profile/aaroth.bsky.social/post/3mjmg53ykxs2zI've recently been getting invitations to talk about how to use AI tools to assist with TCS research. Its something I've been doing a lot, but don't have structured thoughts about how to explain process. But I'm going to try -- first such talk is tomorrow: t.co/wlHPBzXzDm https://t.co/wlHPBzXzDm16 Apr 2026 12:32 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mjmg53ykxs2zhttps://bsky.app/profile/aaroth.bsky.social/post/3mjfniwqerc2pAI Agents like Codex are very good at figuring out taxes, including obscure local ones that Intuit doesn't bother with (looking at you, Philadelphia local taxes). Businesses that provide financial/legal services that involve reasoning through dense but public documentation are in trouble.13 Apr 2026 19:55 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mjfniwqerc2phttps://bsky.app/profile/aaroth.bsky.social/post/3mgvnnikwcc2zAlpha_0 joke from Dogman13 Mar 2026 00:25 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mgvnnikwcc2zhttps://bsky.app/profile/aaroth.bsky.social/post/3mguksyjhus2uVery cool work. Empirical science has many researcher-degrees-of-freedom which makes it hard to interpret specific studies --- these are only a single trajectory through the data analysis multiverse. Human researchers are opaque. But with agents you can explore the whole space! [contains quote post or other embedded content]12 Mar 2026 14:01 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mguksyjhus2uhttps://bsky.app/profile/aaroth.bsky.social/post/3mguhzdqlx22iNeural networks are highly non-convex, so approximate error minimizers need not look anything like each other in parameter space. But we show that nevertheless (for many model sizes) approximate error minimizers must closely agree in function/prediction space despite this!12 Mar 2026 13:11 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mguhzdqlx22ihttps://bsky.app/profile/aaroth.bsky.social/post/3mgniuktkkk2eMichael @mkearnsphilly.bsky.social ) and I wrote a blog post about our experiences using AI for research, and our thoughts on what these developments will mean for research, publication, and education: https://www.amazon.science/blog/how-ai-is-changing-the-nature-of-mathematical-research09 Mar 2026 18:38 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mgniuktkkk2ehttps://bsky.app/profile/aaroth.bsky.social/post/3mgniuktkkk2eMichael @mkearnsphilly.bsky.social ) and I wrote a blog post about our experiences using AI for research, and our thoughts on what these developments will mean for research, publication, and education: https://www.amazon.science/blog/how-ai-is-changing-the-nature-of-mathematical-research09 Mar 2026 18:38 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mgniuktkkk2ehttps://bsky.app/profile/aaroth.bsky.social/post/3mgauuz2pv22oWhich is the better model for math? GPT 5.2, or GPT 5.3 codex (either one on high reasoning)?04 Mar 2026 18:08 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mgauuz2pv22ohttps://bsky.app/profile/aaroth.bsky.social/post/3mbylnwmhi22mExcited about a new paper! Multicalibration turns out to be strictly harder than marginal calibration. We prove tight Omega(T^{2/3}) lower bounds for online multicalibration, separating it from online marginal calibration for which better rates were recently discovered.09 Jan 2026 13:21 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mbylnwmhi22mhttps://bsky.app/profile/aaroth.bsky.social/post/3mb7mldewh22hYes. We already have a set of ingrained red flags for human written papers that signal a lack of care: not citing the relevant literature, not formatting or typesetting math correctly, etc. These don't mean the paper is wrong but they strongly correlate with lack of care. But... [contains quote post or other embedded content]30 Dec 2025 15:01 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3mb7mldewh22hhttps://bsky.app/profile/aaroth.bsky.social/post/3majfrvog2s2o2025 was an eventful/disruptive year for computer science research, for two reasons: 1) a shock to federal funding, and 2) the arrival of AI models capable enough to assist mathematical research. 1) is unambiguously bad and 2) is probably mostly good. I'll write about AI first.21 Dec 2025 19:01 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3majfrvog2s2ohttps://bsky.app/profile/aaroth.bsky.social/post/3m5c7vxj44c2nDid your fairness/privacy/CS&Law/etc paper just get rejected from ITCS? Oh FORC! Submit tomorrow and join us at Harvard this summer. [contains quote post or other embedded content]10 Nov 2025 18:12 +0000at://did:plc:3q2kaxhjkceuc7kj4dmtfstl/app.bsky.feed.post/3m5c7vxj44c2n