Our paper “What Makes Treatment Effects Identifiable? Characterizations and Estimators Beyond Unconfoundedness” received the Best Paper Award at COLT 2025! Huge shout-out to my amazing collaborators: Alkis Kalavasis, Katerina Mamali, @AnayMehrotra, and Manolis Zampetakis.

Anay and Alkis are on the job market this year—keep an eye out! 📄 Paper:

Our paper “What Makes Treatment Effects Identifiable? Characterizations and Estimators Beyond Unconfoundedness” received the Best Paper Award at COLT 2025! Huge shout-out to my amazing collaborators: Alkis Kalavasis, Katerina Mamali, @AnayMehrotra, and Manolis Zampetakis.

We're hiring in CS! Associate/full professor level in AI & NLP: https://t.co/DgquxY0VgS Assistant professor level in databases & data management: https://t.co/yqmxMK0c6K

Our proof builds on a new natural potential function, whose monotonicity is established using a sum-of-squares programming based computer-aided proof. For more details, see my talk at the Simons Institute for our result on EG. https://t.co/eMr8KJmdtT

For both EG and OG, we know that the last-iterate asymptotically converges, but the rate was not known despite having been studied for a long time. We obtain tight last-iterate convergence rates for both EG and OG.

We study the convex-concave min-max optimization (and more generally the monotone variation inequalities) in the constrained setting. The EG algorithm by Korpelevich '1976 and the OG algorithm by Popov '1980 are among the most classical and popular algorithms for such problems.

Very excited about our new paper with @ArgyrisOikonom1 @WeiqiangZheng3. We obtain the tight last-iterate convergence rates for the Extragradient (EG) and Optimistic Gradient (OG) algorithms, settling an open problem raised by @KonstDaskalakis. https://t.co/oBMy8THcVu

Anti-Complexitism « Shtetl-Optimized: http://bit.ly/cvMKeR