Loris D'Antoni @lorisdanto.bsky.social · 20/08/2026New funding opportunity in verification/PL research lnkd.in/p/gcDbkmKylnkd.inCode Metal is launching the Verified Code Translation Research Awards, a new program supporting university researchers advancing the scientific foundations of trustworthy code translation. Much of… |...Code Metal is launching the Verified Code Translation Research Awards, a new program supporting university researchers advancing the scientific foundations of trustworthy code translation. Much of wh... 000
Loris D'Antoni @lorisdanto.bsky.social · 07/07/2026This year's FLOC will include RajeevFest, a special workshop to celebrate the career of my PhD advisor Rajeev Alur! lorisdanto.github.io/rajeevfest.g...lnkd.inLinkedInThis link will take you to a page that’s not on LinkedIn 030
Loris D'Antoni @lorisdanto.bsky.social · 10/06/2026Why is it called a Large Language Model and not a Word Wide Web? 120
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026Paper: Constrained Adaptive Rejection Sampling arxiv.org/abs/2510.01902 Paweł Parys, Sairam Vaidya, Taylor Berg-Kirkpatrick, and Loris D'Antoniarxiv.orgConstrained Adaptive Rejection SamplingLanguage Models (LMs) are increasingly used in applications where generated outputs must satisfy strict semantic or syntactic constraints. Existing approaches to constrained generation fall along a sp... 051
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026This work shows that the structure of constraint checkers can be exploited to dramatically accelerate sampling without sacrificing correctness; reminiscent of how conflict-driven clause learning transformed SAT solving by extracting and reusing information from failed search attempts. 6/ 130
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026On some of our benchmarks, acceptance rates improve from roughly 1 valid sample out of every 100 attempts to nearly 1 out of every 3, while preserving exact sampling guarantees. 5/ 110
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026Constrained Adaptive Rejection Sampling (CARS) exploits this information to incrementally build a representation of the constrained distribution. As more samples are processed, CARS learns to avoid large classes of failures, while still producing samples from the exact target distribution. 4/ 110
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026Every interaction with the checker, whether a sample is accepted or rejected, provides information about the structure of the constrained space and where future failures are likely to occur. 3/ 110
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026In our ICML26 paper, we show that autoregressive constraint checkers reveal rich information that can be used to drastically reduce the rejection rate any autoregressive sampling algorithm. 2/ 120
Loris D'Antoni @lorisdanto.bsky.social · 04/06/2026Getting LLMs to generate samples that satisfy constraints is a fundamental problem in machine learning. The standard approach is rejection sampling: generate a sample, check whether it satisfies the constraint, and if not, try again. 1/ 170
Loris D'Antoni @lorisdanto.bsky.social · 01/06/2026Paper: Continuous Diffusion Models Can Obey Formal Syntax arxiv.org/abs/2602.12468 /the endarxiv.orgContinuous Diffusion Models Can Obey Formal SyntaxDiffusion language models offer a promising alternative to autoregressive models due to their global, non-causal generation process, but their continuous latent dynamics make discrete constraints -- e... 020
Loris D'Antoni @lorisdanto.bsky.social · 01/06/2026The result is a training-free constrained generation framework for continuous diffusion language models that can exactly enforce formal syntax constraints while provably preserving generation quality. 4/ 100
Loris D'Antoni @lorisdanto.bsky.social · 01/06/2026Instead of training a classifier for every constraint, we analytically compute how likely partially denoised states are to satisfy the constraint imposed by the regex, then use those probabilities to compute gradients that guide generation toward constraint satisfaction. 3/ 100
Loris D'Antoni @lorisdanto.bsky.social · 01/06/2026Constrained generation has become a powerful tool for autoregressive language models. But continuous diffusion models generate text in fundamentally different ways, making most existing control techniques inapplicable. 2/ 100
Loris D'Antoni @lorisdanto.bsky.social · 01/06/2026At ICML, Jinwoo Kim will present the first approach for enforcing formal structural constraints like regexes, grammars, and schemas on the output of continuous diffusion language models, entirely without retraining. 1/ 180
Loris D'Antoni @lorisdanto.bsky.social · 27/02/2026Lots of people at SoCalPLS, thanks Mukund Raghothaman for organizing! 020
Loris D'Antoni @lorisdanto.bsky.social · 23/02/2026Two months later, same author, new paper, same story. 060
Loris D'Antoni @lorisdanto.bsky.social · 21/02/2026If Klaebo was a country, he would rank 9th on the Olympics medal table 100
Loris D'Antoni @lorisdanto.bsky.social · 18/01/2026Apparently alcohol is not considered a performance enhancing drug (thus not banned by anti doping) in most sports, but it is considered a performance enhancing drug in the sport of... rifle shooting dailynebraskan.com share.google/rWoVb5DR6mnn...dailynebraskan.comdailynebraskan.com | Student voice of the University of Nebraska-Lincoln since 1901 130
Loris D'Antoni @lorisdanto.bsky.social · 15/01/2026@shriram.bsky.social I remember you had a course where you would give students homework where you had the same language syntax but you would change the semantics and the students had to figure out what was going on. I can't find it though. Did I hallucinate it? 160
Loris D'Antoni @lorisdanto.bsky.social · 22/12/2025For those wondering Baby: you wash your hands before touching them Toddler: you wash your hands after touching them 080
Loris D'Antoni @lorisdanto.bsky.social · 16/12/2025I think we're all guilty of occasionally slipping in a self citation, but adding five/six papers in a completely irrelevant area seems like really bad practice 2390
Loris D'Antoni @lorisdanto.bsky.social · 16/12/2025I was reading a paper (by a full prof!) that included a paragraph in related work that was completely irrelevant and was there just to cite a bunch of the author's own work. This type of behavior makes me not trust the work of such authors. 3374854
Loris D'Antoni @lorisdanto.bsky.social · 07/11/2025Google Scholar has stopped showing one of my POPL papers randomly. Do people know how to fix this? I don't care particularly, but for my student this is his top paper and it's annoying for him 300
Loris D'Antoni @lorisdanto.bsky.social · 23/10/2025Apparently lots of friends will be in SD for NeurIPS. 010
Loris D'Antoni @lorisdanto.bsky.social · 23/09/2025Who teaches an undergraduate principles of programming languages class? Looking for some inspiration to teach one at UCSD 261
Loris D'Antoni @lorisdanto.bsky.social · 16/09/2025The world’s largest federated conference in logic & automated reasoning, FLoC 2026 (www.floc26.org), will be in Lisbon! Thousands will attend—an excellent chance for companies to showcase their presence. Interested in sponsoring? Details www.floc26.org/sponsors/FLo...floc26.orgFLoC 2026: The 9th Federated Logic Conference 065
Loris D'Antoni @lorisdanto.bsky.social · 09/08/2025I've been added to the "girls' mums" group chat of my daughter's school. How to proceed? 150
Loris D'Antoni @lorisdanto.bsky.social · 18/07/2025Today's Wordle celebrates a person who has provided the inspiration behind much of my work 020
Loris D'Antoni @lorisdanto.bsky.social · 13/07/2025I actually only watch anime and do magic tricks in the list. Occasionally build models (but from data and using gradient decent) 020
Loris D'Antoni @lorisdanto.bsky.social · 13/07/2025It's a miracle I've found a partner in my life. 270
Loris D'Antoni @lorisdanto.bsky.social · 29/06/2025Growing fresh tomatoes is hard work, but it really pays off once your 1yo picks them up when you are not watching and squishes them into flat mush 080
Loris D'Antoni @lorisdanto.bsky.social · 11/06/2025I can see why the student thinks I always wear the "hooded long blue long-sleeve shirt" 030
Loris D'Antoni @lorisdanto.bsky.social · 11/06/2025The last question for my CSE230 final was “Draw a comic about anything we learned in this class (e.g., Proving theorems with Coq, types, functional programming, Loris, etc.)” Here are the best ones. 2224
Loris D'Antoni @lorisdanto.bsky.social · 06/06/2025What version of the Turing test is "Can a human pretend to be an AI?" 240
Reposted by Loris D'AntoniSMBC comic Bot @smbc-bot.bsky.social · 15/01/2025 Saturday Morning Breakfast Cereal - Applied www.smbc-comics.com/comic/applied "You are all impure now!" 0185
Loris D'Antoni @lorisdanto.bsky.social · 17/04/2025Teaching my research group how to label their axes 1210
Loris D'Antoni @lorisdanto.bsky.social · 12/04/2025This is wild. And so cool that every so often you can catch her climb in San Diego 0100
Reposted by Loris D'AntoniNicole Micheroni @nicolemicheroni.bsky.social · 12/04/2025Some personal news: the Department of Homeland Security has given me, an immigration lawyer born in Newton, Massachusetts, seven days to leave the U.S. Does anyone know if you can get Italian citizenship through great-grandparents? 2227257389780
Loris D'Antoni @lorisdanto.bsky.social · 10/04/2025Why is it called Apple TV+ and not Severance package? 0110
