Mona Schirmer (@monaschir) profile picture

PhD Candidate @AmlabUva @ELLISforEurope | AI safety & robustness | Alumni @Amazon @WorldBank @ENSAEparis @TUBerlin

Amsterdam, Niederlande
Joined January 2018
Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)Photo by Mona Schirmer (@monaschir)
🚨 Big announcement on something we've been working on for 2 years now - the Center for Human Progress. The world is full of great ideas that ought to "unstick" progress everywhere. What often stops them from working is the human layer: our norms, culture, & behaviors - corruption, misaligned incentives, and identity politics. The center is a think-do tank - cultural evolution meets behavioral science and economics - with new ways to think about cooperation and innovation. 1/
Introducing the Center for Human Progress, a think-and-do tank rooted in cultural evolution, behavioral science, and economics. The world isn't short of great ideas. It's short of the systems to deliver them. The barriers to implementation are typically in the people layer - norms, incentives, and other aspects of culture. 🧵 centerforhumanprogress.org/
17
40
7
240
25,197
Mona Schirmer retweeted
Excited to be at #ICLR2026 in Rio this week! Presenting “Soft Tokens, Hard Truths” Saturday 10.30am (Pavilion 3, P3-#1020). Feel free to DM me to chat about self-improvement, reasoning, code gen. I’m also on the job market for industry research positions.
🔥New preprint: Soft Tokens, Hard Truths Introduces the first scalable continuous-token RL method for LLMs - no reference CoTs needed; scales to hundreds of thought tokens. Best to train soft, infer hard! Pass@1 parity ⚖️, Pass@32 gains 📈& better robustness 🛡️ vs. hard CoT 1/🧵
6
22
3,642
Mona Schirmer retweeted
Excited to share that we're looking for a new colleague at @AMLab_UvA : Assistant Professor in AI for Science 🔬🤖 AMLab is a world-class ML research group embedded in Amsterdam's thriving AI ecosystem: leading research groups, an ELLIS unit, startups, and big tech — all within reach. And Dutch academic labor conditions are genuinely among the best in Europe ❤️ Deadline: May 30 👉 werkenbij.uva.nl/en/vacancie…
2
36
7
202
33,440
Mona Schirmer retweeted
Presenting today at @NeurIPSConf from 11am-2pm (poster number #3904)👇👇
TTA Monitor is coming to #NeurIPS2025! Our monitor flags risk increase of a continuously adapting model without any labels and with false-alarm guarantees. Let’s dive in! 🧵 1/6 📄 Paper: arxiv.org/abs/2507.08721 📌 Poster: neurips.cc/virtual/2025/loc/…
2
1
4
1,758
TTA Monitor is coming to #NeurIPS2025! Our monitor flags risk increase of a continuously adapting model without any labels and with false-alarm guarantees. Let’s dive in! 🧵 1/6 📄 Paper: arxiv.org/abs/2507.08721 📌 Poster: neurips.cc/virtual/2025/loc/…
5
3
2
11
1,558
5/ The monitor reliably detects risk violations across TTA methods and datasets and even detects TTA collapse!
2
71
4/ Our bound is unsupervised and uses only model uncertainty as a loss proxy. Because we calibrate the proxy threshold online, the bound remains valid and tight even under severe shift and model adaptation (e.g. via entropy minimization).
2
65
3/ Our monitor lower-bounds the test risk (🟩) and raises an alarm once the confidence lower bound (🟦) exceeds a pre-defined risk level (⬛). Confidence sequences ensure that the alarm remains anytime-valid with statistical guarantees on the false-alarm rate.
2
70
2/ Test-Time Adaptation (TTA) helps models adjust to distribution shift on the fly without labels. But TTA can silently fail or collapse (e.g., by predicting a single class ⬆️)!
2
83
Mona Schirmer retweeted
🔥New preprint: Soft Tokens, Hard Truths Introduces the first scalable continuous-token RL method for LLMs - no reference CoTs needed; scales to hundreds of thought tokens. Best to train soft, infer hard! Pass@1 parity ⚖️, Pass@32 gains 📈& better robustness 🛡️ vs. hard CoT 1/🧵
6
52
1
277
40,059
Come by our Oral on Monitoring Risks in Test-time Adaptation at the @PUT_TTA_ICML25 #ICML Workshop ⏱️: Friday, 10am PDT 🚩: West Meeting Room 220-222 Work with @MetodJazbec @canaesseth @eric_nalisnick
5
11
1,052
Mona Schirmer retweeted
Come say hi to me, @AlexTimans, and @eric_nalisnick today between 11am-2pm in East Exhibit Hall A-C #4505 . We'll be presenting our Fast yet Safe paper on efficient inference and risk control 👇
Hey! Catch me next week at @NeurIPSConf where we’ll be presenting our Fast yet Safe paper. We show how to equip efficient AI methods like early-exiting or soft speculative decoding with performance guarantees using risk control! arxiv.org/abs/2405.20915 🧵🔽
1
1
17
676
Good morning NeurIPS! Excited to present "Generative Modelling of Structurally Constrained Graphs" with Clément Vignac, @DorinaThanou , and @pafrossard. Come find me at poster #2704 in East Exhibit Hall A-C from 16:30-19:30 if you're interested in constrained graph diffusion!
4
1
14
1,350
Excited to see #ImpactAI debuting at @Googleorg 's Impact Summit! Very grateful to have been part of the journey over the last two years. Check out how #ImpactAI helps to make better decisions in development economics ⬇️
This Post is from an account that no longer exists. Learn more
4
389
Mona Schirmer retweeted
👋I am on the academic job market, looking for tenure-track positions. I work on machine learning, decision making, and social aspects of AI. Let's get in touch if your institution is hiring! 💻stsirtsis.github.io 😀Shares are very much appreciated!
2
19
2
51
9,093