iiinova — where your digital self keeps evolving

Stephan Rabanser @stvrb.bsky.social · 23/07/2025

More on this work: 📄 Our workshop paper: openreview.net/pdf?id=qYI4f... 🖼️ Our award-winning poster: rabanser.dev/assets/poste... 🛠️ Check out the workshop for more new research on efficient on-device machine learning: ttodlerfm.gitlab.io

openreview.net

000

Stephan Rabanser @stvrb.bsky.social · 23/07/2025

Thanks to all my amazing collaborators at Google for hosting me for this internship in Zurich and for making this work possible: Nathalie Rauschmayr, Achin (Ace) Kulshrestha, Petra Poklukar, Wittawat Jitkrittum, Sean Augenstein, Congchao Wang, and Federico Tombari!

100

Stephan Rabanser @stvrb.bsky.social · 23/07/2025

In our work, we introduce Gatekeeper: a novel loss function that calibrates smaller models in cascade setups to confidently handle easy tasks while deferring complex ones. Gatekeeper significantly improves deferral performance across a diverse set of architectures and tasks.

100

Stephan Rabanser @stvrb.bsky.social · 23/07/2025

🏅 Very excited to share that my recent Google internship project on model cascading has received the 𝗕𝗲𝘀𝘁 𝗣𝗼𝘀𝘁𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 at the 𝘛𝘛𝘖𝘋𝘓𝘦𝘳-𝘍𝘔 𝘞𝘰𝘳𝘬𝘴𝘩𝘰𝘱 @ 𝘐𝘊𝘔𝘓! Thanks a lot to the organizers for setting up this amazing workshop!

100

Stephan Rabanser @stvrb.bsky.social · 11/07/2025

📄 Gatekeeper: Improving Model Cascades Through Confidence Tuning Paper ➡️ arxiv.org/abs/2502.19335 Workshop ➡️ Tiny Titans: The next wave of On-Device Learning for Foundational Models (TTODLer-FM) Poster ➡️ West Meeting Room 215-216 on Sat 19 Jul 3:00 p.m. — 3:45 p.m.

000

Stephan Rabanser @stvrb.bsky.social · 11/07/2025

📄 Selective Prediction Via Training Dynamics Paper ➡️ arxiv.org/abs/2205.13532 Workshop ➡️ 3rd Workshop on High-dimensional Learning Dynamics (HiLD) Poster ➡️ West Meeting Room 118-120 on Sat 19 Jul 10:15 a.m. — 11:15 a.m. & 4:45 p.m. — 5:30 p.m.

101

Stephan Rabanser @stvrb.bsky.social · 11/07/2025

📄 Suitability Filter: A Statistical Framework for Classifier Evaluation in Real-World Deployment Settings (✨ oral paper ✨) Paper ➡️ arxiv.org/abs/2505.22356 Poster ➡️ E-504 on Thu 17 Jul 4:30 p.m. — 7 p.m. Oral Presentation ➡️ West Ballroom C on Thu 17 Jul 4:15 p.m. — 4:30 p.m.

101

Stephan Rabanser @stvrb.bsky.social · 11/07/2025

📄 Confidential Guardian: Cryptographically Prohibiting the Abuse of Model Abstention TL;DR ➡️ We show that a model owner can artificially introduce uncertainty and provide a detection mechanism. Paper ➡️ arxiv.org/abs/2505.23968 Poster ➡️ E-1002 on Wed 16 Jul 11 a.m. — 1:30 p.m.

101

Stephan Rabanser @stvrb.bsky.social · 11/07/2025

📣 I will be at #ICML2025 in Vancouver next week to present two main conference papers (including one oral paper ✨) and two workshop papers! Say hi if you are around and want to chat about ML uncertainty & reliability! 😊 🧵 Papers in order of presentation below:

110

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🚀 Dive deeper: Paper ▶️ arxiv.org/abs/2505.23968 Code ▶️ github.com/cleverhans-l... Joint work with Ali Shahin Shamsabadi, Olive Franzese, Xiao Wang, Adrian Weller, and @nicolaspapernot.bsky.social. Talk to us at ICML in Vancouver! 🇨🇦 🧵10/10 #Abstention #Uncertainty #Calibration #ZKP #ICML2025

020

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

📚 Summary of key contributions: 1️⃣ A novel abuse threat for uncertainty quantification / abstention mechanisms. 2️⃣ Theoretical proof of attack feasibility. 3️⃣ Practical attack instantiation using the MIRAGE loss. 4️⃣ ZKP‑based CONFIDENTIAL GUARDIAN defense. 🧵9/10

100

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🔐 How Confidential Guardian works: Auditor supplies a reference dataset which has coverage over suspicious regions. 📂 Model runs inside a ZKP circuit. 🤫 Confidential Guardian releases ECE & reliability diagram—artificial uncertainty tampering pops out. 🔍📈 🧵8/10

120

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🛡️ Can we catch the Mirage? Yes—introducing CONFIDENTIAL GUARDIAN. It assesses a model's calibration properties (match of predicted probabilities and true underlying probabilities) without revealing model weights, using zero‑knowledge proofs of verified inference. 🔒🧾 🧵7/10

210

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

📉 Why is this scary? • Regulators often only monitor accuracy, not uncertainty. • Victims face delays, need to jump through extra hoops. 🔄 • Public trust in model uncertainty erodes. 😡 Abstention, once a virtue, becomes a smokescreen for discriminatory behavior. 🧵6/10

100

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

💥 Meet MIRAGE: A regularizer pushes the model's output distribution towards near‑uniform targets in any chosen region while leaving a small gap to random chance accuracy—confidence crashes 📉, accuracy stays high 📈. Result: systematic “uncertain” labels that hide bias. 🧵5/10

100

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🧠 Theoretical feasibility: We show theoretically that such uncertainty attacks work on any neural network—either repurposing hidden neurons or attaching additional fresh neurons to damp confidence. This means that no model is safe out‑of‑the‑box. 🧵4/10

100

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🎯 Presenting a new threat—Artificial Uncertainty Induction: A dishonest institution can create a model which outputs high uncertainty for inputs it dislikes, quietly shuffling people into review limbo while keeping accuracy and existing audits intact. 😈📉 🧵3/10

110

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

🔍 Background—Cautious Predictions: ML models are often designed abstain from predicting when uncertain to avoid costly mistakes (finance, healthcare, justice, autonomous driving). But what if that safety valve becomes a backdoor for discrimination? 🚪⚠️ 🧵2/10

100

Stephan Rabanser @stvrb.bsky.social · 02/06/2025

📢 New ICML 2025 paper! Confidential Guardian: Cryptographically Prohibiting the Abuse of Model Abstention 🤔 Think model uncertainty can be trusted? We show that it can be misused—and how to stop it! Meet Mirage (our attack💥) & Confidential Guardian (our defense🛡️). 🧵1/10

133

Stephan Rabanser @stvrb.bsky.social · 10/03/2025

Starting off this account with a banger: In September 2025, I will be joining @princetoncitp.bsky.social at Princeton University as a Postdoc working with @randomwalker.bsky.social & @msalganik.bsky.social! I am very excited about this opportunity to continue my work on trustworthy/reliable ML! 🥳

0122