Martin Vechev
Full Professor
Leads the Secure, Reliable and Intelligent Systems Lab, working on certified robustness of neural networks, safety and security of large language models, and machine learning for code.
Martin Vechev's lab at ETH Zurich works on trustworthy & safe ML and large language models. 85 papers since 2023, 37 at top venues, mostly ICLR and ICML. 8 master's students. 26 alumni, most often to Anthropic and Daedalean AI.
At a glance
- papers, 2023–2026
- 22/ year
- at T1 venues
- 44%
- PhD students
- 0
- MS students
- 8
- alumni listed
- 26
Papers by yearfrom the lab homepage, 2023–2026
- 25 papers2026 · T1 9 · collecting
- 23 papers2025 · T1 10
- 23 papers2024 · T1 12
- 14 papers2023 · T1 6
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, PLDI, CCS
Research focuspaper titles that use each topic's words
- Large language models45 papers
- code7 papers
- secure2 papers
- Trustworthy & safe ML0 papers
- intelligent systems0 papers
Paper list
From the lab homepage.
202625 papers
- ICML (T1)Adaptive Generation of Bias-Eliciting Questions for LLMs (opens in a new tab)
- ICML (T1)AutoBaxBuilder: Bootstrapping Code Security Benchmarking (opens in a new tab)
- AI4Math@ICMLBeyond Benchmarks: MathArena as an Evaluation Platform for Mathematics with LLMs (opens in a new tab)
- ICML SpotlightBrokenMath: A Benchmark for Sycophancy in Theorem Proving with LLMs (opens in a new tab)
- ICML (T1)CodeTaste: Can LLMs Generate Human-Level Code Refactorings? (opens in a new tab)
- COLMCoding Agents Don't Know When to Act (opens in a new tab)
- ICLR DL4C @ NeurIPS'25 OralConstrained Decoding of Diffusion LLMs with Context-Free Grammars (opens in a new tab)
- COLMDelay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR (opens in a new tab)
- ICLR (T1)Dual Randomized Smoothing: Beyond Global Noise Variance (opens in a new tab)
- MemAgents @ ICLR Oral & Runner-up Best PaperEvaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents? (opens in a new tab)
- NeurIPS (T1)Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermark (opens in a new tab)
- ICLR * Equal contributionExpressiveness of Multi-Neuron Convex Relaxations in Neural Network Certification (opens in a new tab)
202523 papers
- ICML (T1)A Unified Approach to Routing and Cascading for LLMs (opens in a new tab)
- ICML * Equal contributionAutomated Benchmark Generation for Repository-Level Coding Tasks (opens in a new tab)
- ICML * Equal contributionAverage Certified Radius is a Poor Metric for Randomized Smoothing (opens in a new tab)
- ICML SpotlightBaxBench: Can LLMs Generate Secure and Correct Backends? (opens in a new tab)
- ICML * Equal contributionBlack-Box Adversarial Attacks on LLM-Based Code Completion (opens in a new tab)
- ICLR (T1)Black-Box Detection of Language Model Watermarks (opens in a new tab)
- ICML (T1)CTBENCH: A Library and Benchmark for Certified Training (opens in a new tab)
- ICML (T1)Discovering Spoofing Attempts on Language Model Watermarks (opens in a new tab)
- TMLR (T1)Gaussian Loss Smoothing Enables Certified Training with Tight Convex Relaxations (opens in a new tab)
- ICLR (T1)GRAIN: Exact Graph Reconstruction from Gradients (opens in a new tab)
- ICLR (T1)Large Language Models are Advanced Anonymizers (opens in a new tab)
- NeurIPS Datasets and BenchmarksMathArena: Evaluating LLMs on Uncontaminated Math Competitions (opens in a new tab)
202423 papers
- NeurIPS Datasets and BenchmarksA Synthetic Dataset for Personal Attribute Inference (opens in a new tab)
- ICLR Spotlight, PPPM-AwardBeyond Memorization: Violating Privacy Via Inference with Large Language Models (opens in a new tab)
- NeurIPS (T1)ConStat: Performance-Based Contamination Detection in Large Language Models (opens in a new tab)
- Workshop on Mathematical Reasoning, NeurIPSConstraint-Based Synthetic Data Generation for LLM Mathematical Reasoning (opens in a new tab)
- ICLR SpotlightControlled Text Generation via Language Model Arithmetic (opens in a new tab)
- ICML (T1)CuTS: Customizable Tabular Synthetic Data Generation (opens in a new tab)
- NeurIPS (T1)DAGER: Exact Gradient Inversion for Large Language Models (opens in a new tab)
- NeurIPS NextGenAISafety@ICML24 OralExploiting LLM Quantization (opens in a new tab)
- ICLR * Equal contributionExpressivity of ReLU-Networks under Convex Relaxations (opens in a new tab)
- S&P (T1)From Principle to Practice: Vertical Data Minimization for Machine Learning (opens in a new tab)
- ICML (T1)Guiding LLMs The Right Way: Fast, Non-Invasive Constrained Generation (opens in a new tab)
- ICLR (T1)Hiding in Plain Sight: Disguising Data Stealing Attacks in Federated Learning (opens in a new tab)
202314 papers
- PLDI (T1)Abstract Interpretation of Fixpoint Iterators with Applications to Neural Networks (opens in a new tab)
- Quantum JournalAbstraqt: Analysis of Quantum Circuits via Abstract Stabilizer Simulation (opens in a new tab)
- NeurIPS * Equal contributionAutomated Classification of Model Errors on ImageNet (opens in a new tab)
- ICLR * Equal contribution SpotlightCertified Training: Small Boxes are All You Need (opens in a new tab)
- NeuIPSConnecting Certified and Adversarial Training (opens in a new tab)
- ICLR (T1)Efficient Certified Training and Robustness Verification of Neural ODEs (opens in a new tab)
- ICML (T1)FARE: Provably Fair Representation Learning with Practical Certificates (opens in a new tab)
- CCS (T1)Group and Attack: Auditing Differential Privacy (opens in a new tab)
- ICLR SpotlightHuman-Guided Fair Classification for Natural Language Processing (opens in a new tab)
- Knowlege and Logical Reasoning Workshop -- ICML * Equal contributionLarge Language Models are Zero-Shot Multi-Tool Users (opens in a new tab)
- ACM CCS Distinguished Paper AwardLarge Language Models for Code: Security Hardening and Adversarial Testing (opens in a new tab)
- Neural Conversational AI Workshop, TEACH -- ICML * Equal contributionLMQL Chat: Scripted Chatbot Development (opens in a new tab)
Where alumni went26 listed, 57% to industry
Industry 4Academia 3
Most often to
- Anthropic
- Daedalean AI
- Google DeepMind
Also to
- UIUC
- UC Berkeley
Apply to ETH ZurichRolling
Not announced yet
Computer Science Doctorate (D-INFK)
- Round
- Rolling (no deadline)
- Opens
- —
- GRE
- See program page
- English
- Proficiency in written and spoken English
- Fee
- See program page
Confirmed on the program page.
Similar labsnearest by research summary
- GBCambridge
Ferenc Huszár
7 T1 - USUSC
Bart Andrew Kosko
0 T1 - HKCUHK
Weiyang Liu
- Generative & foundation models
- USUW
Radha Poovendran
- Trustworthy & safe ML
- DETUM
Zeynep Akata
- Trustworthy & safe ML
- USMichigan
Atul Prakash
- Trustworthy & safe ML
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.