134 citations · 776 across the 124 of their papers we have counts for
71 papers · 1 filter
Leveraging Instruction Tuning and Merging for Reasoning Model Adaptation
Yu-Du Feng, Niels Mündler-Sasahara, Mark Vero +1
Reasoning language models (RLMs) demonstrate impressive performance by leveraging test-time compute in the form of reasoning tokens. However, this behavior makes adapting RLMs to n…
Widening the Gap: Exploiting LLM Quantization via Outlier Injection
Xiaohua Zhan, Kazuki Egashira, Robin Staab +2
LLM quantization has become essential for memory-efficient deployment. Recent work has shown that quantization schemes can pose critical security risks: an adversary may release a…
Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR
Kazuki Egashira, Mark Vero, Jasper Dekoninck +3
Reinforcement Learning with Verifiable Rewards (RLVR) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs). While RLVR is designe…
Dual Randomized Smoothing: Beyond Global Noise Variance
Chenhao Sun, Yuhao Mao, Martin Vechev
Randomized Smoothing (RS) is a prominent technique for certifying the robustness of neural networks against adversarial perturbations. With RS, achieving high accuracy at small rad…
SPEAR++: Scaling Gradient Inversion via Sparsely-Used Dictionary Learning
Alexander Bakarsky, Dimitar I. Dimitrov, Maximilian Baader +1
Federated Learning has seen an increased deployment in real-world scenarios recently, as it enables the distributed training of machine learning models without explicit data sharin…
Pay Attention to the Triggers: Constructing Backdoors That Survive Distillation
Giovanni De Muri, Mark Vero, Robin Staab +1
LLMs are often used by downstream users as teacher models for knowledge distillation, compressing their capabilities into memory-efficient models. However, as these teacher models…
Fewer Weights, More Problems: A Practical Attack on LLM Pruning
Kazuki Egashira, Robin Staab, Thibaud Gloaguen +2
Model pruning, i.e., removing a subset of model weights, has become a prominent approach to reducing the memory footprint of large language models (LLMs) during inference. Notably,…
Watermarking Diffusion Language Models
Thibaud Gloaguen, Robin Staab, Nikola Jovanović +1
We introduce the first watermark tailored for diffusion language models (DLMs), an emergent LLM paradigm able to generate tokens in arbitrary order, in contrast to standard autoreg…
Constrained Decoding of Diffusion LLMs with Context-Free Grammars
Niels Mündler, Jasper Dekoninck, Martin Vechev
Large language models (LLMs) have shown promising performance across diverse domains. Many practical applications of LLMs, such as code completion and structured data extraction, r…
SoK: Data Minimization in Machine Learning
Robin Staab, Nikola Jovanović, Kimberly Mai +4
Data minimization (DM) describes the principle of collecting only the data strictly necessary for a given task. It is a foundational principle across major data protection regulati…
Watermarking Autoregressive Image Generation
Nikola Jovanović, Ismail Labiad, Tomáš Souček +2
Watermarking the outputs of generative models has emerged as a promising approach for tracking their provenance. Despite significant interest in autoregressive image generation mod…
Type-Constrained Code Generation with Language Models
Niels Mündler, Jingxuan He, Hao Wang +3
Large language models (LLMs) have achieved notable success in code generation. However, they still frequently produce uncompilable output because their next-token inference procedu…
MixAT: Combining Continuous and Discrete Adversarial Training for LLMs
Csaba Dékány, Stefan Balauca, Robin Staab +2
Despite recent efforts in Large Language Model (LLM) safety and alignment, current adversarial attacks on frontier LLMs can still consistently force harmful generations. Although a…
Watch your steps: Dormant Adversarial Behaviors that Activate upon LLM Finetuning
Thibaud Gloaguen, Mark Vero, Robin Staab +1
Finetuning open-weight Large Language Models (LLMs) is standard practice for achieving task-specific performance improvements. Until now, finetuning has been regarded as a controll…
GRAIN: Exact Graph Reconstruction from Gradients
Maria Drencheva, Ivo Petrov, Maximilian Baader +2
Federated learning claims to enable collaborative model training among multiple clients with data privacy by transmitting gradient updates instead of the actual client data. Howeve…
AlphaIntegrator: Transformer Action Search for Symbolic Integration Proofs
Mert Ünsal, Timon Gehr, Martin Vechev
We present the first correct-by-construction learning-based system for step-by-step mathematical integration. The key idea is to learn a policy, represented by a GPT transformer mo…
Average Certified Radius is a Poor Metric for Randomized Smoothing
Chenhao Sun, Yuhao Mao, Mark Niklas Müller +1
Randomized smoothing (RS) is popular for providing certified robustness guarantees against adversarial attacks. The average certified radius (ACR) has emerged as a widely used metr…
Expressiveness of Multi-Neuron Convex Relaxations in Neural Network Certification
Yuhao Mao, Yani Zhang, Martin Vechev
Neural network certification methods heavily rely on convex relaxations to provide robustness guarantees. However, these relaxations are often imprecise: even the most accurate sin…
Ward: Provable RAG Dataset Inference via LLM Watermarks
Nikola Jovanović, Robin Staab, Maximilian Baader +1
RAG enables LLMs to easily incorporate external data, raising concerns for data owners regarding unauthorized usage of their content. The challenge of detecting such unauthorized u…
Mitigating Catastrophic Forgetting in Language Transfer via Model Merging
Anton Alexandrov, Veselin Raychev, Mark Niklas Müller +3
As open-weight large language models (LLMs) achieve ever more impressive performances across a wide range of tasks in English, practitioners aim to adapt these models to different…
A Synthetic Dataset for Personal Attribute Inference
Hanna Yukhymenko, Robin Staab, Mark Vero +1
Recently, powerful Large Language Models (LLMs) have become easily accessible to hundreds of millions of users world-wide. However, their strong capabilities and vast world knowled…
CTBENCH: A Library and Benchmark for Certified Training
Yuhao Mao, Stefan Balauca, Martin Vechev
Training certifiably robust neural networks is an important but challenging task. While many algorithms for (deterministic) certified training have been proposed, they are often ev…
Back to the Drawing Board for Fair Representation Learning
Angéline Pouget, Nikola Jovanović, Mark Vero +2
The goal of Fair Representation Learning (FRL) is to mitigate biases in machine learning models by learning data representations that enable high accuracy on downstream tasks while…
Exploiting LLM Quantization
Kazuki Egashira, Mark Vero, Robin Staab +2
Quantization leverages lower-precision weights to reduce the memory usage of large language models (LLMs) and is a key technique for enabling their deployment on commodity hardware…