Differentially Private Domain Discovery
WGM-based methods provide efficient domain discovery with near-optimal guarantees for missing mass on Zipfian data.
Avg rating: 6.50 (6–8) · Vinod Raman et al.
Safety, Privacy & Alignment
AI safety, alignment, jailbreaks, adversarial robustness, privacy, differential privacy, and membership inference.
All papers
Min rating
EigenBench: A Comparative Behavioral Measure of Value Alignment
EigenBench measures language model value alignment using model ensemble judgments aggregated with EigenTrust without ground truth labels.
Avg rating: 6.00 (4–10) · Jonathn Chang et al.
Every Language Model Has a Forgery-Resistant Signature
Ellipse signatures function as forgery-resistant model output identifiers based on high-dimensional geometric constraints.
Avg rating: 6.00 (4–8) · Matthew Finlayson et al.
Gaussian certified unlearning in high dimensions: A hypothesis testing approach
Analyzes machine unlearning in high dimensions showing single noisy Newton step with Gaussian noise suffices for privacy-accuracy.
Avg rating: 6.00 (4–8) · Aaradhya Pandey et al.
Hubble: a Model Suite to Advance the Study of LLM Memorization
Releases Hubble suite of open-source LLMs with controlled perturbed variants to systematically study memorization risks.
Avg rating: 7.50 (6–8) · Johnny Wei et al.
LLM Fingerprinting via Semantically Conditioned Watermarks
Introduces semantically conditioned watermarks for robust and stealthy LLM fingerprinting robust to deployment scenarios.
Avg rating: 6.50 (6–8) · Thibaud Gloaguen et al.
Omni-Reward: Towards Generalist Omni-Modal Reward Modeling with Free-Form Preferences
Omni-Reward addresses modality imbalance and preference rigidity with omni-modal reward modeling framework.
Avg rating: 6.50 (6–8) · Zhuoran Jin et al.
PateGAIL++: Utility Optimized Private Trajectory Generation with Imitation Learning
PATEGAIL++ privacy-preserving trajectory generation framework using sensitivity-aware noise allocation for improved privacy-utility trade-off.
Avg rating: 5.00 (4–6) · Yingjie Ma et al.
RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
Introduces RedTeamCUA framework with hybrid web-OS sandbox for adversarial testing of computer-use agents.
Avg rating: 6.00 (4–8) · Zeyi Liao et al.
Reliable Weak-to-Strong Monitoring of LLM Agents
MetamerGen generates scene metamers aligned with human perception using foveal/peripheral features and latent diffusion.
Avg rating: 6.00 (4–8) · Neil Kale et al.
Spherical Watermark: Encryption-Free, Lossless Watermarking for Diffusion Models
Watermarks diffusion models losslessly via spherical mapping preserving Gaussian prior up to third-order moments.
Avg rating: 7.50 (6–8) · Xiaoxiao Hu et al.
Steering the Herd: A Framework for LLM-based Control of Social Learning
Framework studying strategic control of social learning by algorithmic information mediators with theoretical analysis and LLM-based simulations.
Avg rating: 6.50 (2–8) · Raghu Arghal et al.
Uncover Underlying Correspondence for Robust Multi-view Clustering
Proposes CorreGen, generative framework for multi-view clustering under noisy correspondence using EM algorithm.
Avg rating: 7.00 (6–8) · Haochen Zhou et al.
What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data
WIMHF uses sparse autoencoders to extract human-interpretable features from preference data, enabling better understanding and curation of human feedback.
Avg rating: 6.50 (4–8) · Rajiv Movva et al.