Imagen de Portada

Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Bakman, Yavuz, Yaldiz, Duygu Nur, Avestimehr, Salman, Karimireddy, Sai Praneeth
Formato:	Preprint
Publicado:	2026
Materias:	Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Reconsidering LLM Uncertainty Estimation Methods in the Wild
por: Bakman, Yavuz, et al.
Publicado: (2025)

Reject Only Critical Tokens: Pivot-Aware Speculative Decoding
por: Ziashahabi, Amir, et al.
Publicado: (2025)

Conformal Prediction Adaptive to Unknown Subpopulation Shifts
por: Wang, Nien-Shao, et al.
Publicado: (2025)

MARS: Meaning-Aware Response Scoring for Uncertainty Estimation in Generative LLMs
por: Bakman, Yavuz Faruk, et al.
Publicado: (2024)

Uncertainty as Feature Gaps: Epistemic Uncertainty Quantification of LLMs in Contextual Question-Answering
por: Bakman, Yavuz, et al.
Publicado: (2025)

Uncertainty Quantification for Hallucination Detection in Large Language Models: Foundations, Methodology, and Future Directions
por: Kang, Sungmin, et al.
Publicado: (2025)

Un-considering Contextual Information: Assessing LLMs' Understanding of Indexical Elements
por: Oguz, Metehan, et al.
Publicado: (2025)

Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs
por: Yun, Vincent-Daniel, et al.
Publicado: (2026)

CroMo-Mixup: Augmenting Cross-Model Representations for Continual Self-Supervised Learning
por: Mushtaq, Erum, et al.
Publicado: (2024)

Optimization with Access to Auxiliary Information
por: Chayti, El Mahdi, et al.
Publicado: (2022)

TruthTorchLM: A Comprehensive Library for Predicting Truthfulness in LLM Outputs
por: Yaldiz, Duygu Nur, et al.
Publicado: (2025)

On the Limits of Momentum in Decentralized and Federated Optimization
por: Zaccone, Riccardo, et al.
Publicado: (2025)

Do Not Design, Learn: A Trainable Scoring Function for Uncertainty Estimation in Generative LLMs
por: Yaldiz, Duygu Nur, et al.
Publicado: (2024)

LIA: Privacy-Preserving Data Quality Evaluation in Federated Learning Using a Lazy Influence Approximation
por: Rokvic, Ljubomir, et al.
Publicado: (2022)

Collaborative Heterogeneous Causal Inference Beyond Meta-analysis
por: Guo, Tianyu, et al.
Publicado: (2024)

DAVED: Data Acquisition via Experimental Design for Data Markets
por: Lu, Charles, et al.
Publicado: (2024)

Defection-Free Collaboration between Competitors in a Learning System
por: Werner, Mariel, et al.
Publicado: (2024)

Do Data Valuations Make Good Data Prices?
por: Fan, Dongyang, et al.
Publicado: (2025)

VoxGuard: Evaluating User and Attribute Privacy in Speech via Membership Inference Attacks
por: Tsaprazlis, Efthymios, et al.
Publicado: (2025)

A Differentially Private Kaplan-Meier Estimator for Privacy-Preserving Survival Analysis
por: Veeraragavan, Narasimha Raghavan, et al.
Publicado: (2024)

Beyond URLs: Metadata Diversity and Position for Efficient LLM Pretraining
por: Fan, Dongyang, et al.
Publicado: (2025)

Entropy-driven Fair and Effective Federated Learning
por: Wang, Lin, et al.
Publicado: (2023)

Communication-Efficient Heterogeneous Federated Learning with Generalized Heavy-Ball Momentum
por: Zaccone, Riccardo, et al.
Publicado: (2023)

Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning
por: Yaldiz, Duygu Nur, et al.
Publicado: (2026)

f-INE: A Hypothesis Testing Framework for Estimating Influence under Training Randomness
por: Panda, Subhodip, et al.
Publicado: (2025)

Black-Box Behavioral Distillation Breaks Safety Alignment in Medical LLMs
por: Jahan, Sohely, et al.
Publicado: (2025)

Robust Multi-Agent LLMs under Byzantine Faults
por: Lee, Haejoon, et al.
Publicado: (2026)

Safety Game: Inference-Time Alignment of Black-Box LLMs via Constrained Optimization
por: Nguyen, Tuan, et al.
Publicado: (2025)

Deployment-Relevant Alignment Cannot Be Inferred from Model-Level Evaluation Alone
por: Vishwarupe, Varad, et al.
Publicado: (2026)

Differentially Private Federated Learning without Noise Addition: When is it Possible?
por: Zhang, Jiang, et al.
Publicado: (2024)

ATP: Enabling Fast LLM Serving via Attention on Top Principal Keys
por: Niu, Yue, et al.
Publicado: (2024)

A Closer Look at Personalized Fine-Tuning in Heterogeneous Federated Learning
por: Chen, Minghui, et al.
Publicado: (2025)

GEM: A Scale-Aware and Distribution-Sensitive Sparse Fine-Tuning Framework for Effective Downstream Adaptation
por: Kang, Sungmin, et al.
Publicado: (2025)

Towards Distillation Guarantees under Algorithmic Alignment for Combinatorial Optimization
por: Le, Thien, et al.
Publicado: (2026)

Trust-Region Noise Search for Black-Box Alignment of Diffusion and Flow Models
por: Schweiger, Niklas, et al.
Publicado: (2026)

Hawk: Accurate and Fast Privacy-Preserving Machine Learning Using Secure Lookup Table Computation
por: Saleem, Hamza, et al.
Publicado: (2024)

Conformal Alignment: Knowing When to Trust Foundation Models with Guarantees
por: Gui, Yu, et al.
Publicado: (2024)

Information Theoretic Guarantees For Policy Alignment In Large Language Models
por: Mroueh, Youssef
Publicado: (2024)

Asymptotics of Language Model Alignment
por: Yang, Joy Qiping, et al.
Publicado: (2024)

No Certificate for Alignment: Two Independent Impossibilities and the Pareto Frontier of Achievable Safety Guarantees
por: Agarwal, Ayushi
Publicado: (2026)