Three new research papers explore advancements in online conformal prediction and selection, addressing limitations in existing methods. The first paper introduces a unified framework for online conformal prediction that simultaneously controls coverage violation and prediction-set efficiency, even in adversarial settings. The second paper focuses on efficient online conformal selection with limited feedback, demonstrating adversarial validity and stochastic efficiency under bandit-like feedback. The third paper proposes Robust Conformalized Selection (RCS), a method designed to maintain valid false discovery rate control even when calibration data is contaminated with noisy responses, and shows its effectiveness in various applications including large language models. AI
IMPACT These papers advance the theoretical underpinnings of uncertainty quantification in machine learning, potentially improving reliability in applications like LLM alignment and decision-making under uncertainty.
RANK_REASON The cluster consists of three academic papers published on arXiv, detailing theoretical advancements in machine learning algorithms.
- arXiv
- large-language models
- Robust Conformalized Selection
- Adaptive Conformal Inference
- alphaXiv
- arXivLabs
- Candès
- CatalyzeX Code Finder for Papers
- CORE Recommender
- DagsHub
- Gibbs
- Gotit.pub
- Hugging Face
- IArxiv Recommender
- Influence Flower
- Kamesh Munagala
- Online Conformal Prediction
- ScienceCast
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
