Research map / Bandits and online learning / Online convex optimization and games
Agnostic Online Learning with Reliable Abstention
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies agnostic online learning in the adversarial injection model with reliable abstention, where the learner may abstain without penalty on adversarial rounds while minimizing mistakes and abstentions on i.i.d. rounds.
Problem
Online learning when each round is either i.i.d. from an unknown distribution or an arbitrary adversarial injection that the learner cannot identify; the learner must limit misclassifications and abstentions on i.i.d. instances.
Approach
Works in the adversarial injection model of GHMS23 with an agnostic setting and free abstention on adversarial injections, sitting between fully stochastic and fully adversarial online learning. Full text unavailable, so the specific algorithmic contribution is not stated.
Experimental setup
Metrics: total misclassifications on i.i.d. instances abstentions on i.i.d. instances
Builds on: adversarial injection model (GHMS23)
Closest papers by meaning
- Reliable Abstention under Adversarial Injections: Lower Bounds and New Upper Bounds (NeurIPS 2026 · similarity 0.93)
- Online Learning with Abstention (ICML 2018 · similarity 0.89)
- Online Learning in the Random-Order Model (ICML 2025 · similarity 0.86)
- Bandit-Feedback Online Multiclass Classification: Variants and Tradeoffs (NeurIPS 2024 · similarity 0.86)
- Discounted Adaptive Online Learning: Towards Better Regularization (ICML 2024 · similarity 0.85)
- Online Linear Classification with Massart Noise (ICML 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.