![[2602.17103] Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17103] Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners
Summary
This paper explores online learning models using improving agents, focusing on multiclass setups, budgeted agents, and bandit learners, extending previous research in machine learning.
Why It Matters
Understanding the dynamics of online learning with improving agents is crucial for advancing machine learning applications. This research provides insights into how agents can adapt their features for better outcomes, which is relevant for developing more efficient algorithms in various domains.
Key Takeaways
- Introduces a model where agents can modify features for improved labels.
- Extends previous findings by analyzing multiclass learning and bandit feedback.
- Discusses the cost implications for agents making improvements.
Computer Science > Machine Learning arXiv:2602.17103 (cs) [Submitted on 19 Feb 2026] Title:Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners Authors:Sajad Ashkezari, Shai Ben-David View a PDF of the paper titled Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners, by Sajad Ashkezari and 1 other authors View PDF HTML (experimental) Abstract:We investigate the recently introduced model of learning with improvements, where agents are allowed to make small changes to their feature values to be warranted a more desirable label. We extensively extend previously published results by providing combinatorial dimensions that characterize online learnability in this model, by analyzing the multiclass setup, learnability in a bandit feedback setup, modeling agents' cost for making improvements and more. Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML) Cite as: arXiv:2602.17103 [cs.LG] (or arXiv:2602.17103v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2602.17103 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Sajad Ashkezari [view email] [v1] Thu, 19 Feb 2026 06:01:30 UTC (17 KB) Full-text links: Access Paper: View a PDF of the paper titled Online Learning with Improving Agents: Multiclass, Budgeted Agents and Bandit Learners, by Sajad Ashkezari and 1 other authorsView PDFHTML (experimental)TeX Source view license Current browse contex...
