Online Learning with Unknown Constraints

Authors: Karthik Sridharan, Seung Won Wilson Yoo

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical On the theoretical side, our algorithm s regret can be bounded by the regret of the online regression and online learning oracles, the eluder dimension of the model class containing the unknown safety constraint, and a novel complexity measure that characterizes the difficulty of safe learning. We complement our result with an asymptotic lower bound that shows that the aforementioned complexity measure is necessary.
Researcher Affiliation Academia Karthik Sridharan 1 Seung Won Wilson Yoo 1 1Department of Computer Science, Cornell University, Ithaca NY, United States. Correspondence to: Seung Won Wilson Yoo <EMAIL>.
Pseudocode Yes Algorithm 1 General Constrained Online Learning 1: Input: Oracle OL, Oracle OR, A0( ), δ (0, 1), κ, M 2: F0 = {f F : x X, a A0(x), f(a, x) 0} 3: for t = 1, . . . , T do 4: Receive context xt ... Algorithm 2 Online Learning with Long Term Constraints ... Algorithm 3 Oracle OL for Linear Losses ... Algorithm 4 General Online Learning with Vector-Valued Constraints
Open Source Code No The paper does not contain any explicit statements or links indicating the availability of source code for the described methodology.
Open Datasets No The paper is theoretical and does not describe experiments that use specific datasets. Therefore, no access information for publicly available datasets is provided.
Dataset Splits No The paper is theoretical and does not conduct experiments on specific datasets, thus there is no mention of dataset splits.
Hardware Specification No The paper focuses on theoretical contributions and does not report on experimental results requiring specific hardware specifications.
Software Dependencies No The paper focuses on theoretical algorithms and their bounds, and does not specify any software dependencies or version numbers for implementation.
Experiment Setup No The paper primarily presents theoretical results and algorithms, and does not include details on experimental setup such as hyperparameters or training configurations.