Feasible Arm Identification

Authors: Julian Katz-Samuels, Clay Scott

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we demonstrate the effectiveness of our algorithms on synthetic and real-world datasets.
Researcher Affiliation Academia 1 Department of Computer Science and Electrical Engineering, University of Michigan.
Pseudocode Yes Algorithm 1 MD-UCBE: Multi-dimensional Upper Confidence Bound Exploration algorithm; Algorithm 2 MD-SAR: Multi-dimensional Successive Accepts and Rejects algorithm; Algorithm 3 MD-APT: Multi-dimensional Anytime Parameter-Free Thresholding algorithm.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the methodology is openly available.
Open Datasets Yes We investigate this problem by considering the data in Genovese et al. (2013) (see ARCR20 in week 16 in Table 2 and Table 3). [...] We use a real-world dataset for the natural language processing task of affective text analysis (Snow et al., 2008).
Dataset Splits No The paper describes a multi-armed bandit problem with a fixed budget, not a typical supervised learning setup with explicit training, validation, and test splits for data.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No To calculate p pϵq i,t , we use the quadratic programming solver in the CVXOPT package for python. While the software name is mentioned, specific version numbers for CVXOPT or Python are not provided.
Experiment Setup Yes Each experiment has 20 5-dimensional arms and is run for 2000 time steps. We use Gaussian distributions with variance 1/4. For experiments 1, 2, and 3 we use a cube P tx P R5 : 0 ď xi ď 1u. [...] We run the experiment for 1000 time steps (Dose-Finding). [...] We run each algorithm for 4000 time steps (Crowdsourcing).