Disposable Linear Bandits for Online Recommendations

Authors: Melda Korkut, Andrew Li4172-4180

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm s performance on a recommendation task based on synthetically generated data. Compared to a number of benchmarks, including Lin UCB and a natural modification of Thompson sampling, our algorithm (solved via the preceding heuristic) achieves as much as 10% lower regret against all competing algorithms.
Researcher Affiliation Academia Melda Korkut, Andrew Li Tepper School of Business Carnegie Mellon University {melda,aali1}@cmu.edu
Pseudocode Yes Algorithm 1: Generalized LINUCB (UCBG) and Algorithm 2: Alternating Heuristic
Open Source Code No The corresponding data can be found in https://github.com/Melda Kor/Disposable Linear Bandits.
Open Datasets Yes We generated 퐾arms, 푖= 1, . . . , 퐾in 푑 dimensional space where 퐾= 5000 and 푑= 15. Similarly, we generated a set of 휃s that lie in the same space, where the total number of 휃s is 5000. 3The corresponding data can be found in https://github.com/Melda Kor/Disposable Linear Bandits.
Dataset Splits No The paper mentions synthetically generated data and number of instances but does not specify explicit training, validation, or test dataset splits or cross-validation details.
Hardware Specification No No specific hardware details (like GPU/CPU models, memory, or cloud instance types) are provided for running the experiments.
Software Dependencies No No specific software dependencies with version numbers are listed in the paper.
Experiment Setup Yes For all experiments, we set the tuning parameter 훼for the heuristic the same as Lin UCB s 푐. In experiments, 훼, 푐= (1/2)푖where 푖= 3, 4, 5.