Generalized Linear Bandits with Limited Adaptivity
Authors: Ayush Sawarni, Nirjhar Das, Siddharth Barman, Gaurav Sinha
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also perform experiments in Section 5 that validate its superiority both in terms of regret and computational efficiency in comparison to other baseline algorithms proposed in [14] and [6]. |
| Researcher Affiliation | Collaboration | Ayush Sawarni Stanford University ayushsaw@stanford.edu Nirjhar Das Indian Institute of Science Bangalore nirjhardas@iisc.ac.in Siddharth Barman Indian Institute of Science Bangalore barman@iisc.ac.in Gaurav Sinha Microsoft Research India gauravsinha@microsoft.com |
| Pseudocode | Yes | Algorithm 1 B-GLin CB: Batched Generalized Linear Bandits Algorithm |
| Open Source Code | Yes | The experiment code is available at https://github.com/nirjhar-das/GLBandit_Limited_Adaptivity. |
| Open Datasets | No | Arms were sampled uniformly from the d-dimensional unit ball. [...] Arm features were generated similarly as in the logistic bandit simulation. |
| Dataset Splits | No | The paper describes simulation parameters like total rounds T, dimension d, and number of arms K, but does not specify dataset splits (training, validation, test) for static datasets. Bandit problems evaluate cumulative regret over a sequence of rounds rather than using traditional dataset splits. |
| Hardware Specification | Yes | We ran all the experiments on an Azure Data Science VM equipped with AMD EPYC 7V13 64-Core Processor (clock speed of 2.45 GHz) and Linux Ubuntu 20.04 LTS operating system. |
| Software Dependencies | No | The paper states 'We implemented and tested our code in Python', but does not specify the version of Python or any other software libraries with version numbers. |
| Experiment Setup | Yes | The dimension was set to d = 5, number of arms per round to K = 20, and θ was sampled from a d-dimensional sphere of radius S = 5. Arms were sampled uniformly from the d-dimensional unit ball. We ran simulations for T = 20,000 rounds, repeating them 10 times. [...] We adjusted the Switching Criterion I threshold constant in RS-GLin CB to 0.01. |