On Conditional Versus Marginal Bias in Multi-Armed Bandits
Authors: Jaehyeok Shin, Aaditya Ramdas, Alessandro Rinaldo
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Section 4, we demonstrate with several examples in best arm identification and sequential testing problems how the marginal and conditional bias of the sample means of the arms can have opposite signs. These are, we believe, instances of a general, important phenomenon of theoretical and practical relevance. Experiment: We verify these facts with simulations, where we repeat a stopped sequential test 10^5 times. |
| Researcher Affiliation | Academia | 1Department of Statistics and Data Science, Carnegie Mellon University 2Machine Learning Department, Carnegie Mellon University. Correspondence to: Jaehyeok Shin <shinjaehyeok@cmu.edu>. |
| Pseudocode | No | The paper describes algorithms (e.g., lil UCB, Sequential Halving) in prose, but it does not include any explicitly labeled pseudocode blocks or algorithm figures. |
| Open Source Code | No | The paper does not provide any statements about making its source code available, nor does it include links to a code repository. |
| Open Datasets | No | The paper uses simulated data generated from "standard normal distributions" for its experiments. It does not utilize or provide access information for any pre-existing publicly available datasets. |
| Dataset Splits | No | Since the paper relies on simulations rather than pre-existing datasets, it does not describe training, validation, or test data splits in the conventional sense. The experiments involve running simulations a specified number of times. |
| Hardware Specification | No | The paper describes simulations for its experiments but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run these simulations. |
| Software Dependencies | No | The paper discusses theoretical models and simulation results, but it does not specify any software names or version numbers (e.g., Python, PyTorch, R, MATLAB) used for conducting the experiments. |
| Experiment Setup | Yes | Experiment: We verify these facts with simulations, where we repeat a stopped sequential test 10^5 times. In each test, the arm corresponds to a standard normal distribution... we choose this boundary with unusual parameters = 0.2 and M = 10 to manifest the difference between marginal and conditional biases. To best illustrates the bias results, we use an unusual set of algorithm parameters as δ = 0.2, = 0.1, β = 0.5 and λ = 1 in this experiment. |