Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Indexed Minimum Empirical Divergence-Based Algorithms for Linear Bandits
Authors: Jie Bian, Vincent Y. F. Tan
TMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Furthermore, extensive empirical studies reveal that Lin IMED and its variants outperform widely-used linear bandit algorithms such as Lin UCB and Linear Thompson Sampling in some regimes. 6 Empirical Studies This section aims to justify the utility of the family of Lin IMED algorithms we developed and to demonstrate their effectiveness through quantitative evaluations in simulated environments and real-world datasets such as the Movie Lens dataset. |
| Researcher Affiliation | Academia | Jie Bian EMAIL Department of Electrical and Computer Engineering National University of Singapore Vincent Y. F. Tan EMAIL Department of Mathematics Department of Electrical and Computer Engineering National University of Singapore |
| Pseudocode | Yes | Algorithm 1 Lin IMED-x for x {1, 2, 3} Algorithm 2 Sup Lin IMED Algorithm 3 Base Lin UCB |
| Open Source Code | No | The paper does not contain an explicit statement about the release of its own source code, nor does it provide a link to a code repository. It mentions other algorithms but not its own implementation. |
| Open Datasets | Yes | The Movie Lens dataset (Cantador et al. (2011)) is a widely-used benchmark dataset for research in recommendation systems. We specifically choose to use the Movie Lens 10M dataset |
| Dataset Splits | No | The paper mentions using a 'synthetic dataset' and the 'Movie Lens 10M dataset'. For Movie Lens, it describes selecting 'the best K {20, 50, 100} movies' and an assumption about user clicks, but it does not specify any training, validation, or test splits. For the synthetic dataset, it describes parameters like dimensions (d) and number of arms (K) but no splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as CPU models, GPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions algorithms like Lin UCB, Lin TS, and IDS as benchmarks, but it does not specify any software dependencies or library versions used for implementation (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | We set βt(γ) = (R q 3d log(1 + t) + 2)2 (here γ = 1 (1+t)2 and L = 2) for the synthetic dataset with varying and finite arm set and βt(γ) = (R q d log((1 + t)t2) + 20)2 (here γ = 1 t2 and L = 20) for the Movie Lens dataset respectively. The confidence widths p βt(γ) for each algorithm are multiplied by a factor α and we tune α by searching over the grid {0.05, 0.1, 0.15, 0.2, . . . , 0.95, 1.0} and report the best performance for each algorithm; see Appendix G. Both γ s are of order O( 1 t2 ) as suggested by our proof sketch in Eqn. (5). We set C = 30 in Lin IMED-3 throughout. The sub-Gaussian noise level is R = 0.1. |