Bandits with Adversarial Scaling

Authors: Thodoris Lykouris, Vahab Mirrokni, Renato Paes Leme

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Figure 1 we compare those algorithms in a purely stochastic instance with large means. As usually noted in the literature, the performance of Thompson Sampling is vastly superior than all other algorithms. On this instance, UCB, Tsallis and BROAD have similar perfomance, EXP3++ is somewhat worse followed by AAE and AAEAS which are notably worse. This is expected as they are the least adaptive. It is good to keep those in mind as we compare their performance on certain adversarial scaling scenarios.
Researcher Affiliation Industry 1Microsoft Research, New York City, NY, USA 2Google Research, New York, NY, USA.
Pseudocode Yes Algorithm 1 Active Arm Elimination with Adversarial Scaling (AAEAS)
Open Source Code No The paper does not provide any explicit statement about making the source code for its methodology publicly available, nor does it include links to a code repository.
Open Datasets No The paper describes experiments run on simulated stochastic instances with defined parameters (e.g., 'two arms with means µ = [0.5, 0.8]'), rather than utilizing or providing access information for a publicly available or open dataset.
Dataset Splits No The paper describes simulated experiments and algorithmic performance over time, but it does not specify explicit training, validation, or test dataset splits in terms of percentages, sample counts, or predefined citations.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with versions) that would be needed to replicate the experiments.
Experiment Setup No The paper describes the characteristics of the simulated environments (e.g., arm means, cold start duration) and the general adaptive mechanisms of the algorithms (e.g., AAEAS uses δ=1/T), but it does not provide specific hyperparameter values or detailed system-level training settings for all comparative experimental runs beyond these general descriptions.