Multi-Fidelity Best-Arm Identification

Authors: Riccardo Poiani, Alberto Maria Metelli, Marcello Restelli

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we numerically validate IISE, showing the benefits of our method in simulated domains.
Researcher Affiliation Academia Riccardo Poiani DEIB, Politecnico di Milano riccardo.poiani@polimi.it Alberto Maria Metelli DEIB, Politecnico di Milano albertomaria.metelli@polimi.it Marcello Restelli DEIB, Politecnico di Milano marcello.restelli@polimi.it
Pseudocode Yes Our solution (pseudo-code in Algorithm 1 and visual representation in Figure 1) builds on the Successive Elimination algorithm [10].
Open Source Code Yes The code for generating the synthetic bandits and the Yahtzee game is released in the supplementary material.
Open Datasets Yes More specifically, we consider the Yahtzee game [3].
Dataset Splits No The paper describes experimental parameters but does not specify dataset splits (e.g., train/validation/test percentages or counts) in the main text. The problem context (Best-Arm Identification) does not typically involve static dataset splits for training or validation.
Hardware Specification Yes The experiments were run on a machine with an Intel Core i7-4770K CPU (3.50 GHz), 16GB of RAM, and a GeForce RTX 2080 Ti GPU (11GB).
Software Dependencies Yes The algorithms were implemented in Python 3.8 and ran on PyTorch 1.9.
Experiment Setup Yes Synthetic A setting parameters are K = 2000, M = 4, λ = [1, 10, 100, 1000], ξ = [1.15, 0.225, 0.015, 0], γ = [0.3, 0.05, 0.001, 0]; for Synthetic B, instead, K = 1000, M = 5, λ = [16, 64, 256, 1024, 4096], ξ = [1.15, 0.45, 0.105, 0.0105, 0], γ = [0.3, 0.1, 0.01, 0.001, 0]. For both the synthetic bandits and the Yahtzee game, we have chosen the parameters such that the conditions of Assumption 1 hold for our results, we have picked a small δ = 0.001 and σ2 = 1. The thresholds αm are set in two different ways.