Best Arm Identification in Multi-Armed Bandits

By Jean-Yves Audibert, Sébastien Bubeck, R´emi Munos. 2010

1,677 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Taking the Human Out of the Loop: A Review of Bayesian Optimization (2016, methodology): “arm identification problem, which is more suitable for the model selection task [7], [30], [50], [51], [72], [104].”
Non-stochastic Best Arm Identification and Hyperparameter Optimization (2015, neutral): “While clearly not a proof, this argument along with known lower bounds for the stochastic setting (Audibert & Bubeck, 2010; Kaufmann et al., 2014), a subset of the non-stochastic setting, suggest that the above result may be nearly tight in a minimax sense up to log factors.”
Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems (2012, neutral): “Owing to this probabilistic construction, the lower bound proof is based on the same KullbackLeibler divergence as the one used in the proof of the lower bound for stochastic bandits —see Subsection 2.3.”
Learning to Perform Physics Experiments via Deep Reinforcement Learning (2016, positive): “We assign masses to objects in a way that is uncorrelated with their appearance in order to ensure that the task is not solvable without interaction.”
On the Complexity of Best-Arm Identification in Multi-Armed Bandit Models (2014, methodology): “This result is to be compared to the lower bound of Audibert et al. (2010). While Theorem 16 does not really provide a lower bound on κB(ν), the complexity term H(ν) is close to the quantity that appears in Theorem 4 for the fixed-confidence setting (in the Gaussian case), which improves over the te”
From Bandits to Monte-Carlo Tree Search: The Optimistic Principle Applied to Optimization and Planning (2014, background): “This loss bound is not optimal for finitely many armed bandits (since there exists strategies that achieve exponential loss bounds as discussed in [27, 9]), but in the case of X -armed bandits (where the set of arms is larger than the number of rounds n), this may be unimprovable.”
PAC Subset Selection in Stochastic Multi-armed Bandits (2012, background): “On a closely-related exploration problem, Audibert et al. (2010) also observe that a tighter upper bound can be achieved if the problem complexity (a quantity similar to H) is known beforehand.”
Causal Bandits: Learning Good Interventions via Causal Inference (2016, methodology): “We compare the performance of the Algorithm 1, which is speciﬁc to the parallel problem, but does not require knowledge of q , with that of Algorithm 2 and the Successive Reject algorithm of Audibert and Bubeck (2010).”
Estimating the Maximum Expected Value through Upper Confidence Bound of Likelihood (2017, background): “much as possible [6], and to develop a strategy identifying the optimal arm [9].”
Best Arm Identification: A Unified Approach to Fixed Budget and Fixed Confidence (2012, methodology): “, [3, 1]), the number of rounds of the exploration phase is fixed and is known by the forecaster, and the objective is to maximize the probability of returning the best arm(s).”
Automatic Prompt Optimization with "Gradient Descent" and Beam Search (2023, background): “Note that this almost exactly corresponds to the well-studied problem of best arm identification in bandit optimization (Audibert et al., 2010).”
Spectral–Spatial Hyperspectral Image Classification With Edge-Preserving Filtering (2014, methodology): “Most of these EPFs have a similar property, i.e., they can be used for joint filtering, where the content of one image is smoothed based on the edge information from a guidance image.”