[PDF] A Gang of Adversarial Bandits | Semantic Scholar (2024)

The CBA algorithm is proposed, which exploits the assumption that one action corresponding to the learner's abstention from play, has no reward or loss on every trial, and is the first to achieve bounds on the expected cumulative reward for general confidence-rated predictors.

This work provides a rigorous regret analysis for the standard flooding protocol combined with the UCB policy, and proposes a new protocol called Flooding with Absorption (FWA), which is verified empirically that using FWA leads to significantly lower communication costs despite minimal regret performance loss compared to flooding.

The nearest neighbour rule is adapted to the contextual bandit problem and the algorithm is extremely efficient - having a per trial running time polylogarithmic in both the number of trials and actions, and taking only quasi-linear space.

The analysis shows that the regret of $\texttt{MT-CO}_2\texttt{OL}$ is never worse than the bound obtained when agents do not share information, and it is proved that the algorithm can be made differentially private with a negligible impact on the regret.

The adversarial contextual bandit problem in metric spaces is considered, designing an algorithm in which it can hold out any set of contexts when computing the authors' regret term and hence inherits its extreme computational efficiency.

This work characterize regret in terms of the independence number of the strong product between the feedback graph and the communication network, which recovers as special cases many previously known bounds for distributed online learning with either expert or bandit feedback.

AdaTask can be seen as a comparator-adaptive version of Follow-the-Regularized-Leader with a Mahalanobis norm potential, and a variational formulation of this potential reveals how AdaTask jointly learns the tasks and their structure.

This work proves an effective regret of $\mathcal{O}(\sqrt{n^{1+\gamma}})$ when suitable parameterized graph kernels are chosen, and proposes an approximate algorithm FastONL enjoying regret based on this relaxation.

This paper heuristically derive a limiting PDE on Wasserstein space which characterizes the asymptotic behavior of the regret of the forecaster and shows that the problem of obtaining regret bounds and efficient algorithms can be tackled by finding appropriate smooth sub/supersolutions of this parabolic PDE.

This paper provides algorithms for this setting, both for stochastic and adversarial bandits, and shows that their regret smoothly interpolates between the regret in the classical bandit setting and that of the full-information setting as a function of the neighbors' exploration.

A global recommendation strategy which allocates a bandit algorithm to each network node (user) and allows it to "share" signals (contexts and payoffs) with the neghboring nodes, and derives two more scalable variants of this strategy based on different ways of clustering the graph nodes.

The investigations in this work reveal the significant gains that can be obtained even through static network-aware policies, and proposes a randomized policy that explores actions for each user at a rate that is a function of her network position.

An upper confidence bound-based algorithm is developed, RobustAgg ($epsilon), that adaptively aggregates rewards collected by different players and achieves instance-dependent regret guarantees that depend on the amenability of information sharing across players.

This paper formalizes the networked bandit problem and proposes an algorithm that considers not only the selected arm, but also the relationships between arms, in that it decides an arm depending on integrated confidence sets constructed from historical data.

A contextual bandit problem is studied in a highly non-stationary environment and an efficient learning algorithm that is adaptive to abrupt reward changes is proposed and theoretical regret analysis is provided to show that a sublinear scaling of regret in the time length T is achieved.

This work considers similarity information in the setting of contextual bandits, a natural extension of the basic MAB problem, and presents algorithms that are based on adaptive partitions, and take advantage of "benign" payoffs and context arrivals without sacrificing the worst-case performance.

This paper forms the -multi-player multi-armed bandit problem, and develops an upper confidence bound-based algorithm that adaptively aggregates rewards collected by different players, to be the first to develop such a scheme in a multi-player bandit learning setting.

This paper develops a collaborative contextual bandit algorithm, in which the adjacency graph among users is leveraged to share context and payoffs among neighboring users while online updating, and rigorously proves an improved upper regret bound.

A novel algorithm in which agents, whenever they choose, communicate only arm-ids and not samples, with another agent chosen uniformly and independently at random is developed, demonstrating that even a minimal level of collaboration among the different agents enables a significant reduction in per-agent regret.

...

...