New bandit algorithms tackle fairness and continuous K-Max problems · 6 sources tracked
ByPulseAugur Editorial·[10 sources]·
Researchers have developed new algorithms for bandit problems, which are used in applications like recommendation systems. For continuous K-Max bandits, DCK-UCB achieves a sublinear regret bound of $\widetilde{O}(T^{3/4})$, with MLE-Exp offering near-optimal $\widetilde{O}(\sqrt{T})$ regret for exponential distributions. In a related area, UCB-HARE addresses the "price of fairness" in bandits, establishing a tight minimax lower bound of $\Omega(\sigma\sqrt{k^{\max(1,q)}/T})$ and an algorithm that matches it. Additionally, a novel algorithm for Bandit PCA achieves a minimax regret of $r\sqrt{dT}$ up to polylogarithmic factors, improving upon prior bounds and connecting to quantum tomography.
AI
IMPACT
Advances theoretical understanding and algorithmic solutions for complex bandit problems, potentially improving recommendation and decision-making systems.
RANK_REASON
Multiple arXiv papers detailing new algorithms and theoretical guarantees for bandit problems.
arXiv:2607.14706v1 Announce Type: new Abstract: We design and analyze \underline{M}echanism-\underline{E}nforced \underline{S}equential \underline{HA}lving (MESHA), an algorithm for Best Arm Identification (BAI) in strategic linear bandits. In this setting, each arm may strategic…
We design and analyze \underline{M}echanism-\underline{E}nforced \underline{S}equential \underline{HA}lving (MESHA), an algorithm for Best Arm Identification (BAI) in strategic linear bandits. In this setting, each arm may strategically misreport its feature vector to maximize th…
arXiv:2502.13467v2 Announce Type: replace Abstract: The $K$-Max combinatorial multi-armed bandit problem arises in applications such as recommendation and distributed decision making, where the reward is determined by the maximum outcome among $K$ selected arms. When outcomes are…
In bandit problems, standard regret-minimizing algorithms treat exploration as an amortized cost, which can expose early participants to unfair ex-ante losses in settings such as clinical trials. Recent work addresses this by evaluating the sequence of per-round expected rewards …
arXiv stat.ML
TIER_1English(EN)·Dhruv Sarkar, Soumyadeep Dutta, Sayak Ray Chowdhury·
arXiv:2607.13402v1 Announce Type: new Abstract: In bandit problems, standard regret-minimizing algorithms treat exploration as an amortized cost, which can expose early participants to unfair ex-ante losses in settings such as clinical trials. Recent work addresses this by evalua…
arXiv stat.ML
TIER_1English(EN)·Sayak Ray Chowdhury·
In bandit problems, standard regret-minimizing algorithms treat exploration as an amortized cost, which can expose early participants to unfair ex-ante losses in settings such as clinical trials. Recent work addresses this by evaluating the sequence of per-round expected rewards …
arXiv:2607.10936v1 Announce Type: cross Abstract: We study the bandit-feedback version of online principal component analysis (Bandit PCA): in each round $t = 1,\dots,T$, the adversary selects a $d \times d$ symmetric gain matrix $G_t$ with spectrum in $[0,1]$ and rank at most $r…
We study the bandit-feedback version of online principal component analysis (Bandit PCA): in each round $t = 1,\dots,T$, the adversary selects a $d \times d$ symmetric gain matrix $G_t$ with spectrum in $[0,1]$ and rank at most $r$; the learner simultaneously selects a unit vecto…
We study the bandit-feedback version of online principal component analysis (Bandit PCA): in each round $t = 1,\dots,T$, the adversary selects a $d \times d$ symmetric gain matrix $G_t$ with spectrum in $[0,1]$ and rank at most $r$; the learner simultaneously selects a unit vecto…
