New findings on universal learning in contextual bandits with adversarial rewards.
problem Learning in contextual bandits with time-varying, adversarial rewards.
method Characterization of learnable processes and necessary/sufficient conditions for universal learning.
result Optimistic universal learning for contextual bandits with adversarial rewards is impossible in general.
First robust bandit algorithm for contextual bandits with sub-linear regret.
problem Vulnerability of linear contextual bandit algorithms to adversarial attacks.
method Proposes a robust bandit algorithm for stochastic linear contextual bandits under fully adaptive and omniscient attacks.
result Sub-linear regret under various attacks without requiring attack information.
New action poisoning attacks improve LinUCB's performance by changing action signals.
problem Improving understanding of adversarial attacks on contextual bandit algorithms.
method Proposed action poisoning attacks in white-box and black-box settings.
result Action poisoning attacks can force LinUCB to pull a target arm frequently with low cost.
Improved algorithm for contextual bandits with reduced regret.
problem Adversarial contextual bandits with i.i.d. contexts.
method Oracle-efficient relaxation with O ( T 2 3 ( K log ( ∣ Π ∣ ) ) 1 3 ) O(T^{\frac{2}{3}}(K\log(|Π|))^{\frac{1}{3}}) O ( T 3 2 ( K log ( ∣Π∣ ) ) 3 1 ) regret bound. result First to improve regret bound and match original bound for stochastic case.
Improved ε ε ε -greedy handles strategic bidding in PPC auctions.
problem Strategic bidding in PPC auctions with personalization and corruptions.
method Extended ε ε ε -greedy to handle strategic arms in contextual multi-arm bandit. result ε ε ε -greedy is robust to adversarial corruptions and degrades linearly with corruption. A new adversarial attack method using structured search and contextual bandits.
problem Black-box adversarial attacks on deep learning models.
method Structured search space and Bayesian optimization for contextual bandits.
result Achieves state-of-the-art success rates and query efficiencies.
The paper tackles robust policy learning in multitask contextual bandits with adversarial users.
problem Learning optimal policies in multitask contextual bandits with a small fraction of adversarial users.
method Developed efficient robust mean estimators for both uni-variate and high-dimensional random variables.
result Lower bound of i l d e Ω ( min ( S , A ) ⋅ α 2 / ε 2 ) ildeΩ(\min(S,A) \cdot α^2 / ε^2) i l d e Ω ( min ( S , A ) ⋅ α 2 / ε 2 ) per-user interactions to learn an ε ε ε -optimal policy for good users. Develops a strategy to minimize loss in both stochastic and adversarial environments for linear contextual bandits.
problem Linear contextual bandits with adversarial corruption.
method Proposes a novel strategy called Best-of-Both-Worlds (BoBW) RealFTRL, extending RealLinExp3 and FTRL.
result Regret upper bound of $O\left(\min\left\{\frac{(\log(T))^3}{Δ_{*}} + \sqrt{\frac{C(\log(T))^3}{Δ_{*}}},\ \ \sqrt{T}(\log(T))^2
ight\}
ight)$ , showing effectiveness in both stochastic and adversarial environments.
New algorithm reduces regret in contextual bandits with many near-boundary contexts.
problem High regret in contextual bandits with many near-boundary contexts.
method Hierarchical nearest neighbour approach, holding out contexts for computation.
result Eradicates high regret in adversarial contextual bandits.
Improved regret bounds for adversarial linear contextual bandits.
problem Adversarial linear contextual bandits with changing loss functions.
method Truncated continuous exponential weights algorithm over the probability simplex, analyzing with linear bandit setting without contexts.
result Second-order bound of i l d e O ( K d V T ) ilde O(K\sqrt{d V_T}) i l d e O ( K d V T ) and first-order bound of i l d e O ( K d L T ∗ ) ilde O(K\sqrt{d L_T^*}) i l d e O ( K d L T ∗ ) . Improved online Lasso reduces regret in sparse linear contextual bandits.
problem Sparse linear contextual bandit problem with inefficient sampling.
method Perturbed adversary approach to alleviate sampling inefficiency.
result Online Lasso achieves O ( k T log d ) \mathcal{O}(\sqrt{kT\log d}) O ( k T log d ) regret bound. New algorithm resists corruption in linear contextual bandits.
problem Adversarial corruption in linear contextual bandits.
method Variance-aware algorithm with multi-level partition and adaptive confidence sets.
result Regret bound of i l d e O ( C 2 d ∑ t = 1 T σ t 2 + C 2 R d T ) ilde{O}(C^2d\sqrt{\sum_{t = 1}^T σ_t^2} + C^2R\sqrt{dT}) i l d e O ( C 2 d ∑ t = 1 T σ t 2 + C 2 R d T ) . A new algorithm tackles adversarial linear contextual bandits using kernelized loss functions.
problem Online learning in adversarial linear contextual bandits with flexible loss functions.
method Proposes a computationally efficient algorithm using an optimistically biased estimator for reproducing kernel Hilbert space loss functions.
result Achieves near-optimal regret guarantees under polynomial and exponential eigendecay assumptions.
New algorithm reduces linear contextual bandit regret with adversarial corruption.
problem Linear contextual bandit with adversarial reward corruption.
method Optimism in the face of uncertainty principle, weighted ridge regression.
result Achieves nearly optimal regret for both corrupted and uncorrupted cases.
Paper solves stochastic contextual linear bandits using linear bandit algorithms.
problem Stochastic contextual linear bandits with unknown context distribution.
method Establishes a reduction framework to convert to linear bandit problems.
result Achieves nearly optimal regret bound of O ( d T log T ) O(d\sqrt{T\log T}) O ( d T log T ) . New method achieves near-optimal regret without simulator.
problem Adversarial linear contextual bandits with unknown loss vectors.
method Near-optimally reduces regret to sqrt(T) without simulator.
result Achieves regret of sqrt(T) without simulator, improving existing methods.
A fundamental challenge in contextual bandits is to develop flexible, general-purpose algorithms with computational requirements no worse than classical supervised learning tasks such as classification and regression. Algorithms based on regression have shown promising empirical success, but theoretical guarantees have…
New algorithms minimize regret in both adversarial and stochastic contexts.
problem Minimizing regret in linear contextual bandits.
method Best-of-both-worlds algorithms using FTRL with Shannon entropy regularizer.
result Achieves near-optimal regret bounds in both adversarial and stochastic regimes.
New algorithms robust to adversarial data achieve optimal performance.
problem Adversarial robustness in high-dimensional online learning problems.
method Alternating minimization scheme combining least-squares and convex reweighting.
result Achieves optimal robustness guarantees without distributional assumptions.
Develops a Best-of-Both-Worlds algorithm for linear contextual bandits with Tsallis entropy.
problem Linear contextual bandits with i.i.d. contexts.
method Follow-The-Regularized-Leader (FTRL) with Tsallis entropy.
result Achieves $O\left(\log(T)^{\frac{1+β}{2+β}}T^{\frac{1}{2+β}}
ight)$ regret under margin condition.
The paper tackles minimax optimality in continuum contextual bandits with Hölder continuity.
problem Minimizing regret in a continuum of contexts with Hölder continuity.
method Proves a static-to-contextual regret conversion theorem and analyzes various dependency cases.
result Achieves minimax optimal contextual regret for convex and strongly convex bandits.
In this work we consider adversarial contextual bandits with risk constraints. At each round, nature prepares a context, a cost for each arm, and additionally a risk for each arm. The learner leverages the context to pull an arm and then receives the corresponding cost and risk associated with the pulled arm. In additi…
New algorithm reduces regret in RL with adversarial corruption.
problem Adversarial corruption in reinforcement learning.
method Uncertainty-weighted least-squares regression and weighted uncertainty estimator.
result Achieves regret of i l d e O ( T + ζ ) ilde{O}(\sqrt{T} + ζ) i l d e O ( T + ζ ) for contextual bandits. Study on adaptivity constraints in linear contextual bandits with optimal design.
problem Impact of adaptivity constraints on linear contextual bandits.
method Two models of limited adaptivity: batch learning and rare policy switches. Proposed distributional optimal design.
result Achieves minimax-optimal regret with optimal number of policy switches and batches.
Contextual bandit algorithms are applied in a wide range of domains, from advertising to recommender systems, from clinical trials to education. In many of these domains, malicious agents may have incentives to attack the bandit algorithm to induce it to perform a desired behavior. For instance, an unscrupulous ad publ…
New algorithm tackles dynamic query routing to multiple embedding models.
problem Dynamic query routing to multiple embedding models under adversarial conditions.
method Formalized as adversarial contextual linear bandit with low-rank experts, proposed HPG algorithm.
result HPG algorithm achieves linearized policy regret of i l d e O ( s M T ) ilde{\mathcal O}(s\sqrt{M T}) i l d e O ( s M T ) . A federated learning algorithm tackles linear bandits with adversarial actions, achieving optimal regret bounds.
problem Federated linear bandits with finite adversarial action sets.
method FedSupLinUCB algorithm, extending SupLinUCB and OFUL principles.
result Achieves a total regret of i l d e O ( d T ) ilde{O}(\sqrt{d T}) i l d e O ( d T ) , matching minimax lower bound and being order-optimal. Adaptive algorithms minimize regret in matching markets with contextual arm preferences.
problem Minimizing regret in matching markets with context-dependent player utilities.
method Developed adaptive algorithms for stochastic and adversarial contexts, providing upper and lower bounds.
result Achieved sublinear regret bounds for both stochastic and adversarial contexts.
Study on selecting between base algorithms in stochastic bandit problems.
problem Model selection in stochastic environments with contextual information.
method Developed a meta-algorithm-base algorithm abstraction with a smoothing transformation for optimal O ( T ) O(\sqrt{T}) O ( T ) guarantees. result Optimal O ( T ) O(\sqrt{T}) O ( T ) model selection guarantees for stochastic contextual bandit problems. We present efficient algorithms for the problem of contextual bandits with i.i.d. covariates, an arbitrary sequence of rewards, and an arbitrary class of policies. Our algorithm BISTRO requires d calls to the empirical risk minimization (ERM) oracle per round, where d is the number of actions. The method uses unlabeled…
We propose a contextual bandit based model to capture the learning and social welfare goals of a web platform in the presence of myopic users. By using payments to incentivize these agents to explore different items/recommendations, we show how the platform can learn the inherent attributes of items and achieve a subli…
Study online learning with off-policy feedback in adversarial bandit problems.
problem Learning with limited direct feedback in sequential decision making.
method Proposed algorithms that adapt pessimistic reward estimators to handle unknown behavior policy.
result Guaranteed regret bounds scaling with policy mismatch, improving performance against well-covered comparators.
The paper provides robustness guarantees for mode estimation in bandits.
problem Understanding robustness in mode estimation under adversarial data contamination.
method Simple randomization and theoretical analysis of multi-armed bandits.
result Regret guarantees for various modal bandit problems.
Contextual bandits study how reward variance affects regret bounds.
problem Investigating how small reward variance impacts regret bounds in contextual bandits.
method Analyzing two types of adversaries and function approximation complexities.
result Regret bounds are influenced by the eluder dimension and reward variance.
New algorithms for private generalized linear contextual bandits.
problem Private estimation and optimization for generalized linear models under differential privacy.
method Developed algorithms for stochastic and adversarial contexts under shuffle and joint differential privacy.
result Achieved private regret bounds for generalized linear models, differing from non-private rates by factors of d / ε \sqrt{d/\varepsilon} d / ε and d / ε \sqrt{d/\varepsilon} d / ε respectively. Thompson Sampling is one of the oldest heuristics for multi-armed bandit problems. It is a randomized algorithm based on Bayesian ideas, and has recently generated significant interest after several studies demonstrated it to have better empirical performance compared to the state-of-the-art methods. However, many ques…
New algorithm reduces regret in both adversarial and stochastic contexts.
problem Contextual combinatorial semi-bandits with adversarial and corrupted stochastic regimes.
method Follow-the-Regularized-Leader (FTRL) framework with Shannon entropy regularizer, accelerated by Karush-Kuhn-Tucker conditions.
result Achieves O ~ ( T ) \widetilde{\mathcal{O}}(\sqrt{T}) O ( T ) regret in adversarial and O ~ ( ln T ) \widetilde{\mathcal{O}}(\ln T) O ( ln T ) regret in corrupted stochastic regimes. New algorithm achieves best-of-both-worlds performance in various online learning settings.
problem Achieving optimal performance in both adversarial and stochastic online learning settings.
method General reduction from best-of-both worlds to FTRL and OMD algorithms.
result Transformed existing algorithms into new ones with best-of-both-worlds guarantees.
Optimal algorithm for contextual bandits with unknown context distributions.
problem Designing efficient algorithms for contextual bandits with unknown context distributions.
method Cross-learning setting, novel technique for coordinating multiple epochs.
result Nearly tight regret bound of O ~ ( T K ) \widetilde{O}(\sqrt{TK}) O ( T K ) for learning to bid in first-price auctions and sleeping bandits. Two algorithms address limited adaptivity in generalized linear contextual bandits.
problem Limited adaptivity in generalized linear contextual bandits.
method Two algorithms, B-GLinCB and RS-GLinCB, designed for two settings of limited adaptivity.
result Achieved i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) regret in both settings. Improved regret bounds for contextual bandits considering variance sequences.
problem Establishing lower bounds for contextual bandits with variance-dependent regret.
method Developed variance-dependent lower bounds for linear contextual bandits under two settings: fixed and adaptive variance sequences.
result Lower bounds match upper bounds of SAVE algorithm up to logarithmic factors.
Paper tackles domain adaptation for contextual bandits with sub-linear regret.
problem Adapting contextual bandit algorithms across domains with distribution shift.
method Learn a bandit model for the target domain using feedback from the source domain.
result Sub-linear regret bound maintained across domains.
New algorithm learns optimal exploration parameters for contextual bandits.
problem Learning optimal exploration in contextual bandits.
method Proposes two algorithms that learn optimal exploration parameters online based on context and reward.
result Demonstrates improved performance in learning optimal exploration compared to traditional methods.
This paper tackles fair online decision-making in contextual bandits, achieving optimal performance and fairness.
problem Fairness in online decision-making systems under strategic manipulation.
method Develops algorithms for linear and smooth reward functions, maintaining fairness and optimal regret.
result Achieves nearly minimax-optimal regret with strong fairness guarantees, even in the presence of attacks.
A framework for auto-tuning hyper-parameters in contextual bandit algorithms.
problem Auto-tuning hyper-parameters in real-time for contextual bandit algorithms.
method Proposes a Syndicated Bandits framework to learn multiple hyper-parameters dynamically.
result Achieves optimal regret bounds under certain scenarios and handles multiple contextual bandit algorithms.
New algorithms balance collaboration and adversarial behavior in linear bandits.
problem Minimizing regret in a collaborative linear bandit problem with adversarial agents.
method Robust collaborative phased elimination algorithm with tight analyses.
result Achieves near-optimal regret bounds of $O\left(α+ 1/\sqrt{M}
ight) \sqrt{dT}$ for good agents.
Most contextual bandit algorithms minimize regret against the best fixed policy, a questionable benchmark for non-stationary environments that are ubiquitous in applications. In this work, we develop several efficient contextual bandit algorithms for non-stationary environments by equipping existing methods for i.i.d. …
New algorithms combat adversarial attacks in stochastic linear bandits.
problem Adversarial attacks on stochastic linear bandit rewards.
method Two variants of Robust Phased Elimination algorithms, one knowing C C C and one not. result Near-optimal regret in non-corrupted case and additive terms dependent on C C C .