Prompt-based offline methods are commonly used to optimize large language model (LLM) responses, but evaluating these responses is computationally intensive and often fails to accommodate diverse response styles. This study introduces a novel online evaluation framework that employs a multi-agent conversational bandit model to select optimal responses while aligning with user preferences dynamically. To tackle challenges such as high-dimensional features, large response sets, adaptive conversational needs, and multi-device access, we propose MACO, Multi-Agent Conversational Online Learning, which comprises two key components: (1) MACO-A: Executed by local agents, it employs an online elimination mechanism to filter out low-quality responses. (2) MACO-S: Executed by the cloud server, it adaptively adjusts selection strategies based on aggregated preference data. An adaptive preference mechanism triggers asynchronous conversations to enhance alignment efficiency. Theoretical analysis demonstrates that MACO achieves near-optimal regret bounds, matching state-of-the-art performance in various degenerate cases. Extensive experiments utilizing Google and OpenAI text embedding models on the real-world datasets with different response styles, combined with Llama and GPT-4o, show that MACO consistently outperforms baseline methods by at least 8.29% across varying response set sizes and numbers of agents.

Large language models (LLMs) exhibit diverse response behaviors, costs, and strengths, making it challenging to select the most suitable LLM for a given user query. We study the problem of adaptive multi-LLM selection in an online setting, where the learner interacts with users through multi-step query refinement and must choose LLMs sequentially without access to offline datasets or model internals. A key challenge arises from unstructured context evolution: the prompt dynamically changes in response to previous model outputs via a black-box process, which cannot be simulated, modeled, or learned. To address this, we propose the first contextual bandit framework for sequential LLM selection under unstructured prompt dynamics. We formalize a notion of myopic regret and develop a LinUCB-based algorithm that provably achieves sublinear regret without relying on future context prediction. We further introduce budget-aware and positionally-aware (favoring early-stage satisfaction) extensions to accommodate variable query costs and user preferences for early high-quality responses. Our algorithms are theoretically grounded and require no offline fine-tuning or dataset-specific training. Experiments on diverse benchmarks demonstrate that our methods outperform existing LLM routing strategies in both accuracy and cost-efficiency, validating the power of contextual bandits for real-time, adaptive LLM selection.

Vector data trading is essential for cross-domain learning with vector databases, yet it remains largely unexplored. We study this problem under online learning, where sellers face uncertain retrieval costs and buyers provide stochastic feedback to posted prices. Three main challenges arise: (1) heterogeneous and partial feedback in configuration learning, (2) variable and complex feedback in pricing learning, and (3) inherent coupling between configuration and pricing decisions. We propose a hierarchical bandit framework that jointly optimizes retrieval configurations and pricing. Stage I employs contextual clustering with confidence-based exploration to learn effective configurations with logarithmic regret. Stage II adopts interval-based price selection with local Taylor approximation to estimate buyer responses and achieve sublinear regret. We establish theoretical guarantees with polynomial time complexity and validate the framework on four real-world datasets, demonstrating consistent improvements in cumulative reward and regret reduction compared with existing methods.

Large Language Models (LLMs) are revolutionizing how users interact with information systems, yet their high inference cost poses serious scalability and sustainability challenges. Caching inference responses, allowing them to be retrieved without another forward pass through the LLM, has emerged as one possible solution. Traditional exact-match caching, however, overlooks the semantic similarity between queries, leading to unnecessary recomputation. Semantic caching addresses this by retrieving responses based on semantic similarity, but introduces a fundamentally different cache eviction problem: one must account for mismatch costs between incoming queries and cached responses. Moreover, key system parameters, such as query arrival probabilities and serving costs, are often unknown and must be learned over time. Existing semantic caching methods are largely ad-hoc, lacking theoretical foundations and unable to adapt to real-world uncertainty. In this paper, we present a principled, learning-based framework for semantic cache eviction under unknown query and cost distributions. We formulate both offline optimization and online learning variants of the problem, and develop provably efficient algorithms with state-of-the-art guarantees. We also evaluate our framework on a synthetic dataset, showing that our proposed algorithms perform matching or superior performance compared with baselines.

Proximity graphs are the state-of-the-art solution for approximate nearest neighbor (ANN) search, supporting applications such as Web search and retrieval-augmented generation (RAG). Sustaining long-term performance requires adaptive pruning as data and query workloads evolve. However, existing approaches are largely static and uniform. Adaptive pruning faces three key challenges: temporal drift in data and query distributions, spatial heterogeneity across graph regions, and costly feedback due to graph-level evaluations. We present BANCO, a bandit-based framework for adaptive proximity graph pruning. BANCO unifies diverse pruning strategies within a common decision space and optimizes them via a drift-aware batched bandit algorithm. It addresses temporal drift through drift-aware updates, captures spatial heterogeneity using contextual features for region-specific pruning, and reduces evaluation costs through batched feedback aggregation. We establish a dynamic regret bound with sublinear loss and polynomial computational complexity. Extensive experiments on four real-world datasets demonstrate that BANCO helps maintain long-term ANN search efficiency and accuracy under evolving data and workloads.

Combinatorial multi-armed bandit (CMAB) is a fundamental online learning framework that can optimize cumulative rewards in networked systems under uncertainty. Real-world applications like content delivery and channel allocation often feature binary base arm rewards and nonlinear total reward functions. This paper introduces combinatorial logistic bandits (CLogB), a contextual CMAB framework with the base arm reward modeled as a nonlinear logistic function of the context, and the feedback is governed by a general arm-triggering process. We study CLogB with smooth reward functions, covering applications such as online content delivery, online multi-LLM selection, and dynamic channel allocation. Our first algorithm, CLogUCB, uses a variance-agnostic exploration bonus and achieves a regret bound of Õ(d√κKT), where d is the feature dimension, κ reflects logistic model nonlinearity,Kis the maximum number of triggered arms, and Õ ignores logarithmic factors. This improves on prior results by Õ (√κ). We further propose VA-CLogUCB, a variance-adaptive enhancement achieving regret bounds of Õ(d√KT) under standard smoothness conditions and Õ (d√T) under stronger variance conditions, removing dependence on K. For time-invariant feature maps, we enhance computational efficiency by avoiding nonconvex optimization while maintaining Õ(d√T) regret. Experiments on synthetic and real-world datasets validate the superior performance of our algorithms, demonstrating their effectiveness and scalability for real-world networked systems.

The contextual multi-armed bandit (MAB) problem is crucial in sequential decision-making. A line of research, known as online clustering of bandits, extends contextual MAB by grouping similar users into clusters, utilizing shared features to improve learning efficiency. However, existing algorithms, which rely on the upper confidence bound (UCB) strategy, struggle to gather adequate statistical information to accurately identify unknown user clusters. As a result, their theoretical analyses require several strong assumptions about the "diversity" of contexts generated by the environment, leading to impractical settings, complicated analyses, and poor practical performance. Removing these assumptions has been a long-standing open problem in the clustering of bandits literature. In this work, we provide two partial solutions. First, we introduce an additional exploration phase to accelerate the identification of clusters. We integrate this general strategy into both graph-based and set-based algorithms and propose two new algorithms, UniCLUB and UniSCLUB. Remarkably, our algorithms require substantially weaker assumptions and simpler theoretical analyses while achieving superior cumulative regret compared to previous studies. Second, inspired by the smoothed analysis framework, we propose a more practical setting that eliminates the requirement for i.i.d. context generation used in previous studies, thus enhancing the performance of existing algorithms for online clustering of bandits. Extensive evaluations on both synthetic and real-world datasets demonstrate that our proposed algorithms outperform existing approaches.

The combinatorial multi-armed bandit (CMAB) is a fundamental sequential decision-making framework, extensively studied over the past decade. However, existing work primarily focuses on the online setting, overlooking the substantial costs of online interactions and the readily available offline datasets. To overcome these limitations, we introduce Off-CMAB, the first offline learning framework for CMAB. Central to our framework is the combinatorial lower confidence bound (CLCB) algorithm, which combines pessimistic reward estimations with combinatorial solvers. To characterize the quality of offline datasets, we propose two novel data coverage conditions and prove that, under these conditions, CLCB achieves a near-optimal suboptimality gap, matching the theoretical lower bound up to a logarithmic factor. We validate Off-CMAB through practical applications, including learning to rank, large language model (LLM) caching, and social influence maximization, showing its ability to handle nonlinear reward functions, general feedback models, and out-of-distribution action samples that excludes optimal or even feasible actions. Extensive experiments on synthetic and real-world datasets further highlight the superior performance of CLCB.

Co-branding has become a vital strategy for businesses aiming to expand market reach within recommendation systems. However, identifying effective cross-industry partnerships remains challenging due to resource imbalances, uncertain brand willingness, and ever-changing market conditions. In this paper, we provide the first systematic study of this problem and propose a unified online-offline framework to enable co-branding recommendations. Our approach begins by constructing a bipartite graph linking ''initiating'' and ''target'' brands to quantify co-branding probabilities and assess market benefits. During the online learning phase, we dynamically update the graph in response to market feedback, while striking a balance between exploring new collaborations for long-term gains and exploiting established partnerships for immediate benefits. To address the high initial co-branding costs, our framework mitigates redundant exploration, thereby enhancing short-term performance while ensuring sustainable strategic growth. In the offline optimization phase, our framework consolidates the interests of multiple sub-brands under the same parent brand to maximize overall returns, avoid excessive investment in single sub-brands, and reduce unnecessary costs associated with over-prioritizing a single sub-brand. We present a theoretical analysis of our approach, establishing a highly nontrivial sublinear regret bound for online learning in the complex co-branding problem, and enhancing the approximation guarantee for the NP-hard offline budget allocation optimization. Experiments on both synthetic and real-world co-branding datasets demonstrate the practical effectiveness of our framework, with at least 12% improvement.

Conversational recommender systems proactively query users with relevant ''key terms'' and leverage the feedback to elicit users' preferences for personalized recommendations. Conversational contextual bandits, a prevalent approach in this domain, aim to optimize preference learning by balancing exploitation and exploration. However, several limitations hinder their effectiveness in real-world scenarios. First, existing algorithms employ key term selection strategies with insufficient exploration, often failing to thoroughly probe users' preferences and resulting in suboptimal preference estimation. Second, current algorithms typically rely on deterministic rules to initiate conversations, causing unnecessary interactions when preferences are well-understood and missed opportunities when preferences are uncertain. To address these limitations, we propose three novel algorithms: CLiSK, CLiME, and CLiSK-ME. CLiSK introduces smoothed key term contexts to enhance exploration in preference learning, CLiME adaptively initiates conversations based on preference uncertainty, and CLiSK-ME integrates both techniques. We theoretically prove that all three algorithms achieve a tighter regret upper bound of O (√dTlogT) with respect to the time horizon T, improving upon existing methods. Additionally, we provide a matching lower bound Ω(√dT) for conversational bandits, demonstrating that our algorithms are nearly minimax optimal. Extensive evaluations on both synthetic and real-world datasets show that our approaches achieve at least a 14.6% improvement in cumulative regret.

In conversational recommender systems, interactively presenting queries and leveraging user feedback are crucial for efficiently estimating user preferences and improving recommendation quality. Selecting optimal queries in these systems is a significant challenge that has been extensively studied as a sequential decision problem. The expected value of information (EVOI), which computes the expected reward improvement, provides a principled criterion for query selection. However, it is computationally expensive and lacks theoretical performance guarantees. Conversely, conversational bandits offer provable regret upper bounds, but their query selection strategies yield only marginal regret improvements over non-conversational approaches. To address these limitations, we integrate EVOI within the conversational bandit framework by proposing a new conversational mechanism featuring two key techniques: (1) gradient-based EVOI, which replaces the complex Bayesian updates in conventional EVOI with efficient stochastic gradient descent, significantly reducing computational complexity and facilitating theoretical analysis; and (2) smoothed key term contexts, which enhance exploration by adding random perturbations to uncover more specific user preferences. Our approach applies to both Bayesian (Thompson Sampling) and frequentist (UCB) variants of conversational bandits. We introduce two new algorithms, ConTS-EVOI and ConUCB-EVOI, and rigorously prove that they achieve substantially tighter regret bounds, with both algorithms offering a √d improvement in their dependence on the time horizon T, where d is the dimension of the feature space. Extensive evaluations on synthetic and real-world datasets validate the effectiveness of our methods.

In this paper, we investigate the problem of enhancing cooperative LiDAR-based perception accuracy in vehicular edge networks. The key to solving this problem is the selection of connected and autonomous vehicles (CAVs) that can collectively provide maximum perception performance. In specific, extensive motivating experiments on an open benchmark dataset are conducted, which reveal that the cooperative perception accuracy is a submodular combination of selected CAVs, and such selection is non-trivial due to high vehicular mobility as well as unstable vehicular network conditions. Then, we develop an Edge coordinated COoperative Perception (ECOP) framework, taking into account both cooperative vehicle selection and adaptive bandwidth allocation. The novelty of the ECOP design is threefold. First, a new metric named perceptual gain is designed, which properly quantifies the individual perception contributions of each CAV without incurring additional computational overhead. Secondly, an online vehicle selection strategy, which utilizes continual learning to assess the perceptual gain of each CAV, is devised. Theoretical analysis indicates that the proposed vehicle selection strategy can achieve asymptotically diminishing learning regret, highlighting its effectiveness in adapting to vehicular mobility. Finally, an optimal bandwidth allocation method is proposed, which can adapt to heterogeneous and unstable vehicular network conditions. Simulation results demonstrate that, compared with other benchmarks, ECOP can select vehicle sets with the highest cooperative perception accuracy and ensure real-time perception in the presence of fluctuating bandwidth. Furthermore, a case study is presented to visualize the effectiveness of the proposed ECOP framework.

The Probabilistic Maximum Coverage (PMC) problem plays a pivotal role in modeling various network applications, such as mobile crowdsensing, which involves selecting nodes within a graph that probabilistically cover other nodes. Our study focuses on PMC within the framework of online learning, termed the PMC bandit, where the network parameters are initially unknown. In this scenario, the decision-maker is tasked with learning these parameters to maximize the cumulative rewards from covered nodes. Despite prior research on the PMC bandit, we propose a novel variant, dynamic PMC-G bandit, which extends the semi-bandit feedback model to represent applications more accurately. To tackle the complexities of the time-varying combinatorial arm set rather than traditional static, we enhance the Combinatorial Upper Confidence Bound (CUCB) algorithms by developing two innovative variance-aware strategies: the Variance-Adaptive Combinatorial Upper Confidence Bound (VACUCB) for probabilistically triggered arms, and the Action-Based Combinatorial Upper Confidence Bound (ABCUCB) for self-reliant arms, i.e., independent arms with probabilistically triggered outcomes. Based on variance-aware properties, our contributions notably reduce the dependence on the number of nodes$K$selected per round, demonstrating that: (i) VACUCB effectively minimizes the regret associated with the triggered arms, enhancing the CUCB by a factor of$\tilde{O}(K)$; (ii) ABCUCB further diminishes the dependence on$K$in the leading term. Empirical results from synthetic and real-world datasets confirm that our proposed algorithms outperform current benchmarks in three network applications.

Optimizing network protocols is crucial for improving application performance. Recent research works use multi-armed bandit (MAB) online learning methods to address network optimization problems, aiming to improve cumulative payoffs such as network throughput. However, existing MAB frameworks are ineffective since they inherently assume the network environment is static, or they have high complexity in detecting environmental changes. In this work, we advocate using lightweight "network-assist" techniques together with online learning to optimize network protocols, and show it can effectively detect environmental changes and maximize network performance. Furthermore, optimizing network protocols often face two types of decision (or arm) spaces: discrete and continuous choices, while most prior MAB models only handle discrete settings. This paper proposes a framework capable of managing both spaces. To our best knowledge, we are the first to develop an MAB framework that incorporates network-assist signals in handling dynamic environments, while considering the distinct characteristics of discrete and continuous arm spaces. Our framework can achieve optimality by showing its sub-linear regret bound, matching the state-of-the-art results in several degenerate cases. We also illustrate how to apply our framework to two network applications: (1) wireless network channel selection, and (2) rate-based TCP congestion control. We demonstrate the merits of our algorithms via both numerical simulations and packet-level experiments.

The rapid evolution of multimedia and computer vision technologies requires adaptive visual model deployment strategies to effectively handle diverse tasks and varying environments. This work introduces AxiomVision, a novel framework that can guarantee accuracy by leveraging edge computing to dynamically select the most efficient visual models for video analytics under diverse scenarios. Utilizing a tiered edge-cloud architecture, AxiomVision enables the deployment of a broad spectrum of visual models, from lightweight to complex DNNs, that can be tailored to specific scenarios while considering camera source impacts. In addition, AxiomVision provides three core innovations: (1) a dynamic visual model selection mechanism utilizing continual online learning, (2) an efficient online method that efficiently takes into account the influence of the camera's perspective, and (3) a topology-driven grouping approach that accelerates the model selection process. With rigorous theoretical guarantees, these advancements provide a scalable and effective solution for visual tasks inherent to multimedia systems, such as object detection, classification, and counting. Empirically, AxiomVision achieves a 25.7% improvement in accuracy.

In the domain of conversational recommendation systems (CRSs), the development of recommenders capable of eliciting user preferences through conversation has marked a significant advancement. These systems have been enhanced by incorporating conversational key-terms related to items, which streamline the recommendation process by reducing the extensive exploration that traditional interactive recommenders necessitate. Despite these advancements, CRSs still face significant challenges. The vast number of users and the difficulty in accurately capturing preferences lead to persistent inaccuracies, even when direct user interactions are employed to refine the understanding of user preferences. To tackle these challenges, we propose two innovative bandit algorithms: RCLUMB (Robust Clustering of Misspecified Bandits) and RSCLUMB (Robust Set-based Clustering of Misspecified Bandits). These algorithms employ dynamic graphs and evolving cluster sets, respectively, to represent the changing structure of user preferences, thus leveraging collaborative user preferences to accelerate the learning process. Our algorithms are designed to be resilient against errors in preference modeling and the resulting inaccuracies in clustering. We rigorously analyze the performance of our algorithms and establish regret upper bounds of$O(\epsilon _*T\sqrt{md\log T} + d\sqrt{mT}\log T)$under milder assumptions than previous works, matching the state-of-the-art results in several degenerate cases. Through extensive experiments on synthetic and real-world datasets, our algorithms demonstrate superior performance over existing algorithms.

Conversational recommendation systems (CRSs) are increasingly prevalent, but they are susceptible to the influence of corrupted user behaviors, such as deceptive click ratings. These behaviors can skew the recommendation process, resulting in suboptimal results. Traditional bandit algorithms, which are typically oriented to single users, do not capitalize on implicit social connections between users, which could otherwise enhance learning efficiency. Furthermore, they cannot identify corrupted users in a real-time, multi-user environment. In this paper, we propose a novel bandit problem, Online Learning and Detecting Corrupted Users (OLDCU), to learn and utilize unknown user relations from disrupted behaviors to speed up learning and detect corrupted users in an online setting. This problem is non-trivial due to the dynamic nature of user behaviors and the difficulty of online detection. To robustly learn and leverage the unknown relations among potentially corrupted users, we propose a novel bandit algorithm RCLUB-WCU, incorporating a conversational mechanism. This algorithm is designed to handle the complexities of disrupted behaviors and to make accurate user relation inferences. To detect corrupted users with bandit feedback, we further devise a novel online detection algorithm, OCCUD, which is based on RCLUB-WCU’s inferred user relations and designed to adapt over time. We prove a sub-linear regret bound for RCLUB-WCU, demonstrating its efficiency. We also analyze the detection accuracy of OCCUD, showing its effectiveness in identifying corrupted users. Through extensive experiments, we validate the performance of our methods. Our results show that RCLUB-WCU and OCCUD outperform previous bandit algorithms and achieve high corrupted user detection accuracy, providing robust and efficient solutions in the field of CRSs.

Video camera plays a growing important role in advancing industrial control towards a higher level of automation. Thus, video analytics become highly demanded, especially for low-latency and high-accuracy analytic results. Yet, the data volume produced by camera clusters is prohibitively high. In this article, we proposeRespire, a system that can remove redundant frames for reducing the cost of transmission and processing based on edge computing nodes, while maintaining useful frames for high analytic accuracy. Specifically,Respireincorporates a new way for characterizing the spatial–temporal redundancy between frames. Then,Respireprioritizes the uploading of frames for redundancy reduction. As the search space of the entire collected frames is exponential for the set of frames containing the maximal information, we jointly consider offline and online pruning of frames and propose a heuristic algorithm to reduce the search space. Extensive real-world dataset-based experiments demonstrate that the proposed system can significantly reduce communication and computation costs, while providing sufficient information for guaranteed video analytic accuracy.