Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Combinatorial Multi-Armed Bandit and Its Extension to Probabilistically Triggered Arms (1407.8339v6)

Published 31 Jul 2014 in cs.LG

Abstract: We define a general framework for a large class of combinatorial multi-armed bandit (CMAB) problems, where subsets of base arms with unknown distributions form super arms. In each round, a super arm is played and the base arms contained in the super arm are played and their outcomes are observed. We further consider the extension in which more based arms could be probabilistically triggered based on the outcomes of already triggered arms. The reward of the super arm depends on the outcomes of all played arms, and it only needs to satisfy two mild assumptions, which allow a large class of nonlinear reward instances. We assume the availability of an offline (\alpha,\beta)-approximation oracle that takes the means of the outcome distributions of arms and outputs a super arm that with probability {\beta} generates an {\alpha} fraction of the optimal expected reward. The objective of an online learning algorithm for CMAB is to minimize (\alpha,\beta)-approximation regret, which is the difference between the \alpha{\beta} fraction of the expected reward when always playing the optimal super arm, and the expected reward of playing super arms according to the algorithm. We provide CUCB algorithm that achieves O(log n) distribution-dependent regret, where n is the number of rounds played, and we further provide distribution-independent bounds for a large class of reward functions. Our regret analysis is tight in that it matches the bound of UCB1 algorithm (up to a constant factor) for the classical MAB problem, and it significantly improves the regret bound in a earlier paper on combinatorial bandits with linear rewards. We apply our CMAB framework to two new applications, probabilistic maximum coverage and social influence maximization, both having nonlinear reward structures. In particular, application to social influence maximization requires our extension on probabilistically triggered arms.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Wei Chen (1290 papers)
  2. Yajun Wang (29 papers)
  3. Yang Yuan (52 papers)
  4. Qinshi Wang (4 papers)
Citations (275)

Summary

An Analysis of Combinatorial Multi-Armed Bandit Frameworks with Applications

The paper presents a comprehensive paper on the combinatorial multi-armed bandit (CMAB) problem, an extension of the classical multi-armed bandit (MAB) model. This extension addresses scenarios where complex interactions between actions require consideration of multiple arms and probabilistic outcomes. The proposed framework is particularly valuable in dealing with real-world applications characterized by combinatorial structures and non-linear reward functions.

Core Framework and Theoretical Contributions

The CMAB framework introduces a set of base arms, whose outcomes are governed by unknown distributions. These base arms combine to form super arms, which are selected in each round of decision-making. The framework accommodates scenarios where additional base arms can be probabilistically triggered, expanding the model's applicability to more complex systems like social networks.

The paper proposes an algorithm called CUCB (Combinatorial Upper Confidence Bound) designed to operate within this framework. A primary innovation of CUCB is its ability to minimize the (α,β)(\alpha,\beta)-approximation regret, quantified as the difference between the maximum realizable expected reward and the reward obtained by the algorithm under certain approximation constraints. The CUCB algorithm achieves a distribution-dependent regret of O(logn)O(\log n) similar to the classical UCB1 algorithm, and provides rigorous analysis with explicit dependency on the CMAB parameters.

The regret bounds established in the paper demonstrate that the CMAB framework effectively handles a large class of reward functions, both linear and non-linear, extending the technique’s applicability beyond simple bandit problems. Importantly, the results cover distribution-independent scenarios, yielding regret bounds that function for arbitrary distributions over a wide range of problem instances.

Applications and Implications

Two applications are discussed in detail: probabilistic maximum coverage for online advertising and social influence maximization in viral marketing, both representing instances where rewards are inherently non-linear. The CMAB framework provides a structured approach to handling the challenges posed by these applications, particularly the combinations of actions and uncertainties involved in real-world settings.

  • Probabilistic Maximum Coverage: This problem has practical implications in online advertising. The CMAB framework models the probabilistic interactions between web pages and users, aiming to optimize ad placements across an unknown set of user interactions. The model provides robust methods to anticipate the effectiveness of different strategies over repeated experiments.
  • Social Influence Maximization: This application in viral marketing leverages the CMAB framework to address cascading effects across social networks. It models the spread of influence in networks where the path and probability of influence are not fully known. This setup allows for exploring new strategies to enhance information dissemination, optimizing seed node selections in complex network structures.

Theoretical and Practical Implications

From a theoretical standpoint, this work extends the exploration-exploitation trade-off established in simpler MAB models to much more complex decision-making spaces. Practically, the framework's design allows implementation across a variety of domains with structured combinatorial relationships, making it highly versatile in digital advertising, recommendation systems, and network science.

Speculatively, future developments in this line of research may deepen the exploration of CMAB frameworks in the context of adaptive systems and reinforcement learning, where Markovian dependencies or restless bandits might be considered. Addressing computational efficiency and further refining the approximation algorithms could also lead to more responsive real-time systems adaptable to rapidly changing environments.

This paper's framework and algorithmic solutions provide a solid foundation for advancing both theoretical understandings and practical implementations of multi-armed bandit problems in complex, stochastic environments.