PulseAugur
EN
LIVE 08:45:34

New methods enhance contextual bandit algorithms with graph reduction and offline learning · 3 sources tracked

By PulseAugur Editorial · [3 sources] ·

Researchers have developed new methods for contextual bandits, a type of machine learning problem focused on making sequential decisions. One approach, GraphDR-LinUCB, utilizes graph dimensionality reduction to improve performance in recommendation and advertising systems by projecting arm features onto a spectral subspace. This method achieves a regret bound of $\wtO(k\sqrt{T})$, significantly outperforming full-dimensional methods and other graph-aware techniques on several real-world datasets. Another framework, Offline Estimation to Decisions (OE2D), reduces contextual bandit learning to offline regression, enabling near-optimal regret with fewer calls to an oracle, particularly for large action spaces. This framework introduces a new complexity measure, the Decision-Offline Estimation Coefficient (DOEC), which bridges offline and online bandit algorithm design. AI

IMPACT These advancements in contextual bandit algorithms could lead to more efficient and effective decision-making systems in areas like personalized recommendations and targeted advertising.

RANK_REASON The cluster contains two arXiv papers detailing new theoretical frameworks and algorithms for contextual bandits.

Read on arXiv cs.LG →

AI-generated summary · Google Gemini · from 3 sources. How we write summaries →

New methods enhance contextual bandit algorithms with graph reduction and offline learning · 3 sources tracked

COVERAGE [3]

  1. arXiv cs.LG TIER_1 English(EN) · Joyanta Jyoti Mondal, Ibne Farabi Shihab, Anuj Sharma ·

    Graph Dimensionality Reduction for Contextual Bandits: Structure-Specific Regret Bounds under Approximate Smoothness and Noisy Eigenspaces

    arXiv:2606.27917v1 Announce Type: new Abstract: Contextual bandits with graph-structured arms arise in recommendation, citation retrieval, and social advertising, where arms connected on a graph tend to share reward signal. Standard dimensionality reduction ignores this structure…

  2. arXiv cs.LG TIER_1 English(EN) · Anuj Sharma ·

    Graph Dimensionality Reduction for Contextual Bandits: Structure-Specific Regret Bounds under Approximate Smoothness and Noisy Eigenspaces

    Contextual bandits with graph-structured arms arise in recommendation, citation retrieval, and social advertising, where arms connected on a graph tend to share reward signal. Standard dimensionality reduction ignores this structure, inflating exploration cost by a factor of $d/k…

  3. arXiv stat.ML TIER_1 English(EN) · Hao Qin, Chicheng Zhang ·

    Taming the Monster Every Context: Complexity Measure and Unified Framework for Offline-Oracle Efficient Contextual Bandits

    arXiv:2602.09456v2 Announce Type: replace-cross Abstract: We propose an algorithmic framework, Offline Estimation to Decisions (OE2D), that efficiently reduces contextual bandit learning with general reward function approximation to offline regression. The framework allows near-o…

RELATED ENTITIES

RELATED TOPICS