Thompson sampling
PulseAugur coverage of Thompson sampling — every cluster mentioning Thompson sampling across labs, papers, and developer communities, ranked by signal.
- instance of Gotit.pub 90%
- instance of alphaXiv 90%
- instance of ScienceCast 90%
- instance of CORE Recommender 90%
- instance of Influence Flower 90%
- used by Multi-armed bandits for adjudicating documents in pooling-based evaluation of information retrieval systems 90%
- used by CORE Recommender 70%
- used by alphaXiv 70%
- used by Influence Flower 70%
- instance of CatalyzeX 70%
- authored by alphaXiv 70%
- used by Gotit.pub 60%
7 day(s) with sentiment data
-
New algorithm Exp-KL-MS tackles multi-armed bandit challenges
Researchers have developed a new algorithm called Exponential-Kullback Leibler Maillard Sampling (Exp-KL-MS) to address the multi-armed bandit problem. This algorithm aims to simultaneously satisfy multiple performance …
-
LLM compute allocation hinges on prior information, not just policy
A new research paper explores optimal strategies for allocating computational resources to large language models (LLMs) when processing multiple inputs under a limited budget. The study, published on arXiv, suggests tha…
-
New AI agent uses game theory and moral rhetoric for bargaining games
Researchers have developed a novel agent for the GLEE Competition, designed to excel in bargaining, negotiation, and persuasion games. The agent integrates a game-theoretic solver with Bayesian opponent modeling and a l…
-
New AI Frameworks Accelerate Neural Activity Modeling and Chip Design
Researchers have developed Neural-DISCO, a framework for disentangled representation learning that allows for counterfactual modeling of neural activity. This approach enables the selective editing of behavioral or stim…
-
Self-learning router optimizes LLM costs by learning from outcomes
This article proposes a self-learning router for managing multiple Large Language Models (LLMs) to optimize costs and performance. Instead of fixed rules, the router learns from the outcomes of each model's responses, c…
-
New ranking bandit algorithms improve carousel interface recommendations
Researchers have developed new algorithms for ranking bandits in carousel interfaces, which allow recommender systems to observe user browsing depth. The proposed methods, including OD-TS, PBM-TS, and PBM-UCB, update it…
-
New Bayesian Neural Networks Enhance Deep Reinforcement Learning Exploration
Researchers have developed a novel class of Bayesian neural networks (BNNs) that place Dirichlet Process (DP) priors directly on the data-generating distribution, rather than on network parameters. This approach aims to…
-
AI agents to coordinate tasks via sequential auctions, replacing centralized orchestration
A new research paper proposes a decentralized framework for multi-step AI tasks using sequential auctions, moving away from traditional centralized orchestrators. This approach allows self-interested agents to bid on ta…
-
New Bayesian Bandit Method Selects Fair Classifiers
Researchers have developed a new method for selecting classifiers that adaptively balance predictive performance with subgroup fairness. This approach frames the problem as a Bayesian multi-armed bandit using Thompson S…
-
New AI framework optimizes delivery routes and dynamic pricing
Researchers have developed GREAT-ESDD, a novel framework that uses a Graph Edge Attention Network (GREAT) to optimize delivery routes and dynamic pricing for scenarios involving both home and out-of-home delivery option…
-
New arXiv papers refine bandit algorithms for AI optimization
Multiple arXiv papers published in October 2026 explore advanced strategies for linear bandit problems, aiming to improve regret bounds and computational efficiency. Research includes optimizing offline data utilization…
-
New arXiv papers refine Langevin sampling for LLMs and bandits
Two new arXiv papers explore advanced sampling techniques for machine learning. The first paper introduces Robust Langevin Sampling (R-LMC, A-LMC, RA-LMC) to address limitations in existing methods, particularly for fin…
-
New metric proposed for multi-armed bandit sample complexity
This paper introduces a new metric called expected sample complexity for stochastic multi-armed bandit problems, analyzed within a framework called approximately correct in expectation (ACE). The authors demonstrate tha…
-
FlexiFlow system dynamically switches ML models for improved accuracy
Researchers have developed FlexiFlow, a new dataflow system designed to dynamically switch between different machine learning models during inference. This adaptive switching aims to improve overall workflow accuracy an…
-
New research on arXiv explores bandit algorithms, adversarial attacks, and non-stationarity
Multiple research papers published on arXiv explore advancements in bandit algorithms, focusing on theoretical bounds, adversarial attacks, and non-stationary environments. One paper establishes a general lower bound fo…
-
Amazon Payments uses contextual bandits on SageMaker for AI personalization
Amazon Payments has implemented a multi-objective contextual multi-armed bandit model on Amazon SageMaker to personalize content for customers. This approach aims to optimize the selection of personalized content genera…
-
New evaluation method finds AI failures efficiently under budget
Researchers have developed a novel risk-aware adaptive evaluation method to more efficiently identify high-impact failures in interactive AI agents. This approach, detailed in a new arXiv paper, formulates evaluation as…
-
UC Berkeley researchers develop bandit-based pruning for transformers
Researchers from the University of California, Berkeley have developed a novel method for pruning large transformer models, including those used in vision and language tasks. This technique, framed as a damage-aware mul…
-
New arXiv papers advance multi-armed bandit algorithms and regret minimization
Multiple research papers published on arXiv explore advancements in multi-armed bandit algorithms. One paper addresses optimal switching regret for bandits with an oblivious adversary, proposing a single algorithm that …
-
New Thompson Sampling variant \"alpha-TS\" offers generalized regret analysis
Researchers have developed a generalized regret analysis for Thompson sampling, a popular algorithm for solving stochastic multi-armed bandit problems. This new approach, termed \"alpha-TS,\" utilizes a fractional poste…