Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Neural Exploitation and Exploration of Contextual Bandits
Yikun Ban, Yuchen Yan, Arindam Banerjee, Jingrui He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learned exploration network f2 instead of UCB bound (EE-Net vs NeuralUCB) | neural contextual bandit · MNIST | cumulative regret: 943.5 -> 842.3 | Improved |
| EE-Net instead of NeuralTS | neural contextual bandit · Disin | cumulative regret: 523.2 -> 476.4 | Improved |
| EE-Net instead of Neural-epsilon | neural contextual bandit · Yelp | cumulative regret: 5276.1 -> 4403.1 | Improved |
| EE-Net instead of NeuralTS | neural contextual bandit · Movielens | cumulative regret: 1583.1 -> 1472.4 | Improved |
| neural networks instead of linear UCB | contextual bandit · MNIST | cumulative regret: 7863.2 -> 842.3 | Improved |
| decay epsilon over time (Neural-epsilon+) (ablation) | Neural-epsilon · Movielens | cumulative regret: 1573.4 -> 1578.4 | No clear effect |
| decay epsilon over time (Neural-epsilon+) (ablation) | Neural-epsilon · Yelp | cumulative regret: 5276.1 -> 5162.4 | Improved |
| raise exploration network input dimension from 10 to 200 (ablation) | EE-Net · Movielens | cumulative regret: 1472.3 -> 1452.1 | Improved |
| raise exploration network input dimension from 200 to 500 (ablation) | EE-Net · Movielens | cumulative regret: 1452.1 -> 1467.6 | Hurt |
| use label y1 = r - f1 instead of abs or ReLU variants for f2 (ablation) | EE-Net · Movielens and MNIST | cumulative regret | Improved |
| add exploration network f2 (extra training cost) | EE-Net vs NeuralUCB/TS · training time | computation: approximately 38-46% more computation | Hurt |
| replace matrix-inverse UCB with single forward pass of f2 | EE-Net vs NeuralUCB/TS · inference time | decision time: 32-59% faster inference | Improved |
Closest papers by meaning
- Feel-Good Thompson Sampling for Contextual Bandits: a Markov Chain Monte Carlo Showdown (NeurIPS 2025 · similarity 0.89)
- Contextual Bandits with Online Neural Regression (ICLR 2024 · similarity 0.89)
- Exploration via Feature Perturbation in Contextual Bandits (NeurIPS 2025 · similarity 0.89)
- Neural Dueling Bandits: Preference-Based Optimization with Human Feedback (ICLR 2025 · similarity 0.89)
- Feel-Good Thompson Sampling for Contextual Dueling Bandits (ICML 2024 · similarity 0.88)
- Robust Neural Contextual Bandit against Adversarial Corruptions (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
