Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Deep Bandits Show-Off: Simple and Efficient Exploration with Deep Networks
Rong Zhu, Mattia Rigotti
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 |
|---|---|---|---|
| add SAU-Sampling exploration to a neural network instead of NeuralLinear posterior sampling | Neural-SAU-S · Statlog bandit | relative cumulative regret: 1.26 -> 0.62 | Improved |
| replace NeuralLinear with Neural-SAU-S | NeuralLinear · Mushroom bandit | relative cumulative regret: 2.66 -> 2.20 | Improved |
| replace NeuralLinear with Neural-SAU-S | NeuralLinear · Financial bandit | relative cumulative regret: 10.16 -> 5.60 | Improved |
| replace NeuralLinear with Neural-SAU-S | NeuralLinear · Jester bandit | relative cumulative regret: 69.48 -> 61.02 | Improved |
| replace NeuralLinear with Neural-SAU-S | NeuralLinear · Adult bandit | relative cumulative regret: 74.62 -> 78.02 | Hurt |
| use SAU-UCB with linear model instead of LinearPosterior | LinearPosterior · Financial bandit | relative cumulative regret: 10.77 -> 6.51 | Improved |
| use SAU-UCB with linear model instead of LinearPosterior | LinearPosterior · Census bandit | relative cumulative regret: 46.70 -> 39.98 | Improved |
| use SAU-Sampling versus SAU-UCB with neural network (ablation) | Neural-SAU · 7 real-world bandits | relative cumulative regret: 2.20 -> 2.32 (Mushroom) | No clear effect |
| NeuralLinear compared to Neural-SAU variants | NeuralLinear · 7 real-world bandits | cumulative regret: 32% higher than Neural-SAU-S | Hurt |
| use SAU instead of NeuralLinear for runtime | Neural-SAU · 7 real-world bandits | execution time: more than 10 time faster | Improved |
| use SAU-Neural-S at low delta in Wheel Bandit | NeuralLinear · Wheel Bandit delta=0.50 | relative cumulative regret: 18.71 -> 12.49 | Improved |
| use SAU-Neural-S at very high delta 0.99 | NeuralLinear · Wheel Bandit delta=0.99 | relative cumulative regret: 86.03 -> 106.89 | Hurt |
Closest papers by meaning
- Feel-Good Thompson Sampling for Contextual Bandits: a Markov Chain Monte Carlo Showdown (NeurIPS 2025 · similarity 0.90)
- Provable and Practical: Efficient Exploration in Reinforcement Learning via Langevin Monte Carlo (ICLR 2024 · similarity 0.89)
- Scalable Exploration via Ensemble++ (NeurIPS 2025 · similarity 0.88)
- Contextual Thompson Sampling via Generation of Missing Data (NeurIPS 2025 · similarity 0.88)
- Exploration via Feature Perturbation in Contextual Bandits (NeurIPS 2025 · similarity 0.88)
- The Choice of Noninformative Priors for Thompson Sampling in Multiparameter Bandit Models (AAAI 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.
