Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Thresholded Lasso Bandit
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use two-step Lasso thresholding and least-squares estimation on the selected support | Sparse contextual linear bandits · Simulated settings across the tested K, d, and s₀ values | cumulative regret | Helped |
| Use two-step rather than single-step thresholding (ablation) | TH Lasso Bandit · Sparse support estimation | false positives: \mathcal{O}(s_0) to \mathcal{O}(\sqrt{s_0}) | Helped |
| Use two-step rather than single-step thresholding (ablation) | TH Lasso Bandit · Sparse contextual linear bandit | regret upper-bound coefficients: coefficients on \log T and \sqrt{T} terms | Helped |
| Use the plain Lasso estimate for arm selection instead of thresholding and restricted least squares (ablation) | TH Lasso Bandit · Sparse contextual linear bandit, without margin | regret upper bound: \mathcal{O} ( \log d + \sqrt{T \log (dT)}) | Hurt |
| Add thresholding to the Lasso estimate before arm selection (ablation) | TH Lasso Bandit · Sparse contextual linear bandit, without margin | regret upper bound: \mathcal{O}(s_0^2 \log d + \sqrt{s_0 T }) | Helped |
| Use TH Lasso Bandit instead of SA Lasso bandit | Sparse contextual linear bandits · K=2, d=1000, s₀=5; estimation comparison | ℓ₂ parameter-estimation error | Helped |
| Use TH Lasso Bandit instead of DR Lasso bandit or SA Lasso bandit | Sparse contextual linear bandits · K=2, d=1000, s₀=5; estimation comparison | false positives | Helped |
| Use TH Lasso Bandit instead of SA Lasso bandit | Sparse contextual linear bandits · K=2, d=1000, s₀=5; varying A_max | cumulative regret at t=1000 | Helped |
| Vary λ₀ over the tested range (ablation) | TH Lasso Bandit · K=2, d=1000, s₀=5, A_max=10 | cumulative regret at t=1000 | No clear effect |
| Vary λ₀ over the tested range (ablation) | SA Lasso bandit · K=2, d=1000, s₀=5, A_max=10 | cumulative regret at t=1000 | Hurt |
| Remove the margin condition (ablation) | TH Lasso Bandit · Sparse contextual linear bandit | regret upper bound: \mathcal{O}(s_0^2 \log d + s_0\log T) to \mathcal{O}(s_0^2 \log d + \sqrt{s_0 T }) | Mixed |
Closest papers by meaning
- Lasso Bandit with Compatibility Condition on Optimal Arm (ICLR 2025 · similarity 0.94)
- Sparsity-Agnostic Linear Bandits with Adaptive Adversaries (NeurIPS 2024 · similarity 0.91)
- High-dimensional Linear Bandits with Knapsacks (ICML 2024 · similarity 0.91)
- Causal Bandits: The Pareto Optimal Frontier of Adaptivity, a Reduction to Linear Bandits, and Limitations around Unknown Marginals (ICML 2024 · similarity 0.90)
- New Classes of the Greedy-Applicable Arm Feature Distributions in the Sparse Linear Bandit Problem (AAAI 2024 · similarity 0.90)
- Nearly Minimax Optimal Submodular Maximization with Bandit Feedback (NeurIPS 2024 · similarity 0.90)
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.