Research map / Learning theory and algorithms / Sample complexity
PER-ETD: A Polynomially Efficient Emphatic Temporal Difference Learning Method
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 |
|---|---|---|---|
| Restart the follow-on trace every update and use a logarithmically increasing period | ETD(0) · Off-policy value evaluation (theoretical) | sample complexity: exponential sample complexity to be polynomials | Helped |
| Use PER-ETD(0) instead of TD | BAIRD counter-example · BAIRD counter-example | convergence | Helped |
| Use PER-ETD(0) instead of vanilla ETD(0) | BAIRD counter-example · BAIRD counter-example | stability | Helped |
| Increase the period length from a small value | PER-ETD(0) · BAIRD counter-example | bias error | Helped |
| Increase the period length beyond the lower-bias range | PER-ETD(0) · BAIRD counter-example | variance | Hurt |
| Use a properly chosen finite period length | PER-ETD(0) · BAIRD counter-example, large distribution mismatch | convergence | Helped |
| Increase distribution mismatch | PER-ETD(0) · BAIRD counter-example, varying target or behavior policies | convergence rate: $\rho_{max}=1.17, 1.40, 2.80, 4.20,5.60$ | Hurt |
| Set λ=0 rather than other tested values | PER-ETD(λ) with feature Φ1 · BAIRD counter-example, Φ1 | bias error | Helped |
| Set λ=1 rather than other tested values | PER-ETD(λ) with feature Φ2 · BAIRD counter-example, Φ2 | bias error | Helped |
| Set λ=0.4 rather than λ=0 or λ=1 | PER-ETD(λ) with feature Φ3 · BAIRD counter-example, Φ3 | bias error | Helped |
| Use λ=1 instead of selecting λ for the feature geometry | PER-ETD(λ) · BAIRD counter-example, finite period length | bias error | Mixed |
Closest papers by meaning
- Generalized Emphatic Temporal Difference Learning: Bias-Variance Analysis (AAAI 2016 · similarity 0.93)
- A Non-asymptotic Analysis of Non-parametric Temporal-Difference Learning (NeurIPS 2022 · similarity 0.88)
- Consistent On-Line Off-Policy Evaluation (ICML 2017 · similarity 0.88)
- Bridging the Gap Between Average and Discounted TD Learning (ICML 2026 · similarity 0.88)
- The surprising efficiency of temporal difference learning for rare event prediction (NeurIPS 2024 · similarity 0.87)
- Discerning Temporal Difference Learning (AAAI 2024 · similarity 0.85)
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.