Research map / Multi-agent systems and games / Games and equilibria
Deep Reinforcement Learning for Green Security Games with Real-Time Information
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 local modes followed by global-mode training (ablation) | DeDOL · 5×5 and 7×7 grids, Random and Gaussian maps | patroller expected utility: −1.63 -> −0.42; −0.43 -> 0.60; −2.00 -> −0.54; −4.15 -> −2.35 | Helped |
| Add local-mode strategies without global-mode retraining (ablation) | DeDOL · 5×5 Random | patroller expected utility: −1.63 -> −0.25 | Helped |
| Add local-mode strategies without global-mode retraining (ablation) | DeDOL · 5×5 Gaussian | patroller expected utility: −0.43 -> −2.41 | Hurt |
| Add local-mode strategies without global-mode retraining (ablation) | DeDOL · 7×7 Random and Gaussian | patroller expected utility: −2.00 -> −1.72; −4.15 -> −2.62 | Mixed |
| Use exploration with α = 0.15 rather than α = 0 (ablation) | DeDOL-S global mode · 3×3 Random and Gaussian | patroller expected utility: 0.60 -> 0.73; 0.12 -> 0.64 | Helped |
| Use domain-specific heuristic initial strategies instead of random DQN initialization | DeDOL / PSRO · 5×5 and 7×7 grids | patroller expected utility: -8.98 vs -1.63; -9.09 vs 0.60; -10.65 vs -0.54; -10.08 vs -2.35 | Helped |
| Use enhanced DQN (dueling double DQN with gradient clipping) instead of the compared patroller strategies | Best-response approximation · 7×7 grids against heuristic random-walk poacher | expected utility | Helped |
| Use actor-critic instead of enhanced DQN | Best-response approximation · 7×7 grids against heuristic random-walk poacher | expected utility | Hurt |
| Use DeDOL local + global instead of CFR | 3×3 GSG-I · 3×3 Random and Gaussian | patroller expected utility: 0.85 vs 1.01; 0.86 vs 1.05 | Hurt |
| Use DeDOL local + global instead of random sweeping | Patroller · All tested grid sizes and map types | patroller expected utility: -0.04 -> 0.85; -0.09 -> 0.86; -1.91 -> -0.42; -1.16 -> 0.60; -4.06 -> -0.54; -4.25 -> -2.35 | Helped |
Closest papers by meaning
- GUARD: Constructing Realistic Two-Player Matrix and Security Games for Benchmarking Game-Theoretic Algorithms (NeurIPS 2025 · similarity 0.87)
- PolicyGuard: Towards Test-time and Step-level Adversary Defense for Reinforcement Learning Agent (ICML 2026 · similarity 0.85)
- Solving Continuous Mean Field Games: Deep Reinforcement Learning for Non-Stationary Dynamics (NeurIPS 2025 · similarity 0.84)
- Equilibrium Policy Generalization: A Reinforcement Learning Framework for Cross-Graph Zero-Shot Generalization in Pursuit-Evasion Games (NeurIPS 2025 · similarity 0.84)
- Exploring the Efficacy of Multi-Agent Reinforcement Learning for Autonomous Cyber Defence: A CAGE Challenge 4 Perspective (AAAI 2025 · similarity 0.84)
- Learning Mean Field Games on Sparse Graphs: A Hybrid Graphex Approach (ICLR 2024 · similarity 0.83)
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.