Research map / Probabilistic and scientific modeling / Molecules and proteins
NSGZero: Efficiently Learning Non-exploitable Policy in Large-Scale Network Security Games with Neural Monte Carlo Tree Search
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 neural MCTS planning rather than model-free NSG-NFSP | NSGZero · 7x7 and 15x15 synthetic grids | worst-case defender reward: around 0.88 and 0.95 vs 0.54 and 0.48 | Helped |
| Use decentralized resource control rather than centralized control | NSGZero · Manhattan, m=6; Singapore, m=8 | approximate worst-case defender reward: 0.3660 $\pm$ 0.0299 vs 0.1020 $\pm$ 0.0188; 0.4140$\pm$ 0.0306 vs 0.2590 $\pm$ 0.0272 | Helped |
| Increase simulations from 5 to 15 (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | defender win rate: 0.47 to around 0.6 at N=5; around 0.9 at N=15 | Helped |
| Increase simulations beyond 15 (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | defender win rate | No clear effect |
| Use a large PUCT exploration constant (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | defender win rate | Hurt |
| Use temperature 4 when generating policies (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | final defender win rate: about 0.75 vs 0.9 at lower temperatures | Hurt |
| Use very low policy temperature (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | defender win rate | Hurt |
| Train the value network with CE instead of MSE (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | final defender win rate: about 0.75 to 0.9 | Helped |
| Use gamma=1 instead of gamma=0.95 (ablation) | NSGZero · 7x7 grid, fixed random-path attacker | defender win rate | Helped |
Closest papers by meaning
- Tree-Based Stochastic Optimization for Solving Large-Scale Urban Network Security Games (AAAI 2026 · similarity 0.86)
- Deep Reinforcement Learning for Green Security Games with Real-Time Information (AAAI 2019 · similarity 0.86)
- Policy Learning for Continuous Space Security Games Using Neural Networks (AAAI 2018 · similarity 0.85)
- Equilibrium Policy Generalization: A Reinforcement Learning Framework for Cross-Graph Zero-Shot Generalization in Pursuit-Evasion Games (NeurIPS 2025 · similarity 0.84)
- Learning Mean Field Games on Sparse Graphs: A Hybrid Graphex Approach (ICLR 2024 · similarity 0.83)
- Learning Equilibria in Adversarial Team Markov Games: A Nonconvex-Hidden-Concave Min-Max Optimization Problem (NeurIPS 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.