-
Learning Process Model
-
Function curves
- linear
- geometric
-
Bayesian updating
- trial-by-trial behavior not satisfactory
- asymptotic convergence and consistent
- Markov chain
- Machine learner?
-
Desired prop.
- knowledge increases on average
- abstracting out a func. from certain model
-
Stochastic System Model
-
States
- connection/detection
- Transitions
- Reward/Cost measure
- Controls
-
Issues
-
Sample path dependency
- ``long-run behavior'' summary as solution
-
Analysis
- Time-separable trick
-
MDP
-
SSP
- relentless attacker
- Never-Expel dominates
-
Discounted
- prob. of attacker giving up