Dynamic Bayesian networks
Extend BNs to exploit temporal redundancy:
- Markov: future independent of past given present;
- Stationarity: transition model stable over time.
Compact representation for transition model:
DBNs have same scaling problem as BNs