Slides-1, Slides-2. 15.1. Time and Uncertainty States and observations Stationary processes and the Markov assumption 15.2. Inference in Temporal Models (Discrete variables) Filtering: P(X_t | e_{1:t}) by FORWARD Prediction: P(X_{t+k} | e_{1:t}) by FORWARD Smoothing: P(X_k | e_{1:t}), 1<= k <t by FORWARD+BACKWARD Finding the most likely sequence: argmax {X_{1:t} | e_{1:t}) by Viterbi 15.3. Hidden Markov Models (One discrete hidden variable) Simplified matrix algorithms: 15.4. Kalman Filters (Continuous hidden variables) Updating Gaussian distributions A simple one-dimensional example The general case Applicability of Kalman filtering 15.5. Dynamic Bayesian Networks (arbitrary number of hidden variables) Constructing DBNs Exact inference in DBNs (Unrolling, variable elimination, etc.) Approximate inference in DBNs (Particle Filtering) 15.6. Speech Recognition Speech sounds Words Sentences Building a speech recognizer |