Module 8 part 1

OMSCS/Courses/AI/Module_08/Module 8 - Pattern Recognition through Time.md

@@ -0,0 +1,57 @@
+---
+tags:
+  - OMSCS
+  - AI
+---
+# Module 8 - Pattern Recognition through Time
+
+This module follows chapter 14 from AIMA: [[AIMA - Chapter 14 - Probabilistic Reasoning over Time]]
+
+## Warping
+- For detecting patterns which might have different y-intercepts, encode the changes in the signal's values, rather than the signal's raw values
+- For detecting patterns which might stretch and compress, do some time warping
+
+![[Pasted image 20240325101744.png]]
+
+![[Pasted image 20240325101813.png]]
+
+![[Pasted image 20240325101847.png]]
+
+## Dynamic Time Warp
+
+![[Pasted image 20240325101959.png]]
+
+Try to match values between sequences, but try to keep to the diagonal when possible.
+
+Distance in this example is $\sqrt{34}$, much lower compared to the euclidean distance of $\sqrt{170}$.
+
+## Sakoe Chiba Bounds
+![[Pasted image 20240325102302.png]]
+
+Similar to the previous algorithm, but we don't allow the dynamic warping to deviate from some bounds around the diagonal.
+
+## Hidden Markov Models
+- module focuses on 1st order HMMs
+
+![[Pasted image 20240325121112.png]]
+
+### HMMs: I vs We
+![[Pasted image 20240325121514.png]]
+
+![[Pasted image 20240325121551.png]]
+
+### Viterbi Trellis
+![[Pasted image 20240325122002.png]]
+
+![[Pasted image 20240325122317.png]]
+
+![[Pasted image 20240325122630.png]]
+
+- This is the part from the textbook chapter
+
+### New Observation for "I"
+![[Pasted image 20240325130548.png]]
+
+## New Observation for "We"
+![[Pasted image 20240325130609.png]]
+