2019:

L01:
Filled time (much too surprise)
Very rough - needs polishing
Link back to triangle better
Best to add more examples and drop Central limit rant

Add section on supervision/projects so they know who does what!


L02:
Only 25 minutes long
Despite white boarding all algorithms on 'further algorithms' slide!
(Need slides for them)
Need more intuition on crossover
Given number of questions about project leaving this a bit short probably wise.
Add a worked example?
Add golden section?


L03:
Right length, but move branch n' bound to L02 and restore content I didn't have time to do (except for RANSAC, which can also go to L02). More visualisations would help, but otherwise just missing polish.


L09:
20 minutes short, despite extra content - improved flow too much!
Flow could still be better though.
Hard to know what to add - maybe incremental GMM? Don't really cover elsewhere now so good to have one incremental algorithm. Single slide on mean shift?


L10:
Good length - had 5 minutes spare. Content is OK, but jumps around a bit - need to work on flow.


L11:
Given how rushed it was it worked surprisingly well. On time. Flow not great and slides noisy - needs polish and structure.


L12:
Super rushed in prep - improve everything! 20 minutes short - more on contextual bandits?


L13:
15 minutes early, but reasonable. Could probably do a bit more on semi-supervised. CRP needs slide to explain why before gummy bears.


L14:
Maybe add ARMA?
15 minutes early, lots of hand waving - needs more pretty pictures. Baum-Welch demo.
The big walkthrough could do with intermediate equations and differentiating the before/after cubes.


L15:
About right length
Front door explanation slightly wrong - backdoor for one term exists
Bit disjointed at points - need to work on flow
Example hard to follow - needs visuals / spread over more slides
Need to talk about where causal model comes from

