notes

• W1D2 - Comparing Tasks
• decoupled loss function, use of labels, argument to be made that this constitutes supervised learning (I disagree) but explanation on this point and some typos need to need to be made more specific. Attempted rewrite to make explanation better didn't quite hit the mark here and led to some student confusion (see discord)
• W1D3
• A bit of confusion for the tutorial 5 where we eventually did not find the expected output regarding Oval/Walk not being less dynamically similar than Circle/Oval trajectory... Anyone else with this? Also, that last little suggestion on how to go further as a little exercise, which sounds nice to do was a bit confusing on how it is written, making it impossible for us to try it out: To better show the difference between DSA and RSA, we encourage you to run another comparison where we consider each time step to be a pair in the X,Y space and we will look at the the similarity between of Oval with both Circle and Walk."
• Synthetic data plotting issue
• "ok but aside from general concepts, can anyone clarify why oval and random walk trajectories (which are said to be similar in shape) have a higher DSA (=are less similar) than circle and random walk?"
• "also for the future, some revision of tutorial 5 would be helpful - specifically expliciting the comparison between RSA and DSA (which is only thrown there) and giving more elements to understand the bonus on recurrent models (where without looking at collapsed cell it's difficult to understand on what the example network is trained on and where is the recurrent component)"
• Definitely need separate section on what a noise ceiling is, variability in how it is estimated and why sometimes you can get models that can go higher. That would clear up a lot of the plot and figure confusion that seemed to emerge
• W2D5 - Mysteries
• Add Claire Sergent to content creators list (Megan Peters sent Slack message about this)
• MP also spotted the videos were not appearing in the book format correctly (known issue)