Slides

[stemlock]

1. General: More of a subjective opinion, but maybe reduce the time/number of slides spent on background/history of ANNs. As a workshop, I would personally hope to see more applied material, vs a lecture which might focus more on the theory and context. There are plenty of other open source lecture material that covers the latter in depth, so maybe what would be more beneficial would be the live demo portion (increasing the amount of interaction with the workshop attendees).
2. General: Similar to suggestion 1., but the workshop is very slide heavy. May be worth cutting down the material (feels morel like a lecture)
3. Slide 24: No mention of exploding/vanishing gradients during activation functions. May be worthwhile mentioning that variations of ReLU is currently the standard to prevent this.
4. Slide 37: Sample is labeled "Sample 1", whereas in the previous slide it is "Sample 2"
5. Slide 56: No mention of the chain-rule for multi-layer backpropagation and how error depends on derivatives at earlier layers as well. Might be worth mentioning this concept (something like this https://www.untitledkingdom.com/blog/let-me-introduce-you-to-neural-networks-feaa7aa2afe2)