I just wanted to ask what other datasets should we apply the things learned from the first lesson for homework. It would help a lot to apply the learned skills to new datasets.
Any dataset that you think would be interesting! You might be surprised by the results you get 
1 Like
I’m going to just list a few ideas off the top of my head. Some may be less serious, or even on the funny side, and I cannot guarantee that you will actually find a useful dataset for them. Although, you could even build your own, which could also lead to some plus knowledge.
You could try out multiclass classification on eg. money (different coins, banknotes what is their value), pokemon characters (there is a whole lotta bunch!), types of scenery (mountains, fields, city, or even sunset, sunrise, this could also lead to some multilabel classification where one image can have multiple labels), famous monuments/sculptures, famous painters’ works (as in who painted this or that), emotion based on spectrograms of speech (this is some interesting stuff, unfortunately datasets are quite lacking, I worked on this like a year and a half ago at uni), emotions based on facial expressions.
You could also experiments with the size of the datasets. How small you can make it (what is the minimum number of examples per class) while still getting reasonably good predictions (without overfitting)? Which classes suffer the most from reducing the number of examples? What happens if you throw in a few wrongly labelled examples into the classes (like adding some noise to the input)? Does this affect the learning time/results etc.? How? Why?
You don’t necessarily have to reach state-of-the-art results while learning, so you can try out crazy datasets and crazy ideas. I would suggest experimenting around with different ideas, asking questions to yourself and then trying them out. Try to draw conclusions, learn from your experiments.
Play around.
Hopefully my answer was of help, happy coding!
3 Likes