Another definition question : is a latent feature one of the feature of the final feature map ?
1 Like
I am thinking of the inputs to the first layer – the training data itself. Applying dropout to the inputs shrinks the size of the training set, which allows the model to train faster.
Some of these face recognition police technologies look a bit like taken from some cyberpunk novel 
1 Like
The question is very hard when it comes to ‘races’ that could come in different colors. Detecting race is very tricky, I even wonder if it’s even possible to build a good model.
For a visual overview of the forward pass in a ConvNet, I built these spreadsheets and they also have a nice little heat map with conditional formatting 
Supporting blog post here.
12 Likes
How are the values of the convolutional kernels usually selected?
This is answered above, kernels values are weights and are trained with backpropagation.
2 Likes
Not always. Some kernels can be “designed” using fixed values, like edge detectors.
I really love Jeremy and Rachel are talking about this topic. Very impactful and a good thing to remember for us learning this.
13 Likes
That wouldn’t shrink the inputs. That would just randomly feed 0s in place of real data.
You can fix kernel values and then freeze those layers but we do not do that usually since networks find features better than humans.
Watchbird by Robert Sheckley (short story) imagines how machines trained to prevent crime define crime differently that we do.
Science fiction predicting problematic technology.
3 Likes
One approach people take is to create another output that tries to classify based on a variable you don’t want to have bias on (say race or gender), and then uses a loss function that combines doing well on the desired classification task with doing poorly on classification of race or gender.
3 Likes
The model is independent of sample size. Building a model with a small subset of samples is faster than using all the samples.
FB is an out-of-control beast
Has any of this been tested in court? It seems like something that the companies should be held accountable for.
1 Like
Some of us trying to solve this problem.
It might make Rachel’s life a bit easier and help the most upvoted questions get answered.
Just got this notification of discussion on Hacker News. Quite interesting discussion on tacking such issues with your employers.
Which you would do by using a subset of the data in your dataloader. Applying dropout to the inputs would have a very different effect. It would feed in the same amount of data, just randomly zeroing certain inputs.
I did that once today but I didn’t know. Thanks for restating this I must have missed it in former lesson threads.
1 Like