I am wondering about trying to make a dogs/cats/other classifier. The specific problem is that users (imagine the input is eg. photos from a mobile phone) frequently try to confuse the model, so INTENTIONALLY take a picture of for example cars and it really helps for perception of the quality of the model to be able to filter it. How would I tell the model to detect that, so as to be able to tell the user to ‘stop that,this is not a cat or dog’? Obviously, I cant just assume humans will not be humans and behave. 
3 Likes
We will cover this problem when we look at a multi-label task in the next chapters. (This is a great question btw).
2 Likes
Exploratory data analysis for image datasets, is it still relevant or necessary in deep learning, especially when we are using transfer learning?
3 Likes
How would I tell the model to detect that, so as to be able to tell the user to ‘stop that,this is not a cat or dog’?
What you are describing is the issue of out-of-distribution predictions. In general, neural networks cannot be used for completely different domains from the ones that they are trained in. There exist sophisticated ways to detect if your example image is not from the same distribution as the training set, e.g., using self-supervised learning or other out-of-distribution detection methods.
3 Likes
Is there any pretrained weights available other than the ones from Imagenet that we can use? If yes, when should we use others and when Imagenet? Thanks!
7 Likes
How is the size of the training set determined to be able reasonable accuracy while testing ?
I get skips in between the video. Not sure if the issue is in my end. Is someone else feeling the same…
The pretrained models (in fastai) come from PyTorch’s Model Zoo, and generally if they are pretrained it’s on ImageNet.
2 Likes
You could try introducing a third class neither cat nor dog, and include lots of examples of these in your training set.
Can we merge different pre-trained models?
3 Likes
could you please talk a bit about ways to handle and measure “uncertainty” in deep learning.
3 Likes
The issue with that is you get into a problem of infinite possibilities of what should my third class be. It’s a great idea just it has some weaknesses. The idea with multi-label (I imagine) will be similar to this
Not really. It would be hard to get the best weights from multiple models to work together correctly.
That DL for tabular data is best at high cardinality features is fascinating—will we learn more about this in this course, or where can I find out more?
edit: oh, it’s chapter 9!
1 Like
Will use of any Pytorch pretrained model work?
Is DL also good for regression or only classification
1 Like