Dog Rescue Robot

Let’s say you’ve built a robot that searches for lost dogs. When your robot spots a dog, it checks if the breed of the dog it’s looking at matches the breed of a dog that has recently been reported as missing.

Being the good data scientist you are, you made the robot able to tell breeds of dogs apart by creating a convolutional neural network that classifies images and videos of different dog breeds.

There is one problem, your robot has a hard time telling the difference between pugs and pit bulls because the data set you trained it on had these two breeds mislabeled as the other 50% of the time. The problem is even worse if the robot is looking at the dog in less than ideal conditions (rain, fog, from far away, from behind, etc.).

How would you modify your neutral net to account for this fact?