Join GitHub today
GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together.Sign up
Add ability to backprop errors to input (for model interrogation) #2866
Placeholder issue. Will add more detailed description later, but TL;DR: we'd like to be able to ask the model what kind of inputs (e.g., image) will maximize the output (e.g., probability that image is cat) and solve for this using gradient descent in the input space (vs. weight space during learning).
It's almost there, probably needs a "no exception on null" for this method, plus needs to be tested.