Open Internet by MindsNet

Interpretation of Latent Feature Maps in Autoencoders

The author is struggling to accurately interpret which input images are captured in the top-scoring latent feature map of a convolutional autoencoder. Despite trying methods like encoding one image at a time and decoding only the top-scoring latent feature map, the author still encounters false positives, likely due to decoder entanglement.

Computing & Technology, Computer Science, Machine Learning