Understanding Neural Networks Through Deep Visualization (FC8 Regularized Opt)
Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson
Quick links:
« back | ICML DL Workshop paper |
code |
video
Below are the results of running regularized optimization to produce visualizations for each unit on the fc8 layer of a trained AlexNet model. Click on any figure to show optimization using different hyperparameters (see the paper for more details). You can also download all these figures (and visualizations for conv1-5) as a tar file (449M).
25: European fire salamander
72: black and gold garden spider
89: sulphur-crested cockatoo
98: red-breasted merganser
140: red-backed sandpiper
165: black-and-tan coonhound
179: Staffordshire bullterrier
180: American Staffordshire terrier
188: wire-haired fox terrier
202: soft-coated wheaten terrier
203: West Highland white terrier
205: flat-coated retriever
206: curly-coated retriever
209: Chesapeake Bay retriever
210: German short-haired pointer
218: Welsh springer spaniel
229: Old English sheepdog
233: Bouvier des Flandres
238: Greater Swiss Mountain dog
239: Bernese mountain dog
444: bicycle-built-for-two
548: entertainment center
757: recreational vehicle
986: yellow lady's slipper