Experiment with 'real' dataset

vu-minh / tsne_auto_perp_constraints

Auto select perplexity for tsne with user constraints

0 stars 0 forks source link

Open vu-minh opened 6 years ago

vu-minh commented 6 years ago

Can we use CIFAR-10 pretrained model, take out the last fully-connected layer, use this representation for t-SNE?
If yes, we can compare different representations of some network architectures.

vu-minh commented 6 years ago

CIFAR-10 training set has 50K images. Should find a way to use C++ bh-tsne (in parallel)