Deep Incomplete Multi-view Clustering via Pseudo Label Propagation
TL;DR We solve the incomplete multi-view clustering problem borrowing ideas from semi-supervised learning.
TL;DR We solve the incomplete multi-view clustering problem borrowing ideas from semi-supervised learning.
Digits-10: Handwritten digits from UCI; profile correlations & Fourier coefficients as two views.
COIL-20: Rotating toys/objects; intensity, LBP, Gabor form three views.
USPS-MNIST: 1 000 images/class from USPS & MNIST, 10 k samples, two views.
※ 50 % missing features per view to simulate incomplete multi-view learning.
PLP-L: Local pseudo-label propagation | PLP-G: Global pseudo-label propagation | Concat: view concatenation
