It has been shown that subunits from retinal ganglion cell receptive fields and other visual neurons in the cortex can be retrieved using clustering methods (e.g. semiNMF (Liu, et al. 2017) and soft clustering (Shah et al. 2020)).
NMF can be easily combined with splines and produce smooth factors directly using multiplicative updates ( see: Zdunek, et al. 2014) , and k-means clustering can also be augmented with splines by simply replacing the original means with spline-approximated means.
- k-means clustering
- NMF (Lee & Seung, 2001)
- SemiNMF (Ding et al, 2010)
and their corresponding spline-based versions.
K-means clustering with spline-based cluster centroids:
from rfest.MF import KMeans
kms = KMeans(V, k, build_S=True, dims=[20, 20], df=11)
kms.fit(num_iters=100, verbose=10)NMF with spline-based left factor:
from rfest.MF import NMF
nmf = NMF(V, k, build_L=True, dims_L=[5, 20, 15], df_L=7)
nmf.fit(num_iters=100, verbose=10)If spline-based right factor is prefered, then use flags build_R, dims_R and df_R.