search

guruucsd / DifferentialEncoding

Differential Encoding model
1 stars 3 forks source link

Fit the distribution of connections and weights, plot histogram of orientations. #76

Open bcipolli opened 8 years ago

bcipolli commented 8 years ago

Connections can be fit as an oriented Gaussian distribution. We should fit the best Gaussian for each hidden unit, then show the distribution of Gaussians (orientation) to validate that our connectivity procedure samples all directions equally.

Similarly, looks like weights might be Gabor-like (off-center, on-surround). We should fit gabors and plot (a) a distribution of goodness-of-fits, (b) a distribution of frequencies, and (c) a distribution of orientations.

bcipolli commented 8 years ago

Looks like @kendrickkay has some utilities to do the Gabor fits. If those work for us, we can add his knkutils repo as a subtree in our lib directory.