lava.lib.dnf.kernels

lava.lib.dnf.kernels.kernels

class lava.lib.dnf.kernels.kernels.GaussianMixin(amp_exc, width_exc, limit=1.0, shape=None, dominant_width=None)

Bases: ABC

Mixin for kernels that are generated with the gauss function.

Parameters

amp_exc (float) – amplitude of the excitatory Gaussian of the kernel
width_exc (list(float)) – widths of the excitatory Gaussian of the kernel
limit (float) – determines the size/shape of the kernel such that the weight matrix will have the size 2*limit*width_exc; defaults to 1
shape (tuple(int), optional) – will return the weight with this explicit shape; if used, the limit argument will have no effect

class lava.lib.dnf.kernels.kernels.Kernel(weights, padding_value=0)

Bases: object

Represents a kernel that can be used in the Convolution operation.

Parameters

weights (numpy.ndarray) – weight matrix of the kernel
padding_value (float, optional) – value that is used to pad the kernel when the Convolution operation uses BorderType.PADDED

property padding_value: float

Returns the padding value

property weights: ndarray

Returns the weights

class lava.lib.dnf.kernels.kernels.MultiPeakKernel(amp_exc, width_exc, amp_inh, width_inh, limit=1.0, shape=None)

“Mexican hat” kernel (local excitation and mid-range inhibition) for a DNF that enables it to create multiple peaks.

Parameters

amp_inh (float) – amplitude of the inhibitory Gaussian of the kernel
width_inh (list(float)) – widths of the inhibitory Gaussian of the kernel

class lava.lib.dnf.kernels.kernels.SelectiveKernel(amp_exc, width_exc, global_inh, limit=1.0, shape=None)

A kernel that enables creating a selective dynamic neural field (local excitation, global inhibition).

Parameters

amp_exc (float) – amplitude of the excitatory Gaussian of the kernel
width_exc (list(float)) – widths of the excitatory Gaussian of the kernel
global_inh (float) – global inhibition of the kernel; must be negative
limit (float) – determines the size/shape of the kernel such that the weight matrix will have the size 2*limit*width_exc; defaults to 1
shape (tuple(int), optional) – will return the weight with this explicit shape; if used, the limit argument will have no effect