JitFPHomoLinear#
- class brainpy.dnn.JitFPHomoLinear(num_in, num_out, prob, weight, seed=None, sharding=None, mode=None, name=None, transpose=False, atomic=False)[source]#
Synaptic matrix multiplication with the just-in-time connectivity.
It performs the computation of:
\[y = x @ M\]where \(y\) is the postsynaptic value, \(x\) the presynaptic variable, \(M\) the synaptic weights which has the fixed sparse connectivity and weights. Particularly, the connectivity in \(M\) is sampled from a fixed probability \(prob\), and at each connection, the synaptic value is the same \(weight\).
- Parameters:
num_in (
int) – int. The number of the input feature. A positive integer.num_out (
int) – int. The number of the input feature. A positive integer.prob (
float) – float. The connectivity probability.weight (
float) – float. The synaptic value at each position.seed (
Optional[int]) – int. The random seed used to keep the reproducibility of the connectivity.transpose (
bool) – bool. Transpose the JIT matrix or not. Default False.atomic (
bool) – bool. Compute the post-synaptic value with the atomic summation. Default False. May be changed in the future.sharding (
Union[Sequence[str],Device,Sharding,None]) – The sharding strategy.