spd_learn.functional.spd_graph_aggregate#

spd_learn.functional.spd_graph_aggregate(input: Tensor, propagation: Tensor) → Tensor[source]#

Aggregate SPD graph nodes with nonnegative propagation weights.

Parameters:

  • input (torch.Tensor) – SPD nodes of shape (…, nodes, channels, channels).

  • propagation (torch.Tensor) – Row-stochastic weights of shape (…, nodes, nodes), matching dtype/device.

Returns:

Aggregated nodes with broadcast leading axes and unchanged matrix dimensions.

Return type:

torch.Tensor

Notes

Aggregation is independent across samples. PSD inputs need only produce PSD outputs.