convolution_aggregate

pmrf.math.aggregations.convolution_aggregate(x: Array, epsilon: float = 1e-12) → Array

Generalized convolutional aggregation over flattened features.

Parameters:

• x (jnp.ndarray) – Input array of features/losses.

• epsilon (float, default=1e-12) – Small constant to prevent numerical instability.

Returns:

A scalar JAX array representing the aggregated value.

Return type:

jnp.ndarray