Skip to content

Scale free topology #387

Open
Open

Description

@MartinuzziFrancesco

Paper

https://doi.org/10.1063/1.4746765

Topology

From the paper:

Step 1: Growth

Start with an initially randomly connected network containing (m_0) neural units.
At every time step, introduce a new unit and connect it to (m) already-existing neural units, with (m \le m_0).
When the number of units reaches the total (N), the growth process stops.

Step 2: Preferential attachment

The probability (\Pi_i) that a new neural unit will be attached to unit (i) (one of the (m) already-existing units) depends on the degree (k_i) of unit (i), as follows:

$$\Pi_i = \frac{k_i}{\sum_{j \in \{1,2,\ldots,m\}} k_j}, \qquad i = m_0 + 1, \ldots, N.$$

Official implementation

none

Activity

  1. aviralgarg05 commented on Nov 13, 2026

    @aviralgarg05
    Contributor

    Hi, I’m interested in working on this if it’s still available. I’ve reviewed the current initializer conventions and the paper linked in the issue. For a first contribution, would you prefer an undirected Barabási–Albert topology with sampled weights and the standard radius/sparse-output support, or should a directed/weighted variant be included from the beginning?

  2. MartinuzziFrancesco commented on Nov 19, 2026

    @MartinuzziFrancesco
    CollaboratorAuthor

    if you want you can start with the first and add the second with a different pr

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    initsinitializers for internal matrices

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions