https://github.com/pointlander/neural_rank
non-linear page rank
https://github.com/pointlander/neural_rank
machine-learning
Last synced: 15 days ago
JSON representation
non-linear page rank
- Host: GitHub
- URL: https://github.com/pointlander/neural_rank
- Owner: pointlander
- License: bsd-3-clause
- Created: 2020-11-23T23:24:32.000Z (about 5 years ago)
- Default Branch: main
- Last Pushed: 2021-10-28T03:18:17.000Z (about 4 years ago)
- Last Synced: 2024-06-20T21:07:22.076Z (over 1 year ago)
- Topics: machine-learning
- Language: Go
- Homepage:
- Size: 37.1 KB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Neural Rank
This is an implementation of non-linear [page rank](https://en.wikipedia.org/wiki/PageRank). Page rank can be defined by the equation `AX = X` where `A` is an [adjacency matrix](https://en.wikipedia.org/wiki/Adjacency_matrix), and `X` is the vector of the page ranks. This equation can be altered to be non-linear with a non-linear function `f`: `f(Af(AX)) = X`. This is a two layer [recurrent](https://en.wikipedia.org/wiki/Recurrent_neural_network) [auto encoder](https://en.wikipedia.org/wiki/Autoencoder) which can be solved for `X` using [gradient descent](https://en.wikipedia.org/wiki/Gradient_descent). It is recurrent because the `A` weights are reused in the second layer. The implementation of this algorithm tends to produce results similar to page rank. A complex number version is also provided and tends to produce results less like page rank.