Mixed Random Matrix Theory (mixedRMT) : Practice and resources
Introduction
There are many processes that one can apply Random Matrix Theory (RMT), from quantum statistical mechanics to computer networks to deep learning. Primary shortcoming of the existing RMT ensemble approaches that they fix the size of the matrices sampled. This limit the applicability of RMT in many situations, whereby primary components are heterogeneous. This lead to need for building mixed matrix ensembles (MMEs). The analysis of such ensembles is called Mixed RMT.
Formal definition
Due to [suzen21]:
A random matrix ensemble is defined as a mixed ensemble
The degree of the mixture expresses the ratio of different matrix orders in
Some Application areas
- Deep learning: layers produce different size weight matrices.
- Financial instruments: Components that have different-frequency
- Computer networks: Heterogeneous connectivity over different devices.
- Brain and biological networks: Activity in different sub-regions in the brain or biological networks.
How to generate mixed ensemble?
This is probably one of the core issue. One approach would be using Binomial distribution at the given level of degree of mixture [suzen21].
Resources
Currently there are limited works in this direction.
Articles
- [suzen21] Empirical deviations of semicircle law in mixed-matrix ensembles hal-03464130 (2021)
License
This repository and all contributions are licensed under :