Repo for writing and running experiments on the effect of network topology on a reservoir computer's ability to learn the Lorentz equations.
Click here for the masters thesis that started this project (p. 32-end will be the most relevant). Click here for a paper about reservoir computing. Click here for a paper about network specialization.
The code in this repository assumes that the rescomp module is installed.
- Dr. Ben Webb
- DJ Passey
- Joseph Jamieson
- Joey Wilkes
- Write code to generate jobs for the super computer
- Write code to compile data from the jobs
- Write code to extract network information from graphs
- Write statistical analysis pipeline
- Generate hyper parameter overview jobs
- Run all hyper parameter overview jobs for all topologies
- Identify the best combination of parameters for each topology
- Run more extensive tests with the top 1% of parameter combinations
- Analyze overview data
- Analyze top preformers data
- Make plots for paper
- What is the correlation between connected components in the original adjacency matrix and size of edges in Wout?
- Spectral design. Building networks to have certain eigenvalues. Can we start from motifs and connect them in a way that produces the desired structure? Do networks have characteristic motifs that play a role in spectrum as edges are removed.
- If we have a network and we uniformly remove edges until we break the network down into motifs, can we build it again by adding the edges back in some way?