Adding state-of-the-art Pipelines

[Div12345]

Adding this issue for a thread for following up on the discussion in the office hours to add more of the SOTA pipelines.

[jsosulski]

I will add popular ERP classification pipelines I found in the literature.

Maybe we could add some landing page / README.md section title "I want my own classifier to be included in benchmarks, what do I do?" where we explain to create a PR with the corresponding pipelines file and adding dependencies to "external" extras.

[sylvchev]

Good idea! What are the ERP pipeline that you have in mind @jsosulski ? For SSVEP, I'm thinking about CCA, TRCA, MDM and TS-SVM. I'll try to find some reference papers to clearly describe the pipeline. This is really helpful to have links to the papers when you are publishing a benchmarking results.

[jsosulski]

I am of course biased by my own research, but according to Lotte's 2018 Review I would propose:

• SLDA in sensor space on averaged time bins (problem: time bins influence results)
• XDAWN + Riemmanian Logistic Regression (problem: number of XDAWN components / only target or both target and non-target hyperparameters influence results)
• XDAWN + SLDA
• Novel LDA method: e.g. ToeplitzLDA

Interesting:

• DCPM: However, no implementation available
• HDCPA: No implementation available
• EEGNET/ShallowNet: Computational load / moabb integration? We wanted to skip deep architectures for now IIRC