Applied Electives
Apply the ladder to real domains. Electives are independent of each other; each lists its own prerequisites.
You can now…
- Benchmark decoding pipelines without leakage
- Stage sleep and analyze overnight spectra
- Build a real-time pipeline and account for every millisecond in its latency budget
- State what a research analyst may not claim about clinical EEG, and what a simultaneous recording buys and costs
- Compare device classes neutrally and decide what survives at low channel counts
- Estimate the reliability of a resting-state feature before believing it is a biomarker
- Run a reproducible pipeline at scale, with provenance a stranger can act on
Level 7 is not another rung. The first seven levels are a ladder, each one assuming the one below; this level is a fan: eight independent electives, plus one ordered sequence. The eight electives (L7.1–L7.8) carry their own prerequisites, drawn from wherever in the ladder they need them — sleep from time-frequency, real-time from filtering, biomarkers from oscillations and effect sizes, scale from pipelines and reporting — so you can take one, or three, or all eight, in whatever order your work requires. The sequence (L7.10–L7.16, with L7.9 reserved) is different: it builds a neurofeedback or brain–computer-interface application on a four-channel consumer headband, each lesson on the one before, and the consumer-headset path walks it in order after the ladder’s own prerequisites.
What you bring to them is the thing the ladder was for. You can read a raw trace and name what is on it; you can turn a recording into a spectrum and say which parts of it are rhythms; you can build a preprocessing pipeline whose every decision is written down; you can measure an ERP without letting the measurement choose itself; you can tell a connectivity result from a volume-conduction artifact; and you can state what a p-value licenses and what it does not. Every elective here assumes all of that and adds one domain’s own constraints on top — because in each of them the constraint, rather than the analysis, is what decides the answer. A BCI is judged by cross-subject transfer; a real-time loop by a latency budget you can decompose; a consumer headset by a hardware bandwidth that has nothing to do with its sampling rate; a biomarker by a reliability that bounds every effect size you will ever report from it.
Three of these lessons ship with less than was planned, and in each case the reason is the same policy and it is worth reading before you start. L7.4 has no figure of an epileptiform discharge, because no dataset here whose licence permits shipping a snippet contains a scored one, and drawing a plausible spike would fabricate the clinical finding the lesson warns you not to over-read. L7.7 has no widget, no notebook and no figure at all: this site holds no simultaneous EEG–fMRI, EEG–TMS or EEG–eye-tracking recording, so the gradient artifact, the ballistocardiogram and the TMS pulse are described rather than shown. L7.5 lost a dataset outright — ds-mpeng turned out to be CC BY-NC 4.0 rather than the CC BY 4.0 this site’s catalog had recorded, downstream of the consent its participants gave, so it is named, described and linked, and nothing here derives from it. None of the three is a gap in the teaching. They are what it looks like when a site refuses to show you something it cannot honestly show, and the reasoning is in each lesson rather than in a footnote.
The capstone is where the ladder ends: reproduce a published result end to end, from raw, with the plan committed before the analysis and a repository anyone can run. It is the only capstone that asks you to publish rather than to write, and the only one whose rubric is about a stranger’s machine.
Three pages exist for what comes after. /datasets is the directory as a working instrument — every entry with its licence, access classes, acquisition tables and documented caveats, which is where an analysis plan should start rather than end. /cite is how to cite this site and, more usefully, how to cite every dataset it names, with the DOI and the reference text already assembled. /changelog records what changed here and when, which matters because several of the facts in these lessons — a licence, an access class, a field nobody has been able to confirm yet — are the current state of a record rather than a permanent truth, and the honest thing a teaching site can offer is a date.
Lessons
-
L7.1 Decoding and BCI
Benchmark CSP + LDA, Riemannian classifiers and EEGNet with moabb, within- and cross-subject, without leakage.
-
L7.2 Sleep EEG and staging
Sleep stages and graphoelements, a hypnogram with yasa, and spectra across the night.
-
L7.3 Real-time processing and neurofeedback
Streaming with LSL, ring buffers and causal filters, latency budgets, online artifact handling, feedback design, and what neurofeedback evidence supports.
-
L7.4 Clinical EEG primer (reading module)
What clinical review looks like, the appearance of epileptiform discharges in published examples, why spike detection is hard, and what a research analyst must not claim.
-
L7.5 Mobile and consumer EEG
Dry and low-channel-count systems, motion artifacts, which analyses survive with 4–8 channels, and validation against a research system — comparative and product-neutral.
-
L7.6 Resting-state and biomarkers
Resting-state features, test–retest reliability, and the biomarker replication problem, with the confounds of real clinical archives.
-
L7.7 Simultaneous and multimodal EEG (overview)
EEG–fMRI, EEG–TMS and EEG–eye-tracking setups, their artifacts, and the corrections that exist.
-
L7.8 Reproducible pipelines at scale
mne-bids-pipeline on a BIDS dataset, containers, batch execution, provenance tracking, and a reproducibility statement.
-
L7.10 What a four-channel headband measures
A four-electrode headband described from documented facts — which four sites, what rate, what bandwidth, what reference, what the device filtered before you saw the stream — and the analyses that follow from them, for the two geometries the consumer path names.
-
L7.11 Getting a stream: timestamps, dropped samples, drift, recording
A recording is a file and a stream is a promise — samples at a stated rate, in order, complete, on a clock you can align to — and every part of the promise can fail quietly. One interface for a replayed file, a synthetic stream with planted faults and a headset; the block contract; the rate measured against the rate claimed; drops counted and verified; drift in milliseconds; a session recorded in the site's own format.
-
L7.12 Signal quality you can act on live
Offline you inspect what a cleaner removed and run again; online you get one pass, so the honest move is to detect a bad stretch and suspend the feedback. A per-channel quality gate on the raw block, three states with a hold, blinks on a frontal montage and not on an occipital one, the gate scored against the offline artifact timeline on the same recording, and a consumer export's own quality columns — shown, never used as truth.
-
L7.13 The feedback signal: band power in a buffer
"Alpha at Pz" is not a definition. A feedback signal is a band, a set of channels and how they are combined, a reference, a filter and its length, a derivation and a smoothing — each either exact in time or measured. Three ways to a band-power feature per block and what each costs; smoothing quantified as measured latency with the loopback probe; a baseline from calibration and what an adaptive threshold does to a learning claim; the channels of a montage combined with the combination stated.
-
L7.14 A protocol that can fail: sham, blinding, pre-registration, the valid test
The block-level trend test is anti-conservative and its false-positive rate is measured — eight of twelve replayed sessions "significant" where learning is impossible. The four sham conditions and what each does and does not control; a valid test with the session as the unit, null where it should be and powerful where a rise is planted, and why even a valid within-condition test cannot separate learning from the passage of a minute; a pre-registration hashed before the data; a session log with a sealed sham mode; the CRED-nf items one person can satisfy alone and the ones they cannot.
-
L7.15 A first BCI on four channels
Decoding re-taught for four channels — what is decodable at rank four, the chance band an accuracy has to clear, the leakage a split can smuggle in — and a calibrate-then-apply classifier for eyes closed against eyes open on the occipitotemporal replica with a time split, the decoder frozen into arrays; the P300 at 16, 8 and 4 channels on one subject; a steady-state response detected by canonical correlation on a planted flicker; a blink switch, labelled EOG; and the decoder run block by block with the same features as offline.
-
L7.16 Build the application
The path's capstone-shaped lesson. An application is source → gate → feature or decoder → loop → presenter, with the library owning the first four and the protocol file naming every decision in them; it is packaged as a pip project that depends on the library with a replay entry point, so it runs and its CI runs where no headset is; its budget carries a measured row and the loopback probe is within one block of the arithmetic; its gate suspends feedback across the planted artifact windows; its four sham modes are sealed into their logs; its pre-registration is a hash a commit precedes; its learning outcome is the valid test and its report is the six CRED-nf items; and a run on a headset is documented in a record with every value measured, and no efficacy claim.