Middle school
BCI101: An Applied Introduction to Noninvasive Brain-Computer Interfaces
Neurons, to bits, to real-world outcomes.
Record your own brain's electricity, see what an AI model can and can't tell from it, teach a model two of your own thoughts, and use it to switch a light.
Contents
Part 1
Neurons
Your brain's electricity, and how to measure it.
Chapter 1The brain's electricityHow your brain makes electricity, and how a device on your head can pick it up.Chapter 2Waves and rhythmsHow many times a second a wave repeats, how big it is, and the rhythms in your body and brain.Lab 1Alpha FlowerMeasure what your brain does when you close your eyes, and grow a fractal from it.
Chapter 3Signal and noiseBlinks and clenched jaws make signals too. How to tell them from your brain's.Lab 2BCI Flight SimulatorSteer a ship by looking at flickering lights, read from the back of your brain.
Chapter 3Signal and noiseBlinks and clenched jaws make signals too. How to tell them from your brain's.Lab 2BCI Flight SimulatorSteer a ship by looking at flickering lights, read from the back of your brain.Part 2
Bits
Computers that learn from brain signals.
Chapter 4Your brain dataWhat brain data is, who may see it, and how to say yes or no.Chapter 5Models that learnHow a computer learns from examples, and why it must be tested on new ones.Chapter 6Chance and fair testsHow often guessing gets it right, and what makes a test fair.Lab 3Age DetectorAn AI model on the IntoMind One estimates your age from your EEG. How close does it get?
Part 3
Real-world outcomes
From a thought to a light on a board.
Lab 4Mind-Controlled Light SwitchTeach a model two of your own thoughts, test it fairly, and switch a light with it.
Chapter 7Microcontrollers and circuitsWhat a microcontroller is, and how to light an LED from one.Lab 5Mind-Controlled MicrocontrollerYour custom AI head switches a light on a small board, with no computer in the room.
Chapter 8Designing a BCIPlan a brain-computer interface for a real need, and a fair test for it.Chapter 9The capstoneBuild your own two-state brain-computer interface, test it, and show it.
Chapter 7Microcontrollers and circuitsWhat a microcontroller is, and how to light an LED from one.Lab 5Mind-Controlled MicrocontrollerYour custom AI head switches a light on a small board, with no computer in the room.Chapter 8Designing a BCIPlan a brain-computer interface for a real need, and a fair test for it.Chapter 9The capstoneBuild your own two-state brain-computer interface, test it, and show it.High school
BCI101: An Applied Introduction to Noninvasive Brain-Computer Interfaces
Neurons, to bits, to real-world outcomes.
Record your own brain's electricity, find out what an AI model on the device can and can't tell from it, teach a model two of your own mental states, test it fairly, and use it to switch a pin on a board.
Contents
Part 1
Neurons
How the brain makes a signal you can measure from the scalp.
Chapter 1The brain's electricityFrom one neuron's pulse to a voltage on the scalp: why EEG sees large groups of neurons in step, and what it can't see.Chapter 2Waves and rhythmsFrequency, amplitude and sampling, and how a spectrum takes a signal apart.Lab 1Alpha FlowerMeasure alpha as a share of your EEG, and find out what else moves the number, including where the device sits.Chapter 3Signal and noiseArtifacts from eyes and muscles, and how to tell a real difference from the spread.Lab 2BCI Flight SimulatorSteer by looking at flickering lights: how the visual cortex follows a flicker, and how a decoder reads it.
Chapter 3Signal and noiseArtifacts from eyes and muscles, and how to tell a real difference from the spread.Lab 2BCI Flight SimulatorSteer by looking at flickering lights: how the visual cortex follows a flicker, and how a decoder reads it.Part 2
Bits
Models that learn from brain signals, and how to test them fairly.
Chapter 4Your brain dataNeural data, consent, the law so far, and who should be allowed to see it.Chapter 5Models that learnHow a model fits a rule to labeled examples, and why it must be tested on examples it never saw.Chapter 6Chance and fair testsHow often guessing scores well, and how to test a model fairly.Chapter 7Encoders and headsHow the IntoMind One's encoder sums up four seconds of EEG as 76 numbers, and how small heads read them.Lab 3Age DetectorAn AI model estimates your age from your EEG. Measure its error, and compare it with a simpler guess.
Part 3
Real-world outcomes
From a decision on the device to a pin on a board.
Lab 4Mind-Controlled Light SwitchChoose two mental states and a place on the head, teach a model to tell them apart, and test it fairly.Chapter 8Microcontrollers and circuitsPins, current and resistors, and a program that drives an LED from a microcontroller.Lab 5Mind-Controlled MicrocontrollerYour custom AI head switches a board's pin, with no computer in the loop.Chapter 9Designing a BCIPlan a brain-computer interface for a real need, with requirements and a test written first.Chapter 10The capstoneBuild your own two-state brain-computer interface, test it against the test you wrote first, and present it.
Chapter 8Microcontrollers and circuitsPins, current and resistors, and a program that drives an LED from a microcontroller.Lab 5Mind-Controlled MicrocontrollerYour custom AI head switches a board's pin, with no computer in the loop.Chapter 9Designing a BCIPlan a brain-computer interface for a real need, with requirements and a test written first.Chapter 10The capstoneBuild your own two-state brain-computer interface, test it against the test you wrote first, and present it.College
BCI101: An Applied Introduction to Noninvasive Brain-Computer Interfaces
Neurons, to bits, to real-world outcomes.
Record your own brain's electricity and analyze it in your own code, test what an AI model on the device can and can't tell from it, teach a model two of your own mental states and test it against chance, then close the loop on a board you program.
Contents
Part 1
Neurons
The signal from neurons to the scalp, and the spectrum that takes it apart.
Chapter 1The brain's electricityFrom synaptic current to scalp potential: dipoles, volume conduction, the reference, and what four channels can't see.Chapter 2Waves and rhythmsSampling and aliasing, the discrete Fourier transform, leakage and windows, and Welch's estimate of band power.Lab 1Alpha FlowerAlpha's share by Welch's method, eyes open against closed at Oz and Fpz, tested by permuting the blocks.Chapter 3Signal and noiseSignal-to-noise ratio, the spectral signatures of blinks, muscle, movement and line noise, and when to avoid, reject or correct them.Lab 2BCI Flight SimulatorFrequency tagging, the demo's score rebuilt in code, and the information rate a faster decision buys.
Chapter 3Signal and noiseSignal-to-noise ratio, the spectral signatures of blinks, muscle, movement and line noise, and when to avoid, reject or correct them.Lab 2BCI Flight SimulatorFrequency tagging, the demo's score rebuilt in code, and the information rate a faster decision buys.Part 2
Bits
Models that learn from brain signals, evaluated against chance, baselines and their own uncertainty.
Chapter 4Your brain dataConsent and data minimization, what models infer and identify from EEG, the law so far, research ethics and neurorights.Chapter 5Models that learnLeast squares and its normal equations, overfitting when features near the examples, and ridge regression with λ chosen without the test.Chapter 6Chance and fair testsChance levels and p-values, readings that stick together, leakage, permutation tests over rounds, and many tests at once.Chapter 7Encoders and headsThe encoder on the chip, its 76-number embedding, heads rounded to bytes, and the published evaluation read as a specification.Chapter 8Evaluating modelsA model's error means nothing without its baseline, its interval and its test data.Lab 3Age DetectorTest the head's any-site claim on one person by permuting blocks of windows, and its error against the room's most common age.Lab 4Biological Sex DetectorWhat a score means when two groups overlap far more than they differ.
Lab 4Biological Sex DetectorWhat a score means when two groups overlap far more than they differ.Part 3
Real-world outcomes
A closed loop from a decision on the device to a pin, and a study of your own.
Lab 5Mind-Controlled Light SwitchTwo mental states and a site, a ridge head scored on held-out rounds, a permutation test over rounds, and the mirror site.Chapter 9Microcontrollers and circuitsA pin and Ohm's law, Bluetooth Low Energy from advertising to indications, pairing and its threats, and a loop's latency.Lab 6Mind-Controlled MicrocontrollerA MicroPython program closes the loop on a board's pin, and ten timed calls test a model of the delay.Chapter 10Designing a BCIUser-centered design, requirements a test can check, and a threshold chosen by what each mistake costs.Chapter 11Research methodsHypotheses, power, preregistration, and reporting that someone else can repeat.Chapter 12The capstoneFrom a user's need to a working board: requirements, a preregistered test, two sessions, a reproducible analysis, and the loop measured.
Chapter 9Microcontrollers and circuitsA pin and Ohm's law, Bluetooth Low Energy from advertising to indications, pairing and its threats, and a loop's latency.Lab 6Mind-Controlled MicrocontrollerA MicroPython program closes the loop on a board's pin, and ten timed calls test a model of the delay.Chapter 10Designing a BCIUser-centered design, requirements a test can check, and a threshold chosen by what each mistake costs.Chapter 11Research methodsHypotheses, power, preregistration, and reporting that someone else can repeat.Chapter 12The capstoneFrom a user's need to a working board: requirements, a preregistered test, two sessions, a reproducible analysis, and the loop measured.Graduate
BCI101: An Applied Introduction to Noninvasive Brain-Computer Interfaces
Neurons, to bits, to real-world outcomes.
Take six working brain-computer interfaces apart on your own EEG: what each computes, with which parameters, what it leaves out, and how it holds up against the literature and a fair test. Then preregister and run a study of your own.
Contents
Part 1
Neurons
Generators and the forward model, and spectral estimates with their bias and variance.
Chapter 1The brain's electricityGenerators, the forward model, the reference as a linear map, and why four channels close together can't locate a source.Chapter 2Waves and rhythmsThe periodogram's bias and variance, Welch and multitaper estimates, the aperiodic background, and individual alpha frequency.Lab 1Alpha FlowerIndividual alpha frequency two ways, reactivity, and alpha above the background, with their stability within a person and between sites.Chapter 3Signal and noiseThe noise floor and interference, artifacts with their signatures, what filters cost, and what four channels can't separate.Lab 2BCI Flight SimulatorThe demo's decoder rebuilt, canonical correlation and bits a minute against the window, and the half-field test at O1 and O2.
Chapter 3Signal and noiseThe noise floor and interference, artifacts with their signatures, what filters cost, and what four channels can't separate.Lab 2BCI Flight SimulatorThe demo's decoder rebuilt, canonical correlation and bits a minute against the window, and the half-field test at O1 and O2.Part 2
Bits
Linear decoders on a pretrained encoder, tested against dependence, baselines and where the data came from.
Chapter 4Your brain dataResearch ethics and consent, what recordings, embeddings and heads reveal, the law so far, and neurorights and their critics.Chapter 5Models that learnDiscriminants and least squares, ridge and shrinkage, choosing the penalty by nested cross-validation, and filters against patterns.Chapter 6Chance and fair testsChance levels and error bars, dependent windows and leakage, permutation tests that respect the design, nulls below one half, and forking paths.Chapter 7Encoders and headsSelf-supervised pretraining and distillation, the encoder on the device, a head's integer arithmetic, probes, and how many windows a head needs.Chapter 8Evaluating modelsBaselines and the unit of inference, regression to the mean in brain age, calibration and base rates, exposure, and reporting.Lab 3Age DetectorThe age head's bias toward its training mean and its scatter in a room of similar age, two sites compared with an equivalence test.Lab 4Biological Sex DetectorWhat an AUC near 0.6 implies and where averaging stops, confounds simulated and adjusted, and a design that could tell brain from contact.
Lab 4Biological Sex DetectorWhat an AUC near 0.6 implies and where averaging stops, confounds simulated and adjusted, and a design that could tell brain from contact.Part 3
Real-world outcomes
A closed loop's latency and attack surface, requirements set by users, and a preregistered study.
Lab 5Mind-Controlled Light SwitchTwo sessions on two days: each head tested against its relabelings with two penalties, the transfer raw and recentered, and probes for what it learned.Chapter 9Microcontrollers and circuitsThe board's program and its fail-safes, delay as a distribution, false switches per hour, and a threat model for the loop.Lab 6Mind-Controlled MicrocontrollerThe loop timed on the board's own clock against the window model, false switches at rest with their bounds, and the board's prefix tested.Chapter 10Designing a BCIUsers and testable requirements, metrics that mean something, the operating point by expected cost, and the no-control state.Chapter 11Research methodsEstimands and designs, power at two levels and exaggerated effects, preregistration and registered reports, and reporting others can reproduce.Chapter 12The capstoneA question of your own as a registered report: reviewed before the data, tested as planned, and reproduced by another group.
Chapter 9Microcontrollers and circuitsThe board's program and its fail-safes, delay as a distribution, false switches per hour, and a threat model for the loop.Lab 6Mind-Controlled MicrocontrollerThe loop timed on the board's own clock against the window model, false switches at rest with their bounds, and the board's prefix tested.Chapter 10Designing a BCIUsers and testable requirements, metrics that mean something, the operating point by expected cost, and the no-control state.Chapter 11Research methodsEstimands and designs, power at two levels and exaggerated effects, preregistration and registered reports, and reporting others can reproduce.Chapter 12The capstoneA question of your own as a registered report: reviewed before the data, tested as planned, and reproduced by another group.