ME736: Brain-Inspired Neural Networks, BCI Applications, Model Evaluation, and Sleep Neurophysiology

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/47

flashcard set

Earn XP

Description and Tags

Vocabulary practice flashcards covering brain-inspired neural networks, spiking neural networks, BCI decoding, classifier comparison statistics, model cross-validation, polysomnography, sleep staging, and sleep apnea physiology.

48 Terms

1
New cards

Dendrite

A branched projection of a neuron that functions like a receiving antenna, collecting input signals from other cells and conducting them toward the cell body.

2
New cards

Axon

A long, slender nerve fiber that functions analogous to a copper wire, carrying electrical impulses away from the cell body toward target synapses.

3
New cards

Action Potential (Spike)

A discrete, all-or-nothing electrical impulse lasting approximately 1 ms1\,\text{ms} that serves as the primary mechanism of communication between biological neurons.

4
New cards

Refractory Period

A phase lasting approximately 2 ms2\,\text{ms} immediately following an action potential during which a biological neuron cannot fire another spike regardless of input intensity.

5
New cards

AlexNet

A landmark convolutional neural network architecture introduced in 2012 by Krizhevsky, Sutskever, and Hinton that significantly outperformed state-of-the-art methods on ImageNet, initiating the modern deep learning era.

6
New cards

Local Receptive Field

A key mechanism in convolutional neural networks where each artificial neuron connects only to a small spatial region of the input, mirroring biological V1 visual cortex neurons.

7
New cards

Weight Sharing

The feature in convolutional neural networks where identical filter kernels are applied across every spatial location of the input field, mimicking uniform edge detection across the visual field.

8
New cards

Spiking Neural Network (SNN)

A class of neural networks that compute asynchronously using discrete, event-driven spike trains based on internal membrane potential dynamics, offering 10× to 10000×10\times\text{ to }10000\times greater energy efficiency compared to traditional artificial neural networks.

<p>A class of neural networks that compute asynchronously using discrete, event-driven spike trains based on internal membrane potential dynamics, offering $$10\times\text{ to }10000\times$$ greater energy efficiency compared to traditional artificial neural networks.</p>
9
New cards

Motor Imagery (MI)

The mental rehearsal of a motor act without overt physical movement, which reliably produces a decrease in contralateral sensorimotor alpha rhythm (8–12 Hz8\text{--}12\,\text{Hz}, or mu rhythm) and beta rhythm (13–30 Hz13\text{--}30\,\text{Hz}).

10
New cards

Mu Rhythm

An electroencephalographic oscillation in the 8–12 Hz8\text{--}12\,\text{Hz} frequency band recorded over the sensorimotor cortex that attenuates during real or imagined motor movement.

11
New cards

Information Transfer Rate (ITR in MI BCI)

The operational communication speed of a brain-computer interface system; for non-invasive motor imagery EEG, it is typically limited to 3–53\text{--}5 correct actions per minute, or approximately 10–50 bits/min10\text{--}50\,\text{bits/min}.

12
New cards

Continuous Authentication

A biometric security protocol where a system repeatedly verifies user identity in the background using physiological signals like electroencephalography to maintain access without periodic re-login prompts.

13
New cards

Overfitting

A modeling flaw where a classifier or regression algorithm becomes excessively complex and fits the idiosyncratic noise of the training dataset rather than the underlying generative relationship, resulting in poor generalization to unseen data.

14
New cards

Hold-Out Sample

A subset of observations withheld from model training and estimation so that predictive accuracy and generalizability can be evaluated on entirely unseen data.

15
New cards

kk-Fold Cross-Validation

A resampling technique where available data is partitioned into kk equal-sized subsets; the model is iteratively trained on k−1k - 1 subsets and tested on the remaining single held-out subset across kk runs, averaging the resulting performance scores.

<p>A resampling technique where available data is partitioned into $$k$$ equal-sized subsets; the model is iteratively trained on $$k - 1$$ subsets and tested on the remaining single held-out subset across $$k$$ runs, averaging the resulting performance scores.</p>
16
New cards

Leave-One-Out Cross-Validation (LOOCV)

An extreme variant of kk-fold cross-validation where kk equals the total sample size NN, such that each fold reserves exactly one observation for validation while training on all remaining N−1N - 1 observations.

17
New cards

Stratified Random Sampling

A partitioning method applied piecewise across class strata to guarantee that every data fold or split contains an equal or proportional representation of each class, reducing evaluation bias.

18
New cards

McNemar's Test

A non-parametric statistical hypothesis test applied to a 2×22 \times 2 contingency table of paired binary outcomes from two classifiers evaluated on identical test samples, testing whether discordant error rates differ significantly.

19
New cards

Discordant Pairs

The test cases in a 2×22 \times 2 contingency table where two compared models disagree—specifically n01n_{01} (Classifier A correct, Classifier B incorrect) and n10n_{10} (Classifier A incorrect, Classifier B correct).

20
New cards

Corrected Repeated kk-Fold Cross-Validation Test

A modified paired statistical test developed by Nadeau and Bengio (2003) that scales estimated variance using a factor of (1nk+#test#train)\left(\frac{1}{nk} + \frac{\#\text{test}}{\#\text{train}}\right) to compensate for sample dependency across overlapping cross-validation training folds.

21
New cards

Permutation Test

A non-parametric resampling test used to establish empirical chance-level performance by repeatedly shuffling ground-truth labels across BB iterations, recalculating classification accuracy, and computing the empirical pp-value.

<p>A non-parametric resampling test used to establish empirical chance-level performance by repeatedly shuffling ground-truth labels across $$B$$ iterations, recalculating classification accuracy, and computing the empirical $$p$$-value.</p>
22
New cards

Sensitivity (Recall)

The proportion of actual positive cases correctly identified by a classifier, calculated as TPTP+FN\frac{\text{TP}}{\text{TP} + \text{FN}}.

23
New cards

Specificity

The proportion of actual negative cases correctly identified by a classifier, calculated as TNTN+FP\frac{\text{TN}}{\text{TN} + \text{FP}}.

24
New cards

Receiver Operating Characteristic (ROC) Curve

A graphical plot displaying a binary classifier's performance across varying discrimination thresholds, plotting Sensitivity (True Positive Rate) on the yy-axis against 1−Specificity1 - \text{Specificity} (False Positive Rate) on the xx-axis.

25
New cards

Area Under the ROC Curve (ROC AUC)

A threshold-independent metric of classification performance where 0.50.5 indicates random performance and 1.01.0 indicates flawless class discrimination, remaining robust against severe class imbalance.

26
New cards
<p>Non-Linear Decision Boundary</p>

Non-Linear Decision Boundary

A curved or multi-dimensional classification boundary that separates classes that cannot be segregated by a flat planar hyperplane.

27
New cards

Decision Boundary

A mathematically defined surface or hyperplane in feature space (wTx+b=0\mathbf{w}^T \mathbf{x} + b = 0) separating distinct classification regions.

<p>A mathematically defined surface or hyperplane in feature space ($$\mathbf{w}^T \mathbf{x} + b = 0$$) separating distinct classification regions.</p>
28
New cards

Fisher's Linear Discriminant Analysis (LDA)

A dimensionality reduction and linear classification method that finds an optimal projection vector w=S−1(μ2−μ1)\mathbf{w} = \mathbf{S}^{-1}(\boldsymbol{\mu}_2 - \boldsymbol{\mu}_1) that maximizes between-class variance relative to within-class scatter.

29
New cards

Within-Class Scatter Matrix (S\mathbf{S})

The sum of the scatter matrices of individual classes, defined as S=S1+S2\mathbf{S} = \mathbf{S}_1 + \mathbf{S}_2, where Sc=∑xi∈Dc(xi−μc)(xi−μc)T\mathbf{S}_c = \sum_{\mathbf{x}_i \in \mathbf{D}_c}(\mathbf{x}_i - \boldsymbol{\mu}_c)(\mathbf{x}_i - \boldsymbol{\mu}_c)^T.

30
New cards

Polysomnography (PSG)

A comprehensive multi-parametric recording standard used in sleep medicine and research, minimally combining electroencephalogram (EEG), electro-oculogram (EOG), and electromyogram (EMG).

31
New cards

Pyramidal Neurons (EEG Generation)

Large projection neurons primarily found in cortical layers III and V whose parallel alignment and synchronized postsynaptic potentials create measurable electrical dipoles detected by scalp electrodes.

<p>Large projection neurons primarily found in cortical layers III and V whose parallel alignment and synchronized postsynaptic potentials create measurable electrical dipoles detected by scalp electrodes.</p>
32
New cards

Postsynaptic Potential (PSP)

A graded local neuroelectrical voltage shift (excitatory or inhibitory) across a dendrite or soma that lasts long enough to summate across thousands of neurons and manifest as scalp EEG, unlike brief action potentials.

33
New cards

Posterior Dominant Rhythm (PDR / Alpha Rhythm)

A sinusoidal 8–13 Hz8\text{--}13\,\text{Hz} EEG oscillation prominent over the occipital cortex during relaxed wakefulness with eyes closed that attenuates upon eye opening.

34
New cards

Low-Amplitude Mixed-Frequency (LAMF) Activity

An EEG pattern consisting of low-voltage background activity predominantly in the theta range (4–7 Hz4\text{--}7\,\text{Hz}), serving as the defining hallmark of stage N1 sleep and REM sleep.

35
New cards

Sleep Spindles

Bursts of distinct sinusoidal waves in the 11–16 Hz11\text{--}16\,\text{Hz} band (most commonly 12–14 Hz12\text{--}14\,\text{Hz}) lasting ≥0.5 s\ge 0.5\,\text{s}, originating in the thalamic reticular nucleus during stage N2 sleep to gate external sensory input.

<p>Bursts of distinct sinusoidal waves in the $$11\text{--}16\,\text{Hz}$$ band (most commonly $$12\text{--}14\,\text{Hz}$$) lasting $$\ge 0.5\,\text{s}$$, originating in the thalamic reticular nucleus during stage N2 sleep to gate external sensory input.</p>
36
New cards

K-Complex

A well-delineated, sharp negative wave followed immediately by a slower positive deflection lasting ≥0.5 s\ge 0.5\,\text{s}, recorded predominantly over frontal derivations during stage N2 sleep to suppress awakening.

<p>A well-delineated, sharp negative wave followed immediately by a slower positive deflection lasting $$\ge 0.5\,\text{s}$$, recorded predominantly over frontal derivations during stage N2 sleep to suppress awakening.</p>
37
New cards

Slow Waves (Delta Waves)

High-voltage EEG oscillations with frequencies between 0.5 and 2 Hz0.5\text{ and }2\,\text{Hz} and peak-to-peak amplitudes exceeding 75 μV75\,\mu\text{V}, originating purely from synchronous cortical oscillations during stage N3 sleep.

38
New cards

Hypnogram

A specialized time-series chart depicting the cyclical progression of sleep stages (Wake, N1, N2, N3, and REM) over the course of a night.

<p>A specialized time-series chart depicting the cyclical progression of sleep stages (Wake, N1, N2, N3, and REM) over the course of a night.</p>
39
New cards

Rapid Eye Movement (REM) Sleep

A distinct sleep stage characterized by low-amplitude mixed-frequency EEG, rapid eye movements on EOG, complete motor atonia on EMG, and variable autonomic activity.

40
New cards

Unihemispheric Sleep

A physiological phenomenon wherein one cerebral hemisphere exhibits slow-wave non-REM sleep while the contralateral hemisphere displays awake EEG, observed in cetaceans, fur seals, and birds.

41
New cards

Memory Consolidation Hypothesis

A restorative theory proposing that sleep facilitates the long-term stabilization and integration of newly acquired behavioral skills and memories into neural architecture.

42
New cards

Protection Hypothesis

An adaptive theory suggesting sleep evolved primarily to keep diurnal organisms quiet, inactive, and sequestered during periods of high vulnerability to predators and nighttime hazards.

43
New cards

Obstructive Sleep Apnea (OSA)

A common disorder marked by recurrent collapse and obstruction of the upper airway during sleep, resulting in sustained respiratory effort, intermittent hypoxia, and frequent transient cortical arousals.

44
New cards

Central Sleep Apnea (CSA)

A sleep-related breathing disorder characterized by recurring cessations of airflow during sleep caused by a lack of central ventilatory drive without physical upper airway obstruction.

45
New cards

Apnea-Hypopnea Index (AHI)

A clinical diagnostic metric reflecting the average hourly frequency of apneas and hypopneas lasting over 10 s10\,\text{s}, categorized as mild (5–155\text{--}15), moderate (15–3015\text{--}30), or severe (>30> 30).

46
New cards

Paradoxical Breathing

An abnormal asynchronous breathing motion observed during obstructive sleep apnea where the thorax and abdomen move out of phase due to severe upper airway collapse.

<p>An abnormal asynchronous breathing motion observed during obstructive sleep apnea where the thorax and abdomen move out of phase due to severe upper airway collapse.</p>
47
New cards

Wearable OSA Screening Sensors

Non-invasive consumer and clinical form factors used to monitor physiological apnea signals, comprised predominantly of adhesive patches (43%43\%), wrist bracelets (32%32\%), chest bands (14%14\%), rings (7%7\%), and headsets (4%4\%).

<p>Non-invasive consumer and clinical form factors used to monitor physiological apnea signals, comprised predominantly of adhesive patches ($$43\%$$), wrist bracelets ($$32\%$$), chest bands ($$14\%$$), rings ($$7\%$$), and headsets ($$4\%$$).</p>
48
New cards

Continuous Positive Airway Pressure (CPAP)

A non-invasive medical device therapy that delivers pressurized ambient air through a mask to pneumatically splint the upper airway open during sleep.