Glossary

Key terms used across the site. The first mention of any of these on a page links back here, and each entry lists the pages that discuss it.

Activation function

The nonlinearity applied to each neuron’s weighted input (e.g. sigmoid, tanh, ReLU); without it a network could only represent linear maps.

See: Neural Networks and the Multilayer Perceptron

Discussed on: Neural Networks and the Multilayer Perceptron

Agentic AI

A language model placed in a loop with tools, which reasons about a goal, calls tools, observes results, and repeats until a task is done.

See: Deep Learning, Foundation Models, and Agentic AI

Discussed on: Deep Learning, Foundation Models, and Agentic AI

Analysis of variance

A method that partitions total variability into between-group and within-group parts and tests their ratio with an F-statistic.

See: Analysis of Variance

Discussed on: Analysis of Variance, Asynchrony and the Inflationary Effect in Metapopulations

Autoencoder

A network trained to compress data through a narrow bottleneck and reconstruct it, so the bottleneck retains only the essential structure.

See: Variational Autoencoders

Discussed on: Deep Learning, Foundation Models, and Agentic AI, Variational Autoencoders

Backpropagation

The chain rule applied layer by layer to compute a neural network’s loss gradients efficiently, so its weights can be updated by gradient descent.

See: Neural Networks and the Multilayer Perceptron

Discussed on: Quantitative Methods, Chain Rule, Neural Networks and the Multilayer Perceptron, Recurrent Networks and LSTMs

Balanced incomplete block design

A blocked design whose blocks are too small to hold every treatment, arranged so every pair of treatments appears together equally often.

See: Balanced Incomplete Block Designs

Discussed on: Balanced Incomplete Block Designs

Basic reproduction number

The average number of secondary infections caused by one infectious individual in a fully susceptible population.

See: The Effective Reproduction Number and Forecasting

Discussed on: BIO301 Mathematical Biology, BIO340 Infectious Disease Ecology, Force of Infection and Serocatalytic Models, Social and Structural Drivers of Transmission, Within-Host Viral Dynamics and Infectiousness, Evolutionary Epidemiology and the Evolution of Virulence, Approximate Bayesian Computation, Branching Processes, Networks in Ecology and Epidemiology, Eigenvalues and Eigenvectors, Adaptive Dynamics and the Evolution of Virulence, Expected Value, Metapopulation Networks and the Invasion Threshold, Fitting Dynamic Models to Data, The Molecular Clock and Phylodynamics, The Next-Generation Matrix and R₀, Phylodynamics, The Effective Reproduction Number and Forecasting, SEIR and Compartmental Extensions, Compartmental Models in Biology, Outbreak Analytics and Modeling, Statistical Modeling of Infectious Disease Dynamics

Bias–variance tradeoff

The tension between a model too simple to capture the pattern (bias) and one so flexible it fits noise (variance); the best model balances the two.

See: Overfitting, Regularization, and Cross-Validation

Discussed on: Quantitative Methods, Overfitting, Regularization, and Cross-Validation

Bland–Altman plot

A plot of the difference between two measurement methods against their mean, used to assess agreement via the bias and limits of agreement.

See: Bland–Altman Agreement

Discussed on: Bland–Altman Agreement

Calibration

A probabilistic prediction is calibrated when, among cases assigned probability p, a fraction close to p actually occur.

See: Uncertainty, Calibration, and Conformal Prediction

Discussed on: Epidemic Forecasting, Genomic Surveillance, Quantitative Methods, Approximate Bayesian Computation, Convolutional Networks and Image Identification, Gaussian Processes, Hidden Markov Models, The Kalman Filter, Fitting Dynamic Models to Data, The Molecular Clock and Phylodynamics, Neural Networks and the Multilayer Perceptron, p-Values, POMP Models and Plug-and-Play Inference, Posterior Predictive Checks, Encoding Spatial Priors with VAEs (PriorVAE), Proper Scoring Rules, Recurrent Networks and LSTMs, Species Distribution Models: Presence–Absence Data, State-Space Models and Particle Filtering, Transformers and Attention, Tree Ensembles: Random Forests and Gradient Boosting, Uncertainty, Calibration, and Conformal Prediction, Vectorial Capacity from Field Data, Outbreak Analytics and Modeling, Statistical Modeling of Infectious Disease Dynamics

Clustering

Grouping observations so that members of a group resemble each other more than members of other groups, without using labels.

See: Clustering and Unsupervised Learning

Discussed on: The 2001 UK Foot-and-Mouth Epidemic: Spatial Spread and Control, Genomic Surveillance, Quantitative Methods, Clustering and Unsupervised Learning, Hidden Markov Models, Random-Graph Models, Networks and Graphs, Spatial Cluster Detection, Spatial Moment Equations, Spatial Point Processes, Variational Autoencoders, Spatial Epidemiology and Disease Mapping

Coarsened exact matching

A matching method that coarsens covariates into bins, exact-matches on the bins, and prunes unmatched units, directly bounding covariate imbalance.

See: Matching Methods

Discussed on: Quantitative Methods, Matching Methods

Compartmental model

A model that divides a population into states — such as susceptible, infectious, and recovered — and tracks the flows between them.

See: Compartmental Models in Biology

Discussed on: BIO301 Mathematical Biology, Behavior–Disease Coupled Models, Climate Forcing in Transmission Models, Exponential and Logistic Growth, Compartmental Models in Biology, Within-Host Dynamics and the Immune Response, Outbreak Analytics and Modeling, Outbreak Analytics Bootcamp

Conformal prediction

A distribution-free method that turns any model into prediction sets guaranteed to contain the truth with a chosen probability under exchangeability.

See: Uncertainty, Calibration, and Conformal Prediction

Discussed on: Uncertainty, Calibration, and Conformal Prediction

Convolutional neural network

A network that slides small learned filters across an image, building features hierarchically from edges to objects.

See: Convolutional Networks and Image Identification

Discussed on: Convolutional Networks and Image Identification, Graph Neural Networks, Remote Sensing, Object Counting, and Outbreak Detection

Cross-validation

Estimating out-of-sample performance by repeatedly training on part of the data and validating on the held-out remainder.

See: Overfitting, Regularization, and Cross-Validation

Discussed on: Overfitting, Regularization, and Cross-Validation, Splines and Penalized Regression, Tree Ensembles: Random Forests and Gradient Boosting, Reproducibility

Crossover design

A trial in which every subject receives every treatment in sequence, separated by washout, so treatments are compared within subjects.

See: Crossover Designs

Discussed on: Crossover Designs

Cumulative incidence function

In competing-risks analysis, the probability of experiencing a specific cause of failure by a given time, accounting for the competing events.

See: Competing Risks

Discussed on: Competing Risks

Decision tree

A model that predicts by a sequence of yes/no feature splits, partitioning the data into regions with their own predictions.

See: Tree Ensembles: Random Forests and Gradient Boosting

Discussed on: Health Economics and Economic Evaluation of Infectious-Disease Interventions, Cost-Effectiveness Analysis, Tree Ensembles: Random Forests and Gradient Boosting

Dimensionality reduction

Compressing high-dimensional data into a few informative coordinates that preserve its important structure, for visualization, denoising, or modeling.

See: Dimensionality Reduction and Embeddings

Discussed on: Dimensionality Reduction and Embeddings

E-value

The minimum strength of association an unmeasured confounder would need with both exposure and outcome to fully explain away an observed effect.

See: E-Values and Unmeasured Confounding

Discussed on: E-Values and Unmeasured Confounding

Effective reproduction number

The average number of secondary infections per case at a given time, as susceptibility and interventions change.

See: The Effective Reproduction Number and Forecasting

Discussed on: LAMP: Isothermal Amplification, qPCR and RT-qPCR, Detection Probability: Viral Kinetics and Assay Thresholds, The Speed and Strength of Epidemic Control, Antimicrobial Resistance Across Scales, Final Size, the Herd Immunity Threshold, and Overshoot, Asynchrony and the Inflationary Effect in Metapopulations, The Renewal Equation, The Effective Reproduction Number and Forecasting, Population Dynamics of Resistance, Compartmental Models in Biology, Burst Size, Latent Period, and Mutation at the Cellular Scale, Outbreak Analytics and Modeling, Writing for Policy

Entomological inoculation rate

The number of infectious bites received per person per unit time, the product of the human biting rate and the sporozoite rate — the field measure of transmission intensity.

See: Vectorial Capacity from Field Data

Discussed on: Quantitative Methods, Vectorial Capacity from Field Data

Epistemic uncertainty

Epistemic uncertainty comes from a model’s ignorance and shrinks as more (or more representative) data are gathered; aleatoric uncertainty is the irreducible randomness of the world and does not. “Aleatoric” derives from the Latin alea — a dice game and, by extension, chance or risk — the same word in Julius Caesar’s “alea iacta est” (“the die is cast”), reputedly said as he crossed the Rubicon in 49 BC to march on Rome.

See: Uncertainty, Calibration, and Conformal Prediction

Discussed on: Quantitative Methods, Uncertainty, Calibration, and Conformal Prediction

Evidence lower bound

A tractable lower bound on the log-likelihood, maximized to train variational models; it trades reconstruction accuracy against a prior-matching penalty.

See: Variational Autoencoders

Discussed on: Quantitative Methods, Encoding Spatial Priors with VAEs (PriorVAE), Variational Autoencoders

Force of infection

The per-capita rate at which susceptible individuals become infected; estimable from age–seroprevalence data via serocatalytic models.

See: Force of Infection and Serocatalytic Models

Discussed on: The 2001 UK Foot-and-Mouth Epidemic: Spatial Spread and Control, Force of Infection and Serocatalytic Models, Quantitative Methods, Behavior–Disease Coupled Models, Social Contact Matrices and Age-Structured Mixing, Lebesgue Integration, The Renewal Equation, The Effective Reproduction Number and Forecasting, Serocatalytic Models

Foundation model

A large model pre-trained on broad data with a self-supervised objective, then adapted to many downstream tasks by fine-tuning or prompting.

See: Deep Learning, Foundation Models, and Agentic AI

Discussed on: Quantitative Methods, Deep Learning, Foundation Models, and Agentic AI

G-estimation

A g-method that estimates a structural nested model by finding the treatment effect that makes the treatment-free outcome independent of treatment given confounders.

See: G-Estimation

Discussed on: G-Estimation

Generation interval

The time between the infection of an infector and the infections they in turn cause.

See: Epidemiological Intervals and Delays

Discussed on: Epidemiology, Fitting Delay Distributions: Truncation and Censoring, The Speed and Strength of Epidemic Control, Epidemic Forecasting, Epidemiological Intervals and Delays, The Euler–Lotka Equation: From Growth Rate to R₀, Quantitative Methods, The Renewal Equation, The Effective Reproduction Number and Forecasting, SEIR and Compartmental Extensions, Outbreak Analytics and Modeling, Statistical Modeling of Infectious Disease Dynamics, Cover Letters and Reviewer Responses

Gradient boosting

An ensemble that builds decision trees sequentially, each fitting the errors of the current model — the workhorse for tabular prediction (XGBoost, LightGBM).

See: Tree Ensembles: Random Forests and Gradient Boosting

Discussed on: Functional Responses and the Paradox of Enrichment, Tree Ensembles: Random Forests and Gradient Boosting

Graph neural network

A neural network that learns on graph-structured data by message passing — each node aggregating information from its neighbours.

See: Graph Neural Networks

Discussed on: Graph Neural Networks

Incubation period

The time between infection and the onset of symptoms in an individual.

See: Epidemiological Intervals and Delays

Discussed on: Climate and Disease Transmission, Fitting Delay Distributions: Truncation and Censoring, Epidemiological Intervals and Delays, Outbreak Investigation, Measures of Center, Moment Matching, Compartmental Models in Biology, Vectorial Capacity from Field Data, Writing Results and Discussion

Inverse probability weighting

Reweighting units by the inverse probability of their observed treatment to build a pseudo-population in which treatment is unconfounded by measured covariates.

See: Inverse Probability Weighting

Discussed on: G-Estimation, Inverse Probability Weighting, Propensity Scores

k-means

A clustering method that partitions data into k groups by minimizing each point’s squared distance to its cluster mean, favouring spherical clusters.

See: Clustering and Unsupervised Learning

Discussed on: Quantitative Methods, Clustering and Unsupervised Learning

Kullback–Leibler divergence

An asymmetric measure of how one probability distribution differs from another — the expected log-ratio of the two — central to maximum likelihood, variational inference, and information theory.

See: Kullback–Leibler Divergence

Discussed on: Quantitative Methods, Kullback–Leibler Divergence, Moment Matching

Large language model

A foundation model for text, trained to predict tokens over a large corpus and adaptable to summarization, extraction, and drafting.

See: Deep Learning, Foundation Models, and Agentic AI

Discussed on: Research Tools and Methods, Global Health Data and Surveillance, Graduate Certificate in Infectious Disease Modeling and Analytics, Quantitative Methods for Antimicrobial Stewardship, Field Epidemiology and Tropical Medicine, Quantitative Methods, Deep Learning, Foundation Models, and Agentic AI, Bayesian Modeling for Epidemiology with Stan, Introduction to Computation and Math for Biology, Introduction to Epidemiology and Biostatistics with R, Outbreak Analytics Bootcamp, Risk Communication for Epidemiology and Outbreak Response

Latent period

The time between infection and the onset of infectiousness (which may differ from the incubation period).

See: Epidemiological Intervals and Delays

Discussed on: Epidemiological Intervals and Delays, The 2001 UK Foot-and-Mouth Epidemic: Spatial Spread and Control, Within-Host Viral Dynamics and Infectiousness, The Exponential Distribution, SEIR and Compartmental Extensions, Burst Size, Latent Period, and Mutation at the Cellular Scale

Latent space

The low-dimensional space of codes a model compresses data into, where each axis captures a factor of variation.

See: Variational Autoencoders

Discussed on: Kullback–Leibler Divergence, Variational Autoencoders

Latin square

A t×t layout in which each treatment appears once in every row and every column, controlling two blocking factors at once.

See: Latin Square Designs

Discussed on: Latin Hypercube Sampling, Latin Square Designs

Limits of agreement

In a Bland–Altman analysis, the interval (mean difference ± 1.96 SD of the differences) within which 95% of the differences between two methods are expected to lie.

See: Bland–Altman Agreement

Discussed on: Quantitative Methods, Bland–Altman Agreement

Long short-term memory

A recurrent-network cell whose forget, input, and output gates protect a long-term memory, letting it learn dependencies across many time steps.

See: Recurrent Networks and LSTMs

Discussed on: Recurrent Networks and LSTMs

Manifold learning

Nonlinear dimensionality reduction that unrolls data lying on a curved low-dimensional surface, preserving local neighbourhoods for visualization.

See: Dimensionality Reduction and Embeddings

Discussed on: Quantitative Methods, Dimensionality Reduction and Embeddings

Marginal structural model

An outcome model fit in the inverse-probability-weighted pseudo-population, whose coefficients are marginal causal effects, valid with time-varying confounding.

See: Inverse Probability Weighting

Discussed on: Quantitative Methods, G-Estimation, Inverse Probability Weighting

Message passing

The core operation of a graph neural network, in which each node updates its state by aggregating transformed messages from its neighbours.

See: Graph Neural Networks

Discussed on: Quantitative Methods, Graph Neural Networks

Meta-analysis

A statistical synthesis that pools estimates from multiple studies into one overall estimate, weighting each by its precision.

See: Meta-Analysis

Discussed on: Diagnostic Test Accuracy Meta-Analysis, Mendelian Randomization, Meta-Analysis, Meta-Regression, Publication Bias and Small-Study Effects, Scientific Writing, Introduction to Epidemiology and Biostatistics with R, Writing Methods

Meta-regression

A random-effects regression of study effect sizes on study-level covariates, used to explain between-study heterogeneity in a meta-analysis.

See: Meta-Regression

Discussed on: Meta-Regression

Multilayer perceptron

A feedforward neural network of fully connected layers — the basic deep-learning architecture.

See: Neural Networks and the Multilayer Perceptron

Discussed on: Neural Networks and the Multilayer Perceptron

Neural network

A model built from layers of simple units, each a weighted sum passed through a nonlinearity, whose weights are learned from data.

See: Neural Networks and the Multilayer Perceptron

Discussed on: Chain Rule, Deep Learning, Foundation Models, and Agentic AI, Neural Networks and the Multilayer Perceptron, Encoding Spatial Priors with VAEs (PriorVAE), Scientific Machine Learning: Neural ODEs and Physics-Informed Networks, Uncertainty, Calibration, and Conformal Prediction, Variational Autoencoders, Writing an Abstract

Neural ordinary differential equation

A model whose dynamics — the right-hand side of an ODE — are a neural network, trained by differentiating through the ODE solver.

See: Scientific Machine Learning: Neural ODEs and Physics-Informed Networks

Discussed on: Scientific Machine Learning: Neural ODEs and Physics-Informed Networks

Next-generation matrix

A matrix whose dominant eigenvalue gives the basic reproduction number in a structured or multi-type transmission model.

See: The Next-Generation Matrix and R₀

Discussed on: Social and Structural Drivers of Transmission, Social Contact Matrices and Age-Structured Mixing, The Next-Generation Matrix and R₀, Reservoir Ecology, Structured Population Models, Outbreak Analytics and Modeling

Object detection

A computer-vision task that locates and classifies every instance of a class in an image, drawing a box around each — the basis of counting objects such as cars or cells.

See: Remote Sensing, Object Counting, and Outbreak Detection

Discussed on: Remote Sensing, Object Counting, and Outbreak Detection

Overfitting

When a model fits the noise in its training data rather than the underlying signal, generalizing poorly to new data.

See: Overfitting, Regularization, and Cross-Validation

Discussed on: Dimensionality Reduction and Embeddings, Gaussian Processes, Neural Networks and the Multilayer Perceptron, Overfitting, Regularization, and Cross-Validation, Splines and Penalized Regression, Tree Ensembles: Random Forests and Gradient Boosting

Physics-informed neural network

A neural network trained to satisfy a known differential equation by adding the equation’s residual to its loss, alongside a data-misfit term.

See: Scientific Machine Learning: Neural ODEs and Physics-Informed Networks

Discussed on: Scientific Machine Learning: Neural ODEs and Physics-Informed Networks

Prediction interval

In a random-effects meta-analysis, the interval within which the true effect of a new study is expected to fall — reflecting between-study heterogeneity, not just uncertainty in the mean.

See: Meta-Analysis

Discussed on: Prospective Outbreak Detection and Aberration Algorithms, Epidemic Forecasting, Excess Mortality, The Kalman Filter, Meta-Analysis, Recurrent Networks and LSTMs

Principal component analysis

A linear dimensionality-reduction method that projects data onto the orthogonal directions of greatest variance, the eigenvectors of its covariance matrix.

See: Dimensionality Reduction and Embeddings

Discussed on: Quantitative Methods, Dimensionality Reduction and Embeddings, Eigenvalues and Eigenvectors, Matrix and Vector Notation, Population Stratification and PCA Control, Variational Autoencoders

Propensity score

The probability of receiving treatment given measured confounders; a balancing score used to remove confounding by matching, stratification, or weighting.

See: Propensity Scores

Discussed on: Propensity Scores

Pseudo-absence

Points sampling the available environment across a landscape, contrasted against occurrence records in presence-only species distribution models.

See: Species Distribution Models: Presence-Only Data

Discussed on: Quantitative Methods, Species Distribution Models: Presence-Only Data

Publication bias

The distortion arising when studies with null or unfavorable results are less likely to be published, inflating pooled effects in a meta-analysis.

See: Publication Bias and Small-Study Effects

Discussed on: Publication Bias and Small-Study Effects

Random forest

An ensemble that averages many decorrelated decision trees, each grown on a bootstrap sample with random feature subsets, to cut variance.

See: Tree Ensembles: Random Forests and Gradient Boosting

Discussed on: Species Distribution Models: Presence–Absence Data, Tree Ensembles: Random Forests and Gradient Boosting, Uncertainty, Calibration, and Conformal Prediction

Recurrent neural network

A network that processes a sequence one step at a time, carrying a hidden state forward as a running summary of what it has seen.

See: Recurrent Networks and LSTMs

Discussed on: Recurrent Networks and LSTMs

Regularization

Penalizing model complexity (e.g. an L1 or L2 term on the parameters) to reduce overfitting and variance.

See: Overfitting, Regularization, and Cross-Validation

Discussed on: Back-Calculation and Deconvolution of Infection Curves, Overfitting, Regularization, and Cross-Validation

Remote sensing

Measurement of the physical world at a distance — from satellites, aircraft, drones, or fixed cameras — used in epidemiology for climate, population, and activity proxies.

See: Remote Sensing, Object Counting, and Outbreak Detection

Discussed on: Remote Sensing, Object Counting, and Outbreak Detection

Reparameterization trick

Writing a random latent draw as a deterministic function of the distribution’s parameters plus fixed noise, so gradients can flow through the sampling step.

See: Variational Autoencoders

Discussed on: Quantitative Methods, Variational Autoencoders

Repeated measures design

A design in which the same units are measured under several conditions or times, using each unit as its own control.

See: Repeated Measures Designs

Discussed on: Bland–Altman Agreement, Hierarchical (Multilevel) Models, Repeated Measures Designs

Resistance ratio

The ratio of the lethal concentration (e.g. LC50) of a resistant population to that of a susceptible reference strain, a graded measure of insecticide-resistance intensity.

See: Insecticide-Resistance Monitoring

Discussed on: Quantitative Methods, Insecticide-Resistance Monitoring

Self-attention

The transformer operation in which every element of a sequence attends to every other, weighting them by learned query–key similarity.

See: Transformers and Attention

Discussed on: Quantitative Methods, Transformers and Attention

Serial dilution

A stepwise dilution of a sample by a fixed factor at each step, producing a geometric series of concentrations used for titration and standard curves.

See: Dilutions, Titers, and Standard Curves

Discussed on: Dilutions, Titers, and Standard Curves

Serial interval

The time between symptom onset in an infector and symptom onset in the people they infect.

See: Epidemiological Intervals and Delays

Discussed on: Fitting Delay Distributions: Truncation and Censoring, Epidemiological Intervals and Delays, The Euler–Lotka Equation: From Growth Rate to R₀, Outbreak Investigation, Transmission Tree Reconstruction (Who Infected Whom), Moment Matching, The Renewal Equation, The Effective Reproduction Number and Forecasting, Scientific Writing, Writing Methods, Writing Results and Discussion

SHAP

A method that attributes a model’s prediction to its features using Shapley values from game theory, so contributions sum exactly to the output.

See: Model Interpretability and SHAP

Discussed on: Quantitative Methods, Model Interpretability and SHAP, Tree Ensembles: Random Forests and Gradient Boosting

Softmax

The softmax maps a vector of scores z=(z1,,zK)z = (z_1, \dots, z_K) to probabilities pi=ezi/jezjp_i = e^{z_i} / \sum_j e^{z_j}, which are non-negative and sum to one. It is the output layer of multi-class classifiers and multinomial logistic regression, and the operation that turns attention scores into weights inside a transformer. A temperature τ\tau (dividing the scores by τ\tau before exponentiating) sharpens the distribution as τ0\tau \to 0 and flattens it as τ\tau \to \infty. In practice it is computed via the numerically stable log-sum-exp trick — subtracting the maximum score before exponentiating — to avoid overflow.

See: Transformers and Attention

Discussed on: Convolutional Networks and Image Identification, Deep Learning, Foundation Models, and Agentic AI, Transformers and Attention, Floating-Point Arithmetic & Numerical Stability

Species distribution model

A model relating a species’ occurrence to environmental covariates to map where it can live, from presence-only or presence-absence data.

See: Species Distribution Models: Presence-Only Data

Discussed on: Species Distribution Models: Presence–Absence Data, Species Distribution Models: Presence-Only Data

Split-plot design

An experiment with two randomization scales — a hard-to-change factor on whole plots and an easier one on sub-plots — giving two error strata.

See: Split-Plot Designs

Discussed on: Split-Plot Designs

Standard curve

A curve fit to standards of known concentration (often a four-parameter logistic) and inverted to read the concentration of unknowns from their assay signal.

See: Dilutions, Titers, and Standard Curves

Discussed on: Diagnostics & Surveillance, ELISA, qPCR and RT-qPCR, Dilutions, Titers, and Standard Curves

Stepped-wedge design

A cluster trial in which groups cross one-way from control to intervention on a staggered, randomized schedule until all are exposed.

See: Stepped-Wedge Designs

Discussed on: Stepped-Wedge Designs

Summary ROC curve

The summary of a diagnostic test accuracy meta-analysis, from a bivariate random-effects model that jointly pools sensitivity and specificity and their threshold correlation.

See: Diagnostic Test Accuracy Meta-Analysis

Discussed on: Diagnostics & Surveillance, Quantitative Methods, Diagnostic Test Accuracy Meta-Analysis

Titer

The reciprocal of the highest (or 50%) dilution of a sample that still produces a defined effect, such as neutralization or agglutination.

See: Dilutions, Titers, and Standard Curves

Discussed on: Bland–Altman Agreement, Dilutions, Titers, and Standard Curves

Transfer learning

Reusing a model pre-trained on a large general dataset and fine-tuning it on a smaller task-specific one, so generic features need not be relearned.

See: Convolutional Networks and Image Identification

Discussed on: Quantitative Methods, Convolutional Networks and Image Identification

Transformer

A neural architecture that replaces recurrence with attention, letting every element of a sequence attend to every other — the basis of modern foundation models.

See: Deep Learning, Foundation Models, and Agentic AI

Discussed on: Quantitative Methods, Deep Learning, Foundation Models, and Agentic AI, Transformers and Attention

Type M error

In design analysis, the factor by which a statistically significant estimate overstates the true effect on average; large in underpowered studies.

See: Type M and Type S Errors

Discussed on: Quantitative Methods, Type M and Type S Errors

Type S error

In design analysis, the probability that a statistically significant estimate has the opposite sign from the true effect; rises as power falls toward the significance level.

See: Type M and Type S Errors

Discussed on: Type M and Type S Errors

Unsupervised learning

Learning structure from data without labels — clustering, dimensionality reduction, density estimation, and anomaly detection.

See: Clustering and Unsupervised Learning

Discussed on: Quantitative Methods

Variational autoencoder

A generative autoencoder that learns a smooth probabilistic latent space by maximizing the ELBO, allowing sampling and anomaly detection.

See: Variational Autoencoders

Discussed on: Clustering and Unsupervised Learning, Dimensionality Reduction and Embeddings, Encoding Spatial Priors with VAEs (PriorVAE), Variational Autoencoders

Vectorial capacity

The expected number of infectious bites that eventually arise from all mosquitoes biting one infectious person on a single day, given by m·a²·pⁿ/(−ln p).

See: Vectorial Capacity from Field Data

Discussed on: Climate and Disease Transmission, Quantitative Methods, Vectorial Capacity from Field Data