Concept Glossary: M¶

This page lists English glossary entries for this letter. Entry bodies are assembled from term source files.

machine learning¶

Meaning: Machine learning uses data or experience to improve a model for prediction, classification, generation, or judgment instead of hand-writing every rule.
Why it matters: It explains why AI moves from explicit rule writing to data-based model improvement and connects directly to generalization and evaluation.
Related concepts: model, model training, parameter
Core Section: P1-3.2
Appears in: P1-1.3, P4-1.1, P4-index, P1-2.3, P4-1.2, P4-2.1

margin¶

Meaning: The safety gap between a classification boundary and the closest data cases to that boundary. In boundary-based classification, it is read as the minimum distance that should be made large among candidate boundaries.
Why it matters: When several boundaries can separate the same data, separation alone does not tell the reader which boundary is more stable. Margin helps inspect whether the boundary clings too closely to one class and whether small perturbations near the boundary could flip predictions easily.
Related concepts: decision boundary, classification, hyperparameter
Core Section: P4-13.1
Appears in: P4-13.1

matrix¶

Meaning: A matrix is a two-dimensional arrangement of numbers organized by rows and columns.
Why it matters: Matrices are the basic shape behind table data, mini-batches, weights, and many neural-network computations.
Related concepts: vector, matrix multiplication, tensor
Core Section: P2-3.1
Appears in: P2-3.3, P2-11.3, P4-7.2

matrix multiplication¶

Meaning: Matrix multiplication combines rows from one matrix with columns from another to create new values or representations.
Why it matters: It is the core operation behind combining input vectors with weights to produce scores, outputs, or new representations.
Related concepts: weighted sum, matrix, linear transformation
Core Section: P2-3.3
Appears in: P2-11.1

material license¶

Meaning: A material license is the permission condition that says how a work, code file, dataset, image, or other material may be used in an AI book or learning resource. Being public and being freely reusable are not the same; copying, modification, distribution, commercial use, and attribution requirements can each differ.
Why it matters: When building an AI book, using or adapting outside material requires checking the actual use conditions. Understanding licenses prevents it is on the internet, I cited the source, and I may use it from being treated as the same claim.
Related concepts: copyright, quotation, training data
Core Section: P1-15.2
Appears in: P6-21.1

maximum likelihood estimation (MLE)¶

Meaning: Maximum likelihood estimation is a way to choose parameters so the observed data are as plausible as possible under the model. For logistic regression, it can first be read as choosing parameters that assign high probability to the correct class.
Why it matters: It lets classification training be read not only as how many labels were correct, but as how much probability was assigned to the correct answers. MLE also connects maximizing log-likelihood and minimizing log loss as two directions of the same learning objective.
Related concepts: logistic regression, log loss, likelihood
Core Section: P4-11.3
Appears in: P4-11.4

mean¶

Meaning: Mean is the central summary obtained by adding values and dividing by the number of values.
Why it matters: Many losses and summaries use a mean, but a single center value can hide spread and outliers.
Related concepts: distribution, variance, outlier
Core Section: P2-5.2
Appears in: P2-1.1, P2-1.2, P2-2.2, P2-5.3, P2-5.4, P2-6.1, P2-15.1, P3-5.2, P4-6.3

mean squared error, MSE¶

Meaning: Mean squared error averages squared differences between predictions and true values, penalizing large errors more strongly.
Why it matters: MSE connects loss functions and evaluation metrics while showing why large errors can dominate a summary.
Related concepts: error, loss function, mean
Core Section: P2-6.2
Appears in: P2-2.2, P2-15.1, P5-4.2, P4-10.2

document retrieval metadata¶

Meaning: Document retrieval metadata is information about a document or record, such as title, date, source, permission, type, or location, used to filter, trace, or interpret retrieval results.
Why it matters: Search and retrieval need more than semantic similarity; freshness, permission, source, and document type also affect usability.
Related concepts: vector database, filtering, provenance
Core Section: P1-13.4
Appears in: P1-13.2, P1-14.1, P6-12.2, P6-13.1, P6-13.2

metric¶

Meaning: A metric is a numeric criterion for summarizing model performance or a type of error.
Why it matters: Model comparison depends on what the metric rewards or hides, so a high score is meaningful only after the criterion is understood.
Related concepts: model evaluation design, accuracy, recall
Core Section: P4-6.1
Appears in: P2-6.2, P4-index, P2-15.2, P4-3.1, P4-3.2, P4-4.1, P4-5.2, P4-6.2, P4-6.3, P4-6.4, P4-8.2, P4-8.3, P4-9.3, P4-10.2, P4-15.3, P5-4.1

missing value¶

Meaning: A missing value is a value or record that is empty where an observation should exist. It may be a simple blank, but it can also come from sensor failure, missing records, uncertain sample boundaries, or changes in operating conditions.
Why it matters: If all missing values are filled or dropped in the same way, we may miss whether the sample structure or feature meaning has already broken. The concept helps readers separate values that can be filled, information that should remain as a flag, and samples that should be pulled back from comparison. It also connects preprocessing to data-modeling judgment rather than treating it as simple cleaning.
Related concepts: data quality check, sample unit, feature, preprocessing, source data
Core Section: P3-5.5
Appears in: P3-5.5, P4-7.2, P4-7.3

model¶

Meaning: A model is a simplified computational form built for a purpose. In AI it often means a learned structure that maps input to output, but it is not the whole application or system.
Why it matters: The distinction prevents readers from confusing model behavior with system design, input design, review flow, or product quality.
Related concepts: model input, model output, parameter, task definition
Core Section: P1-4.1
Appears in: P1-14.1, P4-1.1, P4-1.2, P4-3.1

Model Context Protocol, MCP¶

Meaning: Model Context Protocol is an open protocol for connecting AI applications with external tools, resources, and prompts in a common way.
Why it matters: MCP makes external capability discovery and invocation more consistent across AI agent and AI-app systems.
Related concepts: tool use, AI agent, permission, least privilege
Core Section: P6-16.1
Appears in: P1-14.3, P1-14.4, P1-14.5, P6-16.2

model input definition¶

Meaning: Model input definition identifies the information a model actually receives for one computation. It is not the whole outside world or the entire raw dataset, but the values brought into the current run.
Why it matters: A model can only use the evidence it is given. Input design determines what cues the model can see, what it cannot see, and how preprocessing or feature design affects performance.
Related concepts: model output, data, model, feature, task definition
Core Section: P1-4.2
Appears in: P1-4.1, P4-index, P3-5.3, P3-5.4

model output definition¶

Meaning: Model output definition identifies the result a model is expected to produce. It may be a number, a category, a sentence, a report, or a candidate list. The key point is that output is not merely the visible final answer; it is also a design choice that reveals what form the problem has been given.
Why it matters: The same real-world situation can become very different modeling tasks depending on the output definition. Defining a problem is closely tied to deciding what counts as input and what should be returned as output. This concept explains why the same data can lead to classification, regression, summarization, recommendation, or comparison-report generation, each with different evaluation criteria.
Related concepts: model input, supervised learning label, model, output structure, task definition
Core Section: P1-4.2
Appears in: P1-8.1, P4-index, P1-10.1, P1-14.1, P1-14.3, P1-14.5, P1-15.3, P4-6.2, P5-3.6

model selection¶

Meaning: Model selection is the process of narrowing candidate model families by problem type, data condition, interpretability, and cost.
Why it matters: It prevents choosing algorithms by fame alone and makes the first experiment order explicit.
Related concepts: task definition, baseline, model evaluation design
Core Section: P4-8.1
Appears in: P4-3.2, P4-8.2, P4-3.1, P4-4.1, P4-5.1, P4-7.2, P4-9.1, P4-9.2, P4-9.3

model training¶

Meaning: Model training is the process of using past examples and a loss signal to adjust a model's internal values or decision rules.
Why it matters: Everyday wording often blurs learning and training, so the broader idea of improvement through experience should be separated from the narrower update procedure that adjusts a model.
Related concepts: parameter, model, generalization, inference
Core Section: P1-3.2
Appears in: P1-4.1, P4-index, P1-5.1, P2-15.2, P5-6.1, P5-6.2, P5-6.3, P5-6.4, P4-2.1

model validation¶

Meaning: Model validation is an intermediate evaluation data split or procedure used to compare settings, preprocessing choices, or model options during development.
Why it matters: Model validation keeps model selection separate from final testing. It helps readers distinguish the place where choices are compared from the place where final performance is reported.
Related concepts: test data, data leakage, generalization, model evaluation design
Core Section: P4-4.2
Appears in: P2-12.3, P4-index, P4-4.1, P4-9.2, P4-15.3, P4-3.1, P4-3.2

multilayer neural network¶

Meaning: A multilayer neural network stacks multiple computation layers so intermediate representations can be transformed step by step.
Why it matters: It explains why depth means staged representation transformation, not just repeating calculations.
Related concepts: weighted sum, representation, activation function
Core Section: P5-2.1
Appears in: P5-2.2, P5-11.1

modeling task¶

Meaning: A concrete task definition that states what input is given, what output is expected, and what success criterion will be used.
Why it matters: It prevents model choice from happening before the problem has been shaped into a learnable or evaluable task.
Related concepts: Read together with the surrounding glossary entries and the linked Sections for this term.
Core Section: P1-4.4

market substitution¶

Meaning: A copyright and policy concern in which a use may replace demand for the original work or its licensed market.
Why it matters: It matters when evaluating whether reuse of protected material can harm the original creator or market.
Related concepts: Read together with the surrounding glossary entries and the linked Sections for this term.
Core Section: P1-15.2

manifold¶

Meaning: A space that may be curved globally but can be treated like a simpler coordinate space in small local regions.
Why it matters: It helps explain why high-dimensional data can sometimes be studied through lower-dimensional structure.
Related concepts: Read together with the surrounding glossary entries and the linked Sections for this term.
Core Section: P2-3.2

motion planning¶

Meaning: The problem of finding a feasible path or sequence of movements for an agent or system under constraints.
Why it matters: It connects search, planning, and robotics examples to the broader idea of choosing actions within a state space.
Related concepts: Read together with the surrounding glossary entries and the linked Sections for this term.
Core Section: P1-7.4