Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Bayes' Theorem
Mathematical formula for calculating the conditional probability of an event given that another event has occurred, theoretical foundation of Naïve Bayes classifiers.
Conditional Independence
Simplifying assumption of Naïve Bayes stating that features are independent of each other given the class, although this assumption is often violated in practice.
Likelihood
Probability of observing the given features given a specific class, calculated as the product of individual probabilities in the Naïve Bayes model.
Prior Probability
Initial probability of belonging to a class before observing features, estimated from relative frequencies of classes in the training set.
Posterior Probability
Updated probability of belonging to a class after observing features, final result of Bayesian calculation used for classification.
Evidence
Normalizing term in Bayes' theorem representing the marginal probability of observing the features, often omitted in comparative classification.
Multinomial Naïve Bayes Classifier
Variant of Naïve Bayes optimized for discrete features and occurrence counts, particularly effective for text classification and document analysis.
Gaussian Naïve Bayes Classifier
Variant assuming that features follow a normal distribution, suitable for continuous data and numerical attributes in the feature space.
Bernoulli Naive Bayes Classifier
A variant that treats features as binary variables indicating the presence or absence of an attribute, ideal for document classification based on keywords.
Laplace Smoothing
A regularization technique that adds a constant to the counts to avoid zero probabilities, essential when estimating conditional probabilities in Naive Bayes.
Generative Model
A type of model that learns the joint distribution P(X,Y) of data and labels, allowing it to generate new samples and compute conditional probabilities.
Expectation-Maximization (EM)
An iterative algorithm for parameter estimation in models with latent data, sometimes used to train variants of Naive Bayes with hidden variables.
Log-score
A metric that evaluates predictive quality using log-probabilities, avoiding numerical underflows when calculating products of probabilities in Naive Bayes.
Naive Assumption
A simplifying postulate of independence between features which, although unrealistic, enables efficient calculations and often good practical performance in classification.
Feature Vector
A vector representation of an observation's attributes, the fundamental element on which the conditional probability calculations in Naive Bayes operate.
Conditional Probability Distribution
A function describing the probability of features given a specific class, modeled independently for each attribute in the Naive Bayes approach.
Likelihood Ratio
Ratio of conditional probabilities between different classes, used in efficient implementations of Naïve Bayes to avoid redundant calculations.
Multiclass Classification
Natural extension of Naïve Bayes beyond binary classification, using Bayes' theorem to calculate posterior probabilities for each possible class.
Kernel Density Estimation
Alternative non-parametric method for estimating probability distributions in Naïve Bayes when Gaussian assumptions are not valid.