JSDivergence (Divergence de Jensen-Shannon)
JSDivergence, ou Divergence de Jensen-Shannon, is a statistical method used to quantify the similarity between two distributions de probabilité. It is particularly useful in fields such as apprentissage automatique, théorie de l'information, and statistics. Unlike other divergence measures like Divergence de Kullback-Leibler, JSDivergence is symmetric, meaning that the divergence from distribution A to B is the same as from B to A. This property makes it more interpretable and applicable in various scenarios.
Mathématiquement, la JSDivergence est définie comme la moyenne des divergences de Kullback-Leibler de chaque distribution à une distribution mixte. Plus précisément, étant donné deux distributions de probabilité discrètes P et Q, la JSDivergence se calcule selon la formule :
JSD(P || Q) = 0,5 * KL(P || M) + 0,5 * KL(Q || M)
où M est la moyenne de P et Q, définie comme M = 0,5 * (P + Q), and KL denotes the Kullback-Leibler divergence.
This divergence takes values between 0 and 1, where a value of 0 indicates that the two distributions are identical, while a value closer to 1 indicates greater dissimilarity. The symmetric nature of JSDivergence makes it particularly effective for tasks such as clustering, classification, and l'évaluation de modèles en apprentissage automatique.