Bekhzod Anvarov
JEPA Representation Learning for Intrusion Detection on Real-World Drifting Network Traffic.
Rel. Guido Marchetto, Alessio Sacco, Cristiano Serra. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (2MB) | Preview |
Abstract
Today, the use of the Anomaly Detection Classifier (ADC) is standard practice in Intrusion Detection Systems. The ADC is trained to map raw features of network flow records to binary labels using a dataset collected in a single time period. However, network traffic keeps evolving after the training data is collected. The same feature values come to carry different labels, so the feature-label relation the ADC was fitted on no longer holds, a shift known as concept drift. The raw features are the only input the ADC has. Under drift, the ADC misclassifies traffic from later periods and its accuracy falls.
What is needed is a model that can correctly classify flows under concept drift without refitting on new traffic
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
