Adaptive approach for density-approximating neural network models for anomaly detection

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21340%2F21%3A00344025" target="_blank" >RIV/68407700:21340/21:00344025 - isvavai.cz</a>

Výsledek na webu

<a href="https://link.springer.com/chapter/10.1007/978-3-030-57805-3_39" target="_blank" >https://link.springer.com/chapter/10.1007/978-3-030-57805-3_39</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-030-57805-3_39" target="_blank" >10.1007/978-3-030-57805-3_39</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Adaptive approach for density-approximating neural network models for anomaly detection

Popis výsledku v původním jazyce

We propose an adaptive approach for density-approximating neural network models, the alternative use of neural models in anomaly detection. Instead of modeling anomaly indirectly through reconstruction error as is common in auto-encoders, we propose to use a neural model to efficiently approximate anomaly as inferred by k-Nearest Neighbor, which is popular due to its good performance as anomaly detector. We propose an adaptive approach to model the space of kNN inferred anomalies to obtain a neural model with comparable accuracy and considerably better time and space complexity. Moreover, the neural model can achieve even better accuracy in case of noisy data as it allows better control of over-fitting through control of its expressivity. The key contribution over our previous results is the adaptive coverage of kNN induced anomaly space through modified Parzen estimate, which then enables generating arbitrarily large training set for neural model training. We evaluate the proposed approach on real-world computer network traffic data provided by Cisco Systems.

Název v anglickém jazyce

Adaptive approach for density-approximating neural network models for anomaly detection

Popis výsledku anglicky

We propose an adaptive approach for density-approximating neural network models, the alternative use of neural models in anomaly detection. Instead of modeling anomaly indirectly through reconstruction error as is common in auto-encoders, we propose to use a neural model to efficiently approximate anomaly as inferred by k-Nearest Neighbor, which is popular due to its good performance as anomaly detector. We propose an adaptive approach to model the space of kNN inferred anomalies to obtain a neural model with comparable accuracy and considerably better time and space complexity. Moreover, the neural model can achieve even better accuracy in case of noisy data as it allows better control of over-fitting through control of its expressivity. The key contribution over our previous results is the adaptive coverage of kNN induced anomaly space through modified Parzen estimate, which then enables generating arbitrarily large training set for neural model training. We evaluate the proposed approach on real-world computer network traffic data provided by Cisco Systems.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2021

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Advances in Intelligent Systems and Computing

ISBN

9783030578046

ISSN

2194-5357

e-ISSN

—

Počet stran výsledku

11

Strana od-do

415-425

Název nakladatele

Springer Nature

Místo vydání

—

Místo konání akce

Burgos

Datum konání akce

16. 9. 2020

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—