|
Anomaly Detection in System Log Files Using Machine Learning
Moresová, Eva ; Burgetová, Ivana (referee) ; Matoušek, Petr (advisor)
Detekcia anomálií v logoch je dôležitý proces, ktorý pomáha detekovať poruchy systému, pokusy o prienik do systému a ďalšie škodlivé správanie, prípadne týmto udalostiam umožňu\-je predchádzať. Moderné systémy však produkujú logy v množstvách, ktoré nie je možné analyzovať ručne. Preto sa na tento účel používa množstvo automatizovaných metód, od techník založených na pravidlách, až po prístupy používajúce hlboké učenie. Cieľom tejto diplomovej práce je porovnať niekoľko metód detekcie anomálií v logoch a určiť, ktorá z nich je najviac vhodná pre použitie na veľkých log súboroch z praxe. Reprezentantom takýchto dát je zbierka logov z produkčného AAA servera, ktoré boli poskytnuté firmou AT&T. Okrem AT&T logov boli metódy aplikované a vyhodnotené na dvoch ďalších anotovaných datasetoch, z ktorých jeden bol obohatený o synteticky generované anomálie. Táto práca využíva tri metódy detekcie anomálií: lokálny odľahlý faktor, zhlukovací algoritmus DBSCAN a OPTICS framework. Prvé dve metódy skúmajú logy na úrovni jednotlivých záznamov, zatiaľ čo posledná analyzuje celé sekvencie logov. Všetky metódy dosiahli výsledky porovnateľné s prácami, ktoré realizujú podobné prístupy.
|
|
Application of Unsupervised Learning Methods in Graph Similarity Search
Sabo, Jozef ; Burgetová, Ivana (referee) ; Křivka, Zbyněk (advisor)
Goal of this master's thesis was in cooperation with the company Avast to design a system, which can extract knowledge from a database of graphs. Graphs, used for data mining, describe behaviour of computer systems and they are anonymously inserted into the company's database from systems of the company's products users. Each graph in the database can be assigned with one of two labels: clean or malware (malicious) graph. The task of the proposed self-learning system is to find clusters of graphs in the graph database, in which the classes of graphs do not mix. Graph clusters with only one class of graphs can be interpreted as different types of clean or malware graphs and they are a useful source of further analysis on the graphs. To evaluate the quality of the clusters, a custom metric, named as monochromaticity, was designed. The metric evaluates the quality of the clusters based on how much clean and malware graphs are mixed in the clusters. The best results of the metric were obtained when vector representations of graphs were created by a deep learning model (variational graph autoencoder with two relation graph convolution operators) and the parameterless method MeanShift was used for clustering over vectors.
|
|
Acceleration of Neurostimulation Using Artificial Intelligence Methods
Gaňo, Martin ; Chlebík, Jakub (referee) ; Jaroš, Jiří (advisor)
Treatment using transcranial ultrasound is a rapidly arising domain of medicine. This method brings options for non-invasive brain therapies, including ablation, neuromodulation, or potentially opening the blood-brain barrier for the following treatment. The health officer needs to constantly receive feedback on the ultrasound wavefield in the human skull in real-time to accomplish the cure using these techniques. The traditional methods for simulating monochromous ultrasound waves are computationally too expensive. That is why their usage would be infeasible for these purposes, and it brings the need for alternative methods. This work proposed and implemented a method to solve the Helmholtz equation in 3D space using a neural network achieving a faster convergence rate. The neural network design uses lightweight architecture based on UNet. The main interest of this work is neuromodulation because, in this application, it is possible to ignore several variables and phenomena that would not be negligible in other use cases. Omitting them from the calculations increased the chances of accomplishing computations in a reasonable time. The method is fully unsupervised and uses exclusively artificially generated spherical harmonics and physics-based loss for training, with no required ground truth labels. Results showed a faster calculation with acceptable error than other traditional methods.
|
|
Acceleration of Neurostimulation Using Artificial Intelligence Methods
Gaňo, Martin ; Chlebík, Jakub (referee) ; Jaroš, Jiří (advisor)
Treatment using transcranial ultrasound is a rapidly arising domain of medicine. This method brings options for non-invasive brain therapies, including ablation, neuromodulation, or potentially opening the blood-brain barrier for the following treatment. The health officer needs to constantly receive feedback on the ultrasound wavefield in the human skull in real-time to accomplish the cure using these techniques. The traditional methods for simulating monochromous ultrasound waves are computationally too expensive. That is why their usage would be infeasible for these purposes, and it brings the need for alternative methods. This work proposed and implemented a method to solve the Helmholtz equation in 3D space using a neural network achieving a faster convergence rate. The neural network design uses lightweight architecture based on UNet. The main interest of this work is neuromodulation because, in this application, it is possible to ignore several variables and phenomena that would not be negligible in other use cases. Omitting them from the calculations increased the chances of accomplishing computations in a reasonable time. The method is fully unsupervised and uses exclusively artificially generated spherical harmonics and physics-based loss for training, with no required ground truth labels. Results showed a faster calculation with acceptable error than other traditional methods.
|
|
Application of Unsupervised Learning Methods in Graph Similarity Search
Sabo, Jozef ; Burgetová, Ivana (referee) ; Křivka, Zbyněk (advisor)
Goal of this master's thesis was in cooperation with the company Avast to design a system, which can extract knowledge from a database of graphs. Graphs, used for data mining, describe behaviour of computer systems and they are anonymously inserted into the company's database from systems of the company's products users. Each graph in the database can be assigned with one of two labels: clean or malware (malicious) graph. The task of the proposed self-learning system is to find clusters of graphs in the graph database, in which the classes of graphs do not mix. Graph clusters with only one class of graphs can be interpreted as different types of clean or malware graphs and they are a useful source of further analysis on the graphs. To evaluate the quality of the clusters, a custom metric, named as monochromaticity, was designed. The metric evaluates the quality of the clusters based on how much clean and malware graphs are mixed in the clusters. The best results of the metric were obtained when vector representations of graphs were created by a deep learning model (variational graph autoencoder with two relation graph convolution operators) and the parameterless method MeanShift was used for clustering over vectors.
|