Advances in Streaming Novelty Detection

Defended in October 2022

Abstract:

Due to the massive growth of machie learning applications motivated by the outstanding results obtained in a wide variety of areas that range from medicine, biology and economics to engineering and physics, new challenging problems have been encountered in recent years. In this dissertation, a humble contribution to the Streaming Novelty Detection (SND) literature is proposed. Firstly, by giving a taxonomy of terms to learning scenarios that tries to give a short step into the standardization of the field. Secondly, two methodological contributions to the SND problem are given. In a nutshell, SND referes to the problem where new classes can emerge in an unsupervised manner throughout a stream. As a first methodological contribution, a probabilistic framework that is based on a mixture of Gaussian distributions is proposed, where also an in-depth revision of the evaluation methodology is performed. As a second contribution, a method that deals with time series data is proposed. It uses deep autoencoders and deep Support Vector Data Description networks as main actors to deal with the SND problem.

• Link to the dissertation: Advances in Streaming Novelty Detection
• Defence presentation: Presentation of Advances in Streaming Novelty Detection

SNDProb: A Probabilistic Approach for Streaming Novelty Detection

Published in IEEE Transactions on Knowledge and Data Engineering Journal

Abstract:

A probabilistic framework for streaming novelty detection is proposed and illustrated with a mixture of Gaussian distributions that models the set of classes. Instances are predicted based on the probability of belonging to each of the classes. Those for which the model cannot provide confident predictions are introduced into a fixed-sized buffer. When the buffer is full, an Expectation Maximization (EM) algorithm is run to search for new emerging classes in the buffer, and update the current model. The EM algorithm has to deal with an scenario where both probability distributions and instances are available. To overcome this issue, the probability distributions (classes) are weighted. The weights are inferred using a meta-regression model which has been pretrained and supplied with the proposed algorithm. Experiments have been run using synthetic datasets to have a close control over the class arrival strategies, the shape, and the overlapping degree between classes. It is shown that when the assumptions of the probabilistic model are fulfilled, the proposed method outperforms literature non-parametric approaches. Furthermore it obtains competitive results in the case of non-Gaussian classes. The experiments reveal, for the first time, the high sensitivity of the novelty detection algorithms to the class arrival strategies.

• Publication Date: April 2022
• Impact Factor (2020): 6.977
• Link to Journal: https://ieeexplore.ieee.org/document/9761762
• Code: https://github.com/andercarreno/SNDProb

Extensión de un Paquete de Algoritmos Metaheurísticos en R para la Docencia

Graduate dissertation

Want to know more about this thesis? Read the 1^st chapter! Introduction (Spanish)

• Date: July 2016
• Honors mark achieved