I am part of the Applied Research team of the Artificial Intelligence Lab, working on the MEMOBRAIN project. This collaboration between AIMS (an interdisciplinary research group on the intersection between clinical, medical and engineering sciences at UZ Brussels) and the AI Lab focuses on developing interpretable ways to model functional connectivity within brains. Previously, I also worked on the Cyber Security Artificial Intelligence (CSAI) project, where I applied my ongoing PhD research topic (on sequence learning and anomaly detection) to create models and techniques that provide a degree of intelligence for cyber security beyond automated data analytics.

In the summer of 2019, I was a part of a new project on behaviour anomaly detection in a cyber security context. For this project, I wrote a data pre-processor for machine learning models. I resumed my work in the summer of 2020, where I worked on detecting anomalies using finite automata. This project was a collaboration between the VUB AI Lab and NVISO, a Brussels-based cybersecurity firm.

We propose the hierarchical-alphabet automaton (HAA), a novel finite-state acceptor for languages of nested words. These languages can be specified using nested regular expressions (NRE), which are typically compiled to non-deterministic nested word automata (NWA). However, the traditional NWA suffers from exponential state explosion during determinisation and does not explicitly distinguish patterns of different nesting layers, which hinders interpretability. We introduce an alternative model using a partially ordered set of deterministic finite automata (DFA) to explicitly model nesting layers. This hierarchical structure allows us to describe patterns at different nesting depths in separate reusable automata, which introduces modularity and facilitates incremental language construction.

Several existing anomaly detection algorithms that work on sequence data employ similar pipelines that convert subsequences into piecewise approximations before aggregating these into a model. However, choices in sequence segmentation, representation, distance calculation, and modelling approaches affect detection performance across different anomaly types. We propose a framework that formalises these pipeline components and their parameters, enabling systematic evaluation of pipeline configurations across specific sequence learning tasks. Through this formalisation, we introduce a novel hierarchical piecewise approximation model based on nested words. Using the UCR Time Series Anomaly Archive, we demonstrate that specific framework configurations effectively identify subsequences corresponding to specific anomaly definitions, achieving performance comparable to state-of-the-art methods, including deep learning approaches.

The Applied Research team at the Vrije Universiteit Brussel’s Artificial Intelligence Lab introduces their Python framework for implementing and testing a set of related algorithms on sequential data. This demonstration highlights how sequential data can be segmented, discretised, and transformed into aggregate structures to tackle tasks such as anomaly detection and series classification. Applicable across various domains and use cases, this framework aims to become a valuable tool for both researchers and practitioners.