Research
The broad topic of my doctoral research is software anomaly detection. So far I have focused on software logs. The idea is that by using methods such as Next Event Prediction or Machine Learning, we can find the anomalous loglines in a dataset that can have millions of lines. I started my work on this topic in 2022 in the University of Oulu and moved to University of Helsinki in 2023. So far I have three publications that I intend to include in my PhD:
How to Configure Masked Event Anomaly Detection on Software Logs?
The first paper was based on previous work that showed the initial results of pinpointing anomalous events with next event prediction. The goal of the paper was to optimize that approach by including multiple variables for the configuration of the experiments.
DOI: https://doi.org/10.1109/ICSME55016.2022.00050
Event-level Anomaly Detection on Software logs: Role of Algorithm, Threshold, and Window Size
In the second paper, the goal was to use ground truth labels from the BGL dataset to be able to assess accuracy metrics such as precision, recall and F-Score, and compare deep learning models with the simpler N-Gram model. Also, threshold adjustment played a role in the paper.
DOI: https://doi.org/10.1109/QRS62785.2024.00070
Speed and Performance of Parserless and Unsupervised Anomaly Detection Methods on Software Logs
On the third paper, the aim was to assess the performance (both time and accuracy metrics) of unsupervised anomaly detection methods. Additionally, we used different representation methods which could further simplify the process.
DOI: https://doi.org/10.1109/QRS62785.2024.00071
Future work
Currently, I’m finalizing a journal extension of my third paper. This will include new datasets and analysis on different preprocessing approaches. I am also looking into ways to combine system performance metrics with the software logs to improve prediction accuracy.
About me
In the 1st FAST Sprint, when thinking about research personalities, I said that I’m a turtle researcher who moves slowly and when I encounter a threat I withdraw to my shell. To be fair, I was joking, but at the same time every good joke has a grain of truth. To me, humor is the shell, and then my statement becomes a liar’s paradox: If I was joking (i.e. lying), then I was telling the truth, but if I was being sincere then I wasn’t telling the truth. Be that as it may, what did I actually mean by being a turtle? Perhaps the Aesop’s fable where tortoise and hare are racing but when the hare gets overconfident he starts to sleep and loses the race? Or Zeno’s paradox of Achilles and Tortoise racing but as the tortoise has a headstart, Achilles always has to reach the point where the tortoise was and by then the tortoise has gone further, so in the end Achilles will never go past the tortoise? I think both of these descriptions, and many others, add up to the turtle archetype that I’m happy to personify. But let’s be real: In reality there are hare and Achilles researchers who, in fact, progress fast and efficiently. I am happy for them too, and I’m trying to stay right on their heels!
There’s another apt reference to turtle in the idiom “If you see a turtle on a fence post, you know he didn’t get up there by himself”. This would be a good opportunity to talk about the impostor syndrome that’s very prevalent in our field and academia in general. However, whatever doubt I might have of how did this turtle get on the fence post, I like to direct it as sense of gratitude toward my supervisor and co-workers (past and present) who have pushed me forward.
Note: One part of this page was created with generative AI. You can probably guess which.
