Anastasia Natsiou, Sagar Saxena
Huawei Ireland, one of the leading tech firms in AI research organises an annual University Challenge for postgraduate students to compete on a technical solution using Machine Learning techniques. This competition has been for us a learning opportunity to understand and apply machine learning methodologies to solve a complex industrial problem. This challenge has not only helped us to hone our technical prowess but to perform as a team and learn from each other as well. The excitement to compete and win lucrative awards offered by Huawei acted as an incentive.
The Huawei Time Series Anomaly detection competition aims at identifying anomalies in a data set provided by Huawei Ireland. This data set included labeled sensor-data recorded at specific time intervals for anomalous behavior. Time series analysis is an active area of research in the field of machine learning. But anomaly detection in time-series has been one of the most challenging aspects. This may be due to the imbalances in the dataset-distribution formed by labeling anomalous data, which are very small compared to normal data. We are trying to come up with an algorithm that would help in predictive maintenance by predicting anomalous behavior depending on KPIs provided.
Although the competition is still on, has already taught us a myriad of different aspects that we were effectively unaware of to date. Such as learning about anomalous behavior in temporal data and methodologies proposed in the literature to be effective, working in a team, sharing knowledge through brainstorming techniques, to learn from each other, managing work remotely, and meeting deadlines for predefined milestones. Thus, the competition for us is an implicit opportunity to get an insight into this domain-application of ML while learning how to maintain better software engineering practices for code-readability.