We feel honored to announce that Prof. João Gama, Full Professor at the School of Economics, University of Porto, Portugal as the Invited talk speaker.
Prof. João Gama is a Full Professor at the School of Economics, University of Porto, Portugal. He received his Ph.D. in Computer Science from the University of Porto in 2000. He is EurIA Fellow, IEEE Fellow, Fellow of the Asia-Pacific AI Association, and member of the board of directors of the LIAAD, a group belonging to INESC TEC. He is an Editor of several top-level Machine Learning and Data Mining journals. He is ACM Distinguish Speaker. He served as Program Chair of ECMLPKDD 2005, DS09, ADMA09, EPIA 2017, DSAA 2017, served as Conference Chair of IDA 2011, ECMLPKDD 2015, DSAA’2021, and a series of Workshops on KDDS and Knowledge Discovery from Sensor Data with ACM SIGKDD, and ACM SIGAPP. His main research interests are in knowledge discovery from data streams, evolving data, probabilistic reasoning, and causality. He published more than 300 reviewed papers in journals and major conferences. He has an extensive list of publications in data stream learning.
From Fault Detection to Anomaly Explanation: A Case Study on Predictive Maintenance
Predictive Maintenance applications are increasingly complex, with interactions between many components. Based on deep-learning techniques, black-box mod- els are popular approaches due to their predictive accuracy. This paper pro- poses a neural-symbolic architecture that uses an online rule-learning algorithm to explain when the black-box model predicts failures. The proposed system solves two problems in parallel: i) anomaly detection and ii) explanation of the anomaly. For the first problem, we use an unsupervised state-of-the-art LSTM-autoencoder. For the second problem, we train a rule learning system that learns a mapping from the input features to the reconstruction error of the autoencoder. Both systems run online and in parallel. The LSTM-autoencoder signals an alarm for the examples with a reconstruction error that exceeds a threshold. The rule that triggers for that example explains the failure signal by indicating which sensors contribute to the alarm and allows the identification of the component involved in the failure. We evaluate the proposed system in a real-world case study of Metro do Porto. We present examples of explanations that illustrate their utility.