Mourad Khayati

Title: Towards Multimodal Pipelines for Time Series Cleaning

Summary: Time series data collected from real-world systems is inherently noisy, incomplete, and increasingly heterogeneous. Traditional cleaning pipelines have largely operated on raw numerical streams in isolation, treating data repair as independent task. Yet modern monitoring infrastructures produce not only numerical measurements but also logs, events, and contextual signals, a richness that single-modality approaches are ill-equipped to exploit.

In this talk, I argue for a shift towards multimodal pipelines for time series cleaning, where complementary data sources are fused to guide and improve each stage of the cleaning process. I will present an overview of recent advances in multimodal data fusion, and discuss how cross-modal signals can be leveraged to enable more robust data repair.

I further introduce TS3D, a temporal multimodal dataset built for distributed database system analysis, as a concrete framework for evaluating such pipelines. Drawing on lessons from tools such as ImputeGAP and A-DARTS, I will highlight the open challenges that remain, including modality alignment, scalability, and the need for reproducible evaluation frameworks, and outline a research agenda for building the next generation of multimodal time series cleaning systems.

Bio: Mourad Khayati is a Senior Lecturer at the Department of Computer Science of the University of Fribourg, Switzerland. He obtained his PhD from the University of Zurich, Switzerland, under the supervision of Prof. Michael Böhlen. His research interests lie in the field of Time Series analytics, with a special focus on data cleaning, missing values imputation, and time series data management. Over the years, he developed a number of time series tools, and actively participated in promoting reproducibility. His ImputeBench framework received the Best Experiments and Analysis Award at VLDB 2020, and his ORBITS streaming algorithm for time series imputation received a mention at the VLDB 2021 Reproducibility Highlights. He regularly serves on the review boards of data engineering and database venues, including the VLDB Journal, TKDE, KDD, and EDBT. He recently received the Outstanding PC Award from KDD 2025. He has served as a Senior Program Committee member for KDD 2026 and 2025, and CIKM 2020, and has mentored PhD students at several conferences, such as EDBT and CIKM.