David Campos of Aalborg University

The widespread availability of digital devices enables the collection of data from multiple environments, from simple tracking of physical activities to the monitoring of industrial processes. When analyzing this time-series data, the use of machine learning models is continually growing, as they often provide better insights, enabled by their large size and complexity. However, deploying these models at the edge, closer to where data is collected, poses a challenge, as they struggle in computationally constrained environments. To overcome this challenge, this work makes the following contributions:

• LightTS: A method for compressing large time-series machine learning models into lightweight versions while maintaining equivalent performance.
• QCore: A strategy for building small, representative subsets of training data to enable the continual on-edge calibration of lightweight models.
• QCore+: An extended data compression strategy that enhances support for additional time-series tasks while streamlining the deployment of lightweight models. 
• TimeBlocks: A framework for disentangling large models into small, modular blocks that can be combined to build lightweight models for multiple tasks.

By reducing model size and data requirements, these contributions enable the efficient deployment of analytic models in computationally constrained environments.

Biography:

David Campos is a PhD Fellow in the Department of Computer Science at Aalborg University. His research primarily focuses on time series analysis. He has already published several papers in top-tier venues, including SIGMOD and VLDB.

Address:Selma Lagerløfs Vej 300, , Aalborg, Arhus Amt, Denmark, 9220