Neural Compression for Multimodal Vehicle Data - Investigating Different Architectures for Neural Compression Models for Multimodal Time Series Data

Sammanfattning

With the growing amount of data generated by modern vehicles, the need for edgeefficient data processing and compression strategies becomes more apparent. As theamount of data grows so does the potential for machine learning models to extractinsightful information from this data at later stages. Achieving high compressionratios while retaining only the essential components of the original data thereforebecomes a primary goal of many compression strategies. Currently, the existingsolutions for such a task on a resource-constrained device such as a vehicle are limitedand often come with tradeoffs. This thesis investigates compression algorithmsbased on neural networks (commonly referred to as neural compression) which haveshown promising results in other fields, such as image processing. Four neural codecarchitectures are explored and tested on industry-relevant datasets with compressionratio, data reconstruction quality and downstream utility retention as the mainmetrics investigated.