Self Supervised Learning of Embeddings for Semi-Supervised Track Classification

Total::Insight™ is a geospatial and time distribution Decision Support System (DSS) that includes a correlator capable of consuming many sporadic time domain signals and converting them into feature rich tracks. The problem here is how to create an AI/ML embedding that is domain relevant from the tracks.The project objective is to develop, refine and industrialize new AI/ML-enabled track analysis capabilities for the Total::Insight product. The development is challenging as, in most big data problems, the data is unstructured, sparse, unlabelled and diverse. The project scope is to develop new descriptive features, include time dependent track segmentation, develop preliminary time series foundational models, and perform at scale engineering to generate embeddings. The project outcome will enable Total::Insight™ to efficiently perform efficient classification and similarity searches of tracks with limited labeled data.
For this project, we will budget 3 IUs over 4 months, in addition to 4 Larus full-time senior and intermediate employees.

Faculty Supervisor:

Emil Petriu

Student:

Partner:

Larus Technologies

Discipline:

Engineering

Sector:

Artificial Intelligence; Aerospace; Advanced Manufacturing

University:

University of Ottawa

Program: