For activity detection on biomedical time-series data,
biomedical signals, e.g., surface electromyography (sEMG) data, systems are modeled as a switching linear dynamical
system with random variables, including discrete and continuous

We present a formalism for representing a system’s
joint probability density function as a hybrid factor graph:

Graphical, factorized representation of temporal joint probability density functions of many variables
Gaussian and discrete distributions

We show that the formalism can be successfully applied to detect
activities based on sEMG data. The modularity of factor graphs enables the straightforward adoption and extension of the formalism expanding its scope of application to other domains.

Effectively represent switching behavior in systems
Combination of discrete and continuous variables required
State (discrete): active/inactive; Observations (continuous): sEMG signal
Realtime exact inference algorithm for estimating state from observations in general hybrid factor graphs

Exact inference algorithm based on tree-based belief propagation using dedicated message passing patterns
Inference results yield activity estimations in terms of probability distributions instead of binary decisions

M. Stender, J. Graßhoff, T. Braun, R. Möller and P. Rostalski, "A Hybrid Factor Graph Model for Biomedical Activity Detection", In: 2021 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI) (IEEE BHI 2021), July 2021.

M. Stender, M. Hartwig, T. Braun, R. Möller, „Increasing State Estimation Accuracy in the Inference Algorithm on a Hybrid Factor Graph Model”, In:  Proceedings of the 35rd International Florida Artificial Intelligence Research Society Conference (FLAIRS-35), 2022, AAAI, North Miami Beach, Florida, USA, May 15-18, 2022.