This is the implementation of [1,2,3], that is a partially hidden autoregressive Markov model (ARPHMM). Partially hidden, partially supervised, weakly hidden, weakly supervised, weak prior, soft labels, noisy labels, uncertain and imprecise labels, etc focus on taking account of a prior on the latent space, here for an ARHMM.
The code provided allows to reproduce the results of the paper for the simulated data.
- Download https://github.com/emmanuelramasso/CHMM_with_partial_labels
- Run example in
testARPHMM_salves_silmulees.m
The idea of including partial knowledge on states is from [1] and [2] applied to prognostics, while its application to acoustic emission signals is from [3].
[1] On partially supervised learning and inference in dynamic Bayesian networks for prognostics with uncertain factual evidence: Illustration with Markov switching models, Pablo Juesas, Emmanuel Ramasso, Sébastien Drujont, Vincent Placet, Proceedings of the European Conference of the PHM Society, Vol. 3 No. 1 (2016), https://doi.org/10.36001/phme.2016.v3i1.1642
[2] Autoregressive Hidden Markov Models with partial knowledge on latent, space applied to aero-engines prognostics, Pablo Juesas, Emmanuel Ramasso, Sébastien Drujont, Vincent Placet, arxiv https://arxiv.org/abs/2105.00211, 2024.
[3] Ramasso, E., Butaud, P., Jeannin, T., Sarasini, F., Placet, V., Godin, N., ...Gabrion, X. (2020). Learning the representation of raw acoustic emission signals by direct generative modelling and its use in chronology-based clusters identification. Eng. Appl. Artif. Intell., 90, 103478. doi: 10.1016/j.engappai.2020.103478