Unsupervised Neural Hidden Markov Models with a Continuous latent state space
This work addresses the need for flexible and interpretable models in sequence analysis, though it appears incremental as it builds on existing HMM and neural network methods.
The paper tackles the problem of modeling sequences with underlying latent variables by introducing an unsupervised neural Hidden Markov Model with a continuous latent state space, achieving comparable performance to existing neural architectures like LSTMs and GRUs while providing interpretable results.
We introduce a new procedure to neuralize unsupervised Hidden Markov Models in the continuous case. This provides higher flexibility to solve problems with underlying latent variables. This approach is evaluated on both synthetic and real data. On top of generating likely model parameters with comparable performances to off-the-shelf neural architecture (LSTMs, GRUs,..), the obtained results are easily interpretable.