Scaling Factorial Hidden Markov Models: Stochastic Variational Inference without Messages
Authors: Yin Cheng Ng, Pawel M. Chilinski, Ricardo Silva
NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the validity of our algorithm and the scalability claim with experiments using real and simulated data. |
| Researcher Affiliation | Academia | Yin Cheng Ng Dept. of Statistical Science University College London y.ng.12@ucl.ac.uk; Pawel Chilinski Dept. of Computing Science University College London ucabchi@ucl.ac.uk; Ricardo Silva Dept. of Statistical Science University College London r.silva@ucl.ac.uk |
| Pseudocode | No | The paper describes algorithmic steps in prose but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions using an automatic differentiation tool (Autograd) but does not state that the code for the described methodology is open-source or provide a link to it. |
| Open Datasets | Yes | Bach Chorales Data Set [16] and Household Power Consumption Data Set [16] are used, with reference [16] being "M. Lichman. UCI machine learning repository, 2013." |
| Dataset Splits | No | The paper specifies training and test data (e.g., "The training and testing data consist of 30 and 36 sequences from the Bach Chorales data set respectively." and "...keep the first 10^6 data points... for training and set aside the remaining series as test data."), but does not provide explicit details for a separate validation split. |
| Hardware Specification | Yes | Given fixed computing budget of 2 hours per sequence on a 24 cores Intel i7 workstation |
| Software Dependencies | No | The paper mentions using "Python automatic differentiation tool" and "Rmsprop" for optimization, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | The recognition networks in the experiments have 1 hidden layer with 30 tanh hidden units, and rolling window size of 5. The rolling window size is set to 21 and we allow the algorithm to complete 150,000 SGD iterations with 10 subchains per iteration before terminating. |