Inertial Hidden Markov Models: Modeling Change in Multivariate Time Series
Authors: George Montanez, Saeed Amizadeh, Nikolay Laptev
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | we derive and test two methods of regularizing hidden Markov models for this task. Our methods are compared with a recent hierarchical Dirichlet process hidden Markov model (HDP-HMM) and a baseline standard hidden Markov model, of which the former suffers from poor performance on moderate-dimensional data and sensitivity to parameter settings, while the latter suffers from rapid state transitioning, over-segmentation and poor performance on a segmentation task involving human activity accelerometer data from the UCI Repository. The regularized methods developed here are able to perfectly characterize change of behavior in the human activity data for roughly half of the real-data test cases, with accuracy of 94% and low variation of information. |
| Researcher Affiliation | Collaboration | George D. Monta nez Machine Learning Department Carnegie Mellon University Pittsburgh, PA USA gmontane@cs.cmu.edu; Saeed Amizadeh Yahoo Labs Sunnyvale, CA USA amizadeh@yahoo-inc.com; Nikolay Laptev Yahoo Labs Sunnyvale, CA USA nlaptev@yahoo-inc.com |
| Pseudocode | No | The paper contains mathematical derivations and descriptions of algorithms but no explicit section or figure labeled "Pseudocode" or "Algorithm". |
| Open Source Code | No | The paper mentions a "publicly available HDP-HMM toolbox for MATLAB" used for comparison, but it does not provide source code for the methods developed in this paper. |
| Open Datasets | Yes | The second dataset was generated from real-world fortyfive dimensional human accelerometer data, recorded for users performing five different activities, namely, playing basketball, rowing, jumping, ascending stairs and walking in a parking lot (Altun, Barshan, and Tunc el 2010). |
| Dataset Splits | No | The paper describes how synthetic and real datasets were generated and used for evaluation (test cases), but it does not specify explicit training, validation, and test splits with percentages or sample counts in the traditional sense for its own methods. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments. |
| Software Dependencies | No | The paper mentions "HDP-HMM toolbox for MATLAB" but does not specify version numbers for either the toolbox or MATLAB, which is insufficient for reproducibility. |
| Experiment Setup | Yes | Parameter selection for the inertial HMM methods was done using the automated parameter selection procedure described in the Parameter Modifications section. For faster evaluation, we ran the automated parameter selection process on ten randomly drawn examples, averaged the final ζ parameter value, and used the fixed value for all trials. The final ζ parameters are shown in Tables 1 and 2. ... All HMMs were equipped with Gaussian emission models with full covariance matrices. |