Tree-Structured Recurrent Switching Linear Dynamical Systems for Multi-Scale Modeling
Authors: Josue Nassar, Scott Linderman, Monica Bugallo, Il Memming Park
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through a variety of synthetic and real examples, we show how these models outperform existing methods in both interpretability and predictive capability. |
| Researcher Affiliation | Academia | Josue Nassar Department of Electrical & Computer Engineering Stony Brook University Stony Brook, NY 11794 josue.nassar@stonybrook.edu Scott W. Linderman Department of Statistics Columbia University New York, NY 10027 scott.linderman@columbia.edu Mónica F. Bugallo Department of Electrical & Computer Engineering Stony Brook University Stony Brook, NY, 11794 monica.bugallo@stonybrook.edu Il Memming Park Department of Neurobiology and Behavior Stony Brook University Stony Brook, NY, 11794 memming.park@stonybrook.edu |
| Pseudocode | No | The paper describes the inference procedure in narrative text but does not provide a formal pseudocode or algorithm block. |
| Open Source Code | Yes | Source code is available at https://github.com/catniplab/tree_structured_rslds |
| Open Datasets | Yes | Finally, we apply the proposed method on the data from Graf et al. (2011). |
| Dataset Splits | No | The paper specifies training and testing splits for synthetic data, but no explicit validation split is mentioned for any experiment. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory specifications) are provided in the paper for running experiments. |
| Software Dependencies | No | The paper does not list specific version numbers for software dependencies or libraries used (e.g., 'Python 3.8', 'PyTorch 1.9'). |
| Experiment Setup | Yes | We set the number of leaf nodes to be 4 and ran Gibbs for 1,000 samples; the last 50 samples were kept and we choose the sample that produced the highest log likelihood to produce Fig. 2 where the vector fields were produced using the mode of the conditional posteriors of the dynamics. |