An Algorithm for Learning Switched Linear Dynamics from Data
Authors: Guillaume Berger, Monal Narasimhamurthy, Kandai Watanabe, Morteza Lahijanian, Sriram Sankaranarayanan
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we will describe an evaluation of our approach meant to answer two key questions: (a) Do the theoretical guarantees translate into superior empirical performance when compared to highly optimized MILP solvers? (Namely, we compare our approach against the MILP solver Gurobi [17] over a set of microbenchmarks of varying dimensions, number of modes and data sizes.); (b) Does the approach yield interesting results on real-life datasets? |
| Researcher Affiliation | Academia | Guillaume Berger Monal Narasimhamurthy Kandai Watanabe Morteza Lahijanian Sriram Sankaranarayanan University of Colorado Boulder, Boulder, CO, USA firstname.lastname@colorado.edu |
| Pseudocode | Yes | Algorithm 1: Overall algorithm for switched linear system identification and Algorithm 2: Algorithm to expand a non-terminal leaf node in the tree. |
| Open Source Code | No | The paper's 'Questions for Paper Analysis' section states 'Yes' to including code in supplemental material, but the main body of the paper does not provide an explicit URL, repository link, or clear statement like 'We release our code' for the described methodology. |
| Open Datasets | No | The paper states: 'We generate our own handwriting dataset by having an author trace out the letters a , b , c and d using their fingers on the mouse pad of their laptop.' but does not provide any specific link, DOI, repository name, or formal citation for this or any other dataset used in the experiments. |
| Dataset Splits | No | The paper mentions a 'held-out test dataset' but does not specify exact split percentages or sample counts for training, validation, or test sets. |
| Hardware Specification | Yes | All timings are reported in seconds on a Linux server running Ubuntu 22.04 OS with 24 cores and 64 GB RAM. |
| Software Dependencies | Yes | We implemented the proposed approach in Python 3.8, using Gurobi [17] to encode and solve linear programs. |
| Experiment Setup | Yes | We apply our approach to learn m = 3 matrices that fit the data with ϵ = 0.05 and τ = 0.1. and We then identified the dynamics at each mode using the proposed approach (with m = 3, τ = 0.01, and ϵ = 0.01). and We then identified the dynamics at each mode using the proposed approach (with m = 3, τ = 0.1, and ϵ = 0.1). |