Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Sparse Graphical Linear Dynamical Systems
Authors: Emilie Chouzenoux, Victor Elvira
JMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental validation on various synthetic data showcases the effectiveness of the proposed model and inference algorithm. This work will significantly contribute to the understanding and utilization of time-series data in diverse scientific and engineering applications where incorporating a graphical approach is essential to perform the inference. ... We then perform an extensive experimental validation of the proposed model and inference algorithm by means of experiments on various synthetic datasets. |
| Researcher Affiliation | Academia | Emilie Chouzenoux EMAIL Center for Visual Computing Inria, University Paris Saclay 91190 Gif-sur-Yvette, France Victor Elvira EMAIL School of Mathematics University of Edinburgh EH9 3FD Edinburgh, UK |
| Pseudocode | Yes | Algorithm 1 Kalman Filter... Algorithm 2 RTS Smoother... Algorithm 3 DGLASSO algorithm... Algorithm 4 Proximal splitting method to solve (33)... Algorithm 5 Proximal splitting method to solve (34) |
| Open Source Code | Yes | For reproducibility purpose, the code for DGLASSO algorithm, is made publicly available.2 [Footnote 2: https://pages.saclay.inria.fr/emilie.chouzenoux/Logiciel.html] ... The code is publicly available, for reproducibility purpose.6 |
| Open Datasets | Yes | We now evaluate our method on synthetic datasets arising from causal graph discovery studies in the field of weather variability tracking. Specifically, we consider two sets of 200 sparse matrices A RNx, with Nx = 5 or 10 respectively, representing the ground truth causal graphs used to produce WEATH datasets in the Neurips 2019 data challenge (Runge et al., 2020). |
| Dataset Splits | No | We build test time series (xtest, ytest), not seen by the algorithms, constructed by running the ground truth LG-SSM (i.e., with ground truth matrix parameters (A , P , Q )). |
| Hardware Specification | Yes | All codes are run on a Desktop Dell Latitude computer, with 11th Gen Intel(R) Core(TM) i7-1185G7 at 3.00GHz, equipped with 32Go Ram, using Matlab R2021a software. |
| Software Dependencies | Yes | All codes are run on a Desktop Dell Latitude computer, with 11th Gen Intel(R) Core(TM) i7-1185G7 at 3.00GHz, equipped with 32Go Ram, using Matlab R2021a software. |
| Experiment Setup | Yes | We set K = 103, Rk = σ2 RId Ny for every k {1, . . . , K}, µ0 RNx is a vector of ones, Σ0 = σ2 0Id Nx with (σR, σ0) = (10 1, 10 4). Matrix Hk is set to identity matrix for every k {1, . . . , K}, so that Nx = Ny. ... In all the experiments, we set the precision parameters in DGLASSO algorithm to (ξ, ε) = (10 3, 10 3), with a maximum number of 50 iterations for the outer loop, and 20000 iterations for the inner solvers. ... Moreover, the DGLASSO stepsize parameters are set to (θA, θP ) = (1, 1)... We initialize DGLASSO with P(0) = 10 1Id Nx, and A(0) equal to a stable auto-regressive order one matrix with entries A(0)(n, m) = (10 1)|n m| projected onto the set of matrices with spectral norm equal to 0.99. ... The hyperparameters (λA, λP ) for DGLASSO, and λA for GRAPHEM, are finetuned on the rough grid {1, 5, 10}, to minimize RMSE( b A, A ), on one example randomly chosen, per each dataset. |