Wasserstein Distributionally Robust Kalman Filtering
Authors: Soroosh Shafieezadeh Abadeh, Viet Anh Nguyen, Daniel Kuhn, Peyman Mohajerin Mohajerin Esfahani
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We showcase the performance of the proposed Frank-Wolfe algorithm and the distributionally robust Kalman filter in a suite of synthetic experiments. All optimization problems are implemented in MATLAB and run on an Intel XEON CPU with 3.40GHz clock speed and 16GB of RAM, and the corresponding codes are made publicly available at https://github.com/sorooshafiee/WKF. [...] Figure 5 shows the empirical mean square error 1/500 P500 j=1 xj t ˆxj t 2 across 500 independent simulation runs, where ˆxj t denotes the state estimate at time t in the jth run. |
| Researcher Affiliation | Academia | Soroosh Shafieezadeh-Abadeh Viet Anh Nguyen Daniel Kuhn École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland {soroosh.shafiee,viet-anh.nguyen,daniel.kuhn} @epfl.ch [...] Peyman Mohajerin Esfahani Delft Center for Systems and Control, TU Delft, The Netherlands P.Mohajerin Esfahani@tudelft.nl |
| Pseudocode | Yes | Algorithm 1 Bisection algorithm to solve (7b) [...] Algorithm 2 Frank-Wolfe algorithm to solve (5) [...] Algorithm 3 Robust Kalman filter at time t |
| Open Source Code | Yes | All optimization problems are implemented in MATLAB and run on an Intel XEON CPU with 3.40GHz clock speed and 16GB of RAM, and the corresponding codes are made publicly available at https://github.com/sorooshafiee/WKF. |
| Open Datasets | No | The paper conducts synthetic experiments where data is randomly generated (e.g., 'randomly generate two covariance matrices'), rather than using a publicly available or open dataset. Therefore, no concrete access information for a public dataset is provided. |
| Dataset Splits | No | The paper describes running '10^4 simulation runs' or '500 independent simulation runs' in its synthetic experiments, but it does not specify explicit training/validation/test dataset splits (e.g., percentages or sample counts) nor does it reference predefined splits with citations for a static dataset. |
| Hardware Specification | Yes | All optimization problems are implemented in MATLAB and run on an Intel XEON CPU with 3.40GHz clock speed and 16GB of RAM |
| Software Dependencies | No | The paper states that 'All optimization problems are implemented in MATLAB', but it does not specify a version number for MATLAB or any other software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | In all numerical experiments we simulate the different filters over 1000 periods starting from ˆx0 = 0 and V0 = I2. [...] As for the filter design, the Wasserstein and KL radii are selected from the search grids {a 10 1 : a {1, 1.1, , 2}} and {a 10 4 : a {1, 1.1, , 2}}, respectively. Figure 5 reports the results with minimum steady state error across all candidate radii. [...] the algorithm is stopped as soon as the relative duality gap F(Sk) Sk, f(Sk) /f(Sk) drops below 0.01%. |