Matlab reproducible experiments from the following paper:
Dmitrii Ostrovskii, Zaid Harchaoui. Efficient First-Order Algorithms for Adaptive Signal Denoising. ICML 2018.
We use AdaFilter codes for the efficient implementation of adaptive signal denoising, including them as a submodule.
We also use CVX software for disciplined convex programming.
- Make sure that CVX is installed on your computer (commercial license is not needed). CVX installation instructions can be found here.
- Download or clone the repository, and add the following path in MATLAB:
AlgoRec/AdaFilter/code
The experiments, in the order of appearance in the paper, are launched via the following MATLAB commands:
exp_perf_MP_random(N,ifReproduce);
exp_perf_MP_coherent(N,ifReproduce);
exp_certificates(N,ifReproduce);
exp_complexity(N,ifReproduce);
exp_sigm(N,ifReproduce);
Plots will appear in folders plots-<...>
, where <...>
corresponds to the experiment.
Simulation data for the first four experiments will appear in folder sims-perf
, and for the last experiment in folder sims-sigm
.
-
Running a script with
ifReproduce = 1
will first launch simulations, and then produce plots for the obtained data. After that,ifReproduce = 0
can be used to produce the plots without running the simulations again. -
N
is the number of Monte-Carlo trials. To reproduce the results from the paper, one must setN = 20
forexp_sigm
andN = 10
in all other cases. Smaller values ofN
can be used to obtain faster (and less precise) results.