📶 Signal Deconvolution

A sharp signal is observed only after a Gaussian blur and a little added noise. We reconstruct it by choosing its twelve samples to fit the observation, regularised by a smoothness penalty on jumps between adjacent samples. This is the classic regularised least-squares inverse problem: smooth and near-convex, with one well-defined optimum. Score is reconstruction residual plus smoothness penalty, to minimise. 12-D.

Algorithm

Residual + penalty—
Reconstructions tried0
Best so far—

Leaderboard

AlgorithmScoreEvals
— no runs yet —

What's happening

The JS objective is a line-for-line port of example_applications/signal_deconvolution and agrees to floating-point tolerance. The dashed line is the hidden true signal, the cyan line is the blurred observation, and the bars are the optimizer's reconstruction.