We will discuss recent results in denoising of continuous-valued data.
Our starting point is a recent universal denoiser for discrete data (DUDE) by Weissman et. al.
We extend the approach underlying the construction of the DUDE to our setting of continuous-valued symbols using nonparametric density estimation techniques.

Our approach also provides an alternative to the DUDE for denoising large-alphabet
discrete data, by viewing the latter as continuous-valued. This circumvents
some of the computational and statistical difficulties encountered when employing the DUDE in practice. Our denoising scheme, when applied in this setting, admits an intuitive 'context aggregation' interpretation,in addition to being computationally feasible. We obtain performance bounds and asymptotic universal optimality guarantees for the proposed family of schemes that are analogous to those derived for the DUDE.

Initial experimentation seems to hint at the potential of the proposed schemes for doing well on real data.