[Haskell-cafe] Reactive Programming in Machine Learning

[Tom Nielsen]

As far as I am aware, there has been very little work on combining
these two, but that does not mean that it is a bad idea. I can give
you some pointers from a very personal perspective:

-Machine learning is mostly kernel methods and probabilistic
inference, I can't really say much about how one would combine kernel
methods and FRP, but more about probabilistic inference.

-I work in physiology and we use FRP for data acquisition and for
classifying different kinds of evidence. It appears that most data --
observed or inferred -- can be described as signals or as events. We
have a paper about this but it is not yet published.

-We do data analysis by building probabilistic models (i.e.
probability distributions) for the "reactive" objects, i.e. signals or
events, and applying bayesian inference to learn the parameters of
these models. Lots of this work consists of thinking about what would
be a good probability distribution for a signal or an event.

-Gaussian processes make very good models for real-valued signals,
especially when your underlying model is a stochastic differential
equation (some of which can be rewritten as gaussian processes).

-for events we use point processes such as the poisson processes.

-I would love to work on probabilistic reactive control, but don't
really have the time. You could use sequential monte carlo/particle
filters to iteratively estimate the value of unobserved time-varying
quantities and the use FRP-like systems to wire this up to an output
signal. If you want to learn more about sequential monte carlo, there
are lots of videos on videolectures.net. Nando de Freitas has a good
introduction.

Tom

On Sat, Jul 23, 2011 at 2:30 AM, bob zhang <bobzhang1988 at gmail.com> wrote:
> Hi all,
> I am doing a survey on combining Functional Reactive Programming and
> Machine Learning. Has anyone did relevant research on this topic?
> Any discussion or link is appreciable.
> Best,bob
>
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>