2013. március 29.

Probabilisztikus programozás

Az utóbbi napokban a Twitter a DARPA Probabilistic Programming for Advanced Machine Learning (PPAML) Proposers' Day-től hangos. De miért? Rob Zinkov Why Probabilistic Programming Matters posztjában így válaszolja meg a kérdést

Probabilistic programming is a newer way of posing machine learning problems. As the models we want to create become more complex it will be necessary to embrace more generic tools for capturing dependencies. I wish to argue that probabilistic programming languages should be the dominant way we perform this modeling, and will demonstrate it by showing the variety of problems that can be trivially modeled with such a language.
Probabilistic programming also has the potential to give machine learning to the masses by making it very easy to specify realistic models for frequently heterogenous data. Too often, simple models are used because they are popular and implementations are freely available. By shifting focus to a language we remove these artificial constraints.


Nincsenek megjegyzések: