Probabilistic Programming -- Practical Experiences and Challenges in Voucher Processing

Ahmad Al-Sibahi
University of Copenhagen and

Probabilistic programming is a paradigm that augments traditional programming with constructs from Bayesian probabilistic modelling including random variables, distributions, sampling and conditioning. Compared to traditional machine learning techniques, it has three advantages: domain knowledge can be encoded directly, models are more easily interpretable, and it is possible to quantify the uncertainty in predictions.

In my talk, I will provide a short introduction to probabilistic programming and discuss how we use the technique at I will present two concrete models which we have been working on: one for grouping similar vouchers and another for matching keywords against features. We will discuss the high-level constructs used in these models, the challenges we experienced in describing and running inference for them, and finally the preliminary results of the inference. 

Ahmad Al-Sibahi obtained his Ph.D. in computer science from the IT University of Copenhagen.  He is presently employed as a researcher at and postdoc at DIKU, the Department of Computer Science, University of Copenhagen, working on probabilistic programming. 

HostFritz Henglein (DIKU, tel. +45-30589576)

All are welcome. No registration required.