Decision Science with Probabilistic Programming

Mattia Ferrini

Data Science Deep Learning Functional Programming Science

See in schedule

Generative Models are the Swiss Army Knife for the Decision Scientist. Generative models allow the simulation of scenarios based on different business hypotheses (Bayesian priors). With Probabilistic Programming, decision makers can simulate the impact of business drivers in times of great uncertainty.

Furthermore, Probabilistic Programming Languages provide all the inference tools necessary to identify the assumptions that have most likely generated an outcome. Inference is a statistical tool that enables optimal decision-making based on models that explicitly quantify uncertainty.

This talk addresses the use of Probabilistic Programming Languages in decision science. The talk will briefly introduce to Bayesian Machine Learning, Bayesian inference and inference algorithms through a number of use-cases developed in Pyro. The use cases will be simple yet will have practical relevance: the examples will illustrate scalability and verifiability challenges.

This talk is tailored to the hands-on practitioner and the sole prerequisite is an understanding of basic statistical concepts.

Type: Poster session (45 mins); Python level: Intermediate; Domain level: Intermediate

Mattia Ferrini

Mattia has over 15 years of experience in Machine Learning and Applied mathematics.
Mattia is a former hedge fund manager, a startup founder and is currently leading the AI and Mathematical Modelling team at a large consulting company.