Probabilistic Modelling  

The information age is characterized by the abundance of data, lots of it. It comes from social networks, CCTV cameras, sensors and on-line retailers, and the main challenge is to extract useful information from the deluge of available data. As data is inherently noisy, carries a level of uncertainty and is of large scale one typically relies on computationally-efficient probabilistic methods.

 Probabilistic Modelling
To make sense of data one aims to devise a tractable sound probabilistic model that provides insight into the underlying information that gives rise to the data, and means to forecast and predict future behavior. Examples of probabilistic model questions are: the identification of diseases from recorded symptoms; the classification of hand-written numerals into digits; forecasting the type of movies one prefers from past choices; the interests of an individual from on-line shopping records; and forecasting sale figures from past data.

Probabilistic modelling offers computationally-efficient optimization methods for model construction and for solving difficult problems, based on localized models. It also offers a range of methods for providing the most informative presentation of high-dimensional data.  

 Probabilistic Modelling  1

This module introduces students to the main concepts of probabilistic modelling, inference techniques, the projection of probabilities onto graphs, creating complex models from simpler building blocks and the ability to infer values from localized and computationally efficient operations. More specifically it includes:

  • Probabilities and Bayesian methods 
  • Optimizing models via likelihood maximization 
  • Graphical models 
  • Message passing as an inference and optimization tool 
  • Complex distributions as mixture models 
  • Simple visual informatics tools  
 Probabilistic Modelling  2
The module provides the analytical basic tools needed for understanding, modelling and forecasting data and is highly useful for anyone working with data. 
 Probabilistic Modelling  3