There is a growing need for machine learning approaches which combine the representation power of logic with the ability to reason about uncertainty, an area often referred to as statistical relational learning (SRL).  As the size, heterogeneity and diversity of available data grows, the demand for scalable machine learning methods which combine logic and statistics becomes more pressing.

In this talk, I will review my contributions to this area.  I will survey the general area, and then focus attention on two of our recent projects: 1) coupled conditional classifiers (C^3), and 2) probabilistic soft logic (PSL).  I will describe their mathematical foundations, learning and inference algorithms, and empirical evaluation, showing their power in terms of both accuracy and scalability. 

An additional advantage of SRL methods is their representational power and flexibility.  We have applied our tools to a diverse collection of applications problems including: entity resolution, ontology alignment, key-opinion leader identification, sentiment analysis, role identification, personalized recommendation, trust, group affiliation analysis and bioinformatics problems such as splice-site and drug target prediction.  I will reference these problems throughout the talk.

Along with power, comes peril, both in terms of inaccurate conclusions and inference of sensitive information.  I will briefly describe our work on active learning and visual analytics, which helps users overcome limitations in machine learning systems by allowing them to understand and correct potential problems in the input and inferences made.  I will also describe our work on privacy, which highlights the ease with which SRL methods can infer attributes, even if they are explicitly made private, in online social networking sites.