Semantic based Regularization: a generic framework for bridging Machine Learning and Logic

14 dicembre 2016
Versione stampabile

Time: December 14, 2016, h. 02.00 pm
Location: Room Ofek, polo scientifico e tecnologico “Fabio Ferrari”, Building Povo 1, via Sommarive 5, Povo - Trento 

Speaker 

Michelangelo Diligenti, University of Siena & Google

Abstract

Semantic Based Regularization (SBR) is a general Statistical Relational Learning framework that allows to seamlessly integrate transductive or semi-supervised learning with application-specific background knowledge, which is assumed to be expressed as a collection of first-order logic (FOL) clauses. Semantic Based Regularization transforms the learning problem into a set of continuous constraints, which are used both to accommodate the fitting of the supervised data as in classical supervised learning, and to enforce the solution to respect the background knowledge.

All the classical Statistical Relational Learning tasks (learning, inference, collective classification and link prediction) can be coherently solved using this framework, while maintaining state-of-the-art capabilities in processing the low-level feature-based pattern representations that are typical in machine learning problems. This talk will first provide an overview of the framework and it will present SBRS, a freely available software implementation.

In the second part of the talk, we will discuss the computational issues that can make learning and inference hard in SBR and, finally, we will present the strategies that have been developed to break down the complexity of learning and achieve satisfactory results in many real-world applications.

About the Speaker 

Michelangelo Diligenti received the Ph.D. degree in 2002 from the University of Florence, Italy. Currently, he is Assistant Professor of Computer Science at the University of Siena, Italy, where he teaches the course on Software Architecture and Design. Since 2003, he has been collaborating with Google Inc., working on improving the ranking algorithms for web search. His main research interests are in the field of Machine Learning and, in particular, in aspects regarding Learning-to-Rank and Statistical Relational Learning.

Contact person regarding this talk: Andrea Passerini, andrea.passerini [at] unitn.it