Datum održavanja: četvrtak, 6.2.2020. u 11:00 sati, prostorija O-358
Predavač: prof. dr. sc. Francesco Guerra,  University of Modena and Reggio Emilia, Department of Engineering "Enzo Ferrari"
Naziv predavanja: Machine Learning in production: a Database Perspective

Abstract:

Machine Learning is a mature technology now. In the last years, the research community put a large effort in the development of algorithms and techniques that can address efficiently and effectively ML tasks. This effort led to the release of robust software libraries (e.g., ML.NET, SCIKIT, WEKA) that provide implementations of  functionalities for managing and analyzing the datasets,  and Data Mining / Machine Learning algorithms to get insight from the data. Nevertheless, these libraries do not allow scientists and practitioners to easily manage issues that typically affect  ML systems in production, e.g., the need of guaranteeing valid and clean data in operating conditions. The result is that ML applications have been demonstrated to be highly performing in the development phase but very fragile when operating in real working conditions. The concept of Technical Debt has been introduced in the literature to indicate that  it is easy to incur in massive ongoing maintenance costs at the system level when applying machine learning techniques.This talk aims to analyze the impact of the Technical Debt in executing ML tasks when the data comes from DBMS. The library ML@DB that allows users to perform ML serving tasks natively on DBMS through SQL statements is presented as a possible way to reduce the issues in ML systems in production.