93469 - Human Data Science

Learning outcomes

This course aims to demonstrate ways in which human-centric perspectives lead to new approaches to analyzing, evaluating and understanding machine learning (ML) models. For instance, a conventional focus on traditional computational criteria may be insufficient to identify suitable ML-based systems. Instead, reflecting on the training data and their role and quality, as well as relying on tight action-feedback loops that engage humans, may bring to modified model behaviors that are beneficial to users, helping them to understand those models as more than simple black boxes. Misuses of big data, lack of specific attention to their quality and absence of humans in the loop are discussed with the aim to understand what makes Data Science different.

Course contents

Human data science integrates the study of data science with breakthroughs in humans and algorithms to advance our ecosystem of knowledge, and help everyone make better, more insightful decisions. Case studies will be presented to illustrate the folowing:

 

- Datification and interaction with special data (e.g., multimedia)

- Human-Machine interaction loop. Interactive Interfaces. Bigdata-human interaction

- Human-assisted Bigdata analysis

- Human-Machine collaboration in Bigdata

- Models for humans and machines / Machine learning

- Human factors and ELSI (ethical, legal and social issues) in human-in-the-loop systems

- Human-machine collaboration in real-world problems with applications including: health, transportation, cultural heritage, entertainment etc

Readings/Bibliography

Scientific papers and reports delivered by the Lecturer during the course

Teaching methods

Class lectures and projects

Assessment methods

Aim of the examination process is to assess if adequate skills have been acquired by students in this specific field. The exam typically proceeds as follows. The lecturer proposes, also with the help of seminars delivered by external guests, possible case studies in the field of human data science. Students choose one of those cases and develop it until a correspondent system is developed. At the end, students are required to deliver a public talk where the developed system is analyzed and its characteristics discussed.

Teaching tools

Departmental networked lab and applications 

Office hours

See the website of Marco Roccetti