WORKSHOP: MACHINE LEARNING for DATA & ALGORITHM-SHARING in SOCIAL SCIENCE
27-28 May 2024
IMT School for Advanced Studies Lucca, Italy
Are recent advances in Machine Learning making the sharing of algorithms preferable to the sharing of data? We are delighted to announce the Workshop on Machine Learning for Data & Algorithm-sharing in social science. The workshop, organized by the IMT School for Advanced Studies Lucca, provides a unique opportunity for researchers and practitioners to delve into the transformative power and challenges of algorithm-sharing methods, including, but not limited to, cutting-edge federated learning. These state-of-the-art methodologies are poised to revolutionize empirical research in social science, particularly in contexts where conventional data-sharing is impeded by privacy, security, and commercial constraints.
The workshop is organized and hosted by the IMT School for Advanced Studies Lucca and is open to up to 11 participants.
For further information and to apply, write us at das_workshop[at]imtlucca[dot]it
KEYNOTE SPEAKERS
Nathalie Baracaldo
Senior Research Scientist, IBM Research
Bruno Carballa Smichowski
Researcher, Joint Research Centre - European Commission
Jermain Kaminski
Assistant Professor in Entrepreneurship and Innovation, Maastricht University
Gabriele Tolomei
Associate Professor of Computer Science, Sapienza University of Rome
Research across various domains increasingly relies on centralized approaches to data analysis, where data generated are transferred from their origins to remote entities for processing. However, this centralized paradigm can sometimes hinder research when data cannot be shared because of privacy concerns or conflicts with EU GDPR standards.
Algorithm-sharing approaches, such as Federated Learning (FL), Decentralized Machine Learning (DML), and Secure Multi-Party Computation (SMPC), offer a promising solution. These approaches enable collaborative training and computation of predictive models and statistics among cooperating nodes, without revealing private local data. By preserving data privacy and reducing communication costs, these technologies can be applied across various domains, including healthcare, finance, education, business, economics, and social networks. They provide a decentralized way to analyze data while respecting privacy and security.
Despite their potential, these methods have seen limited adoption in socio-economic research. Data-sharing remains the dominant paradigm, relegating algorithm-sharing approaches mostly to meta-analyses. This workshop aims to bridge this gap by exploring new research frontiers aligned with GDPR standards and the growing demand for data protection in the EU.
The workshop seeks to bring together international scholars and practitioners interested in exploring the applications of algorithm-sharing in social science, including economics and cognate disciplines. The main aim is to showcase clever solutions to overcome commercial and privacy constraints to the sharing of data, along with algorithm-sharing applications in social science. By doing so, the workshop aims to answer the following questions:
Which are the ideal applications of data and algorithm-sharing, and when is algorithm-sharing preferable to data-sharing?
How and when can data and algorithm sharing facilitate the collaboration between stakeholders, such as researchers, policymakers, and entrepreneurs, and accommodate data privacy, security and commercial concerns?
Which is the state-of-the-art of adopted algorithm-sharing solutions in public and private domains?
The workshop consists of four main formats:
Keynote speeches: we will have three keynote sessions providing insights from leading experts into the current developments of data and algorithm-sharing techniques.
Paper sessions: participants will present and discuss ongoing research projects, engage in constructive comments, and receive feedback from peers and senior scholars.
Roundtable discussion: the workshop will conclude with a roundtable discussion where participants will reflect on the main challenges and opportunities of data vs. algorithm sharing and identify future research directions and collaborations.
Proof of concept session: participants will have the opportunity to familiarize themselves with algorithm-sharing methods and software through a special “proof of concept” session using Federated Learning methods on real data.
Social events: the best ideas are born at coffee and dinner tables. The workshop will include social moments to let participants network and exchange ideas in an informal and relaxed environment.