Promoting Women's Participation in Data Science
We are pleased to announce an upcoming event dedicated to promoting women's participation in Data Science and related Ph.D. programs.
Please find the program of the event below:
15.00 Welcome address
15.10 A Theoretical Analysis of Recommendation Loss Functions under Negative Sampling
Dr. Giulia Di Teodoro
Bio:
Giulia Di Teodoro is a postdoctoral researcher at the University of Pisa, within the Information Engineering Department. She earned her degree in Management Engineering with honors from Sapienza University of Rome, Italy, where she also completed her PhD in Data Science in 2024. Her doctoral thesis focused on precision medicine for HIV and diabetes as well as the interpretability of machine learning models. Giulia has publications in the fields of Precision Medicine, Explainable Artificial Intelligence, and Bioinformatics. Her research interests include Mixed-Integer Linear Programming (MILP), Precision Medicine, and Recommendation Systems.
Abstract: Recommender Systems (RSs) are pivotal in diverse domains such as e-commerce, music streaming, and social media. This work conducts a comparative analysis of prevalent loss functions in RSs: Binary Cross-Entropy (BCE), Categorical Cross-Entropy (CCE), and Bayesian Personalized Ranking (BPR). Exploring the behaviour of these loss functions across varying negative sampling settings, we reveal that BPR and CCE are equivalent when one negative sample is used. Additionally, we demonstrate that all losses share a common global minimum. Evaluation of RSs mainly relies on ranking metrics known as Normalized Discounted Cumulative Gain (NDCG) and Mean Reciprocal Rank (MRR). We produce bounds of the different losses for negative sampling settings to establish a probabilistic lower bound for NDCG. We show that the BPR bound on NDCG is weaker than that of BCE, contradicting the common assumption that BPR is superior to BCE in RSs training. Experiments on three datasets and two models empirically support these theoretical findings.
15.50 Doing research in the industry and a path towards academia.
Dr. Aleksandra Piktus
Bio:
Aleksandra is currently a second year PhD student in data science at Sapienza and a research engineer working on large language models at Cohere. Before that she was a researcher at HuggingFace and the Facebook AI Research lab in London. She started her career as a software engineer at Facebook where she worked on reducing the spread of misinformation on the platform and on search. Her interests include knowledge-intensive NLP, retrieval augmentation and LLM pre-training data exploration.
16.30 Engineering...? why?
Prof. Marilena Venditelli
Bio:
Marilena Vendittelli obtained her PhD in Systems Engineering in 1997 from the University of Rome "La Sapienza". She won two Marie Curie Research Training Grants in 1996 and 1998, respectively, for research on motion planning and control of nonholonomic systems during her postdoctoral period at LAAS-CNRS in Toulouse (France). From 1998 to 2016 she was at DIAG, first as a post-doc student and then as Assistant Professor (2001-2016). From 2017 to 2019 she was an associate professor at the Department of Information, Electronics and Telecommunications Engineering and in 2020 she (re)joined the DIAG. Over the years she has been a Visiting Scholar at Carnegie Mellon University (2005), the Courant Institute of New York University (2012), the Simons Institute of UC Berkeley (2016).
Abstract: The most common reaction a girl encounters when she says she wants to study engineering at university is one of surprise, sometimes mixed with admiration, followed by the question, "Why?" In this talk, I will humorously revisit the beginnings of my engineering career and the non-technical challenges a girl faces in this field. Then, I will briefly present some of my recent engineering projects, which are the result of my research in the field of robotics.