Events & Projects

Lecture Held by Associate Professor Eran Toch

On Thursday, April 24, 2025, we hosted a lecture by Associate Professor Eran Toch from Tel Aviv University. Professor Toch serves as the Head of the Industrial Engineering Department within the Faculty of Engineering at the university.

Professor Toch's research focuses on how human-computer interaction and machine learning can help people better manage their online experiences, privacy, and security. His work involves the computational analysis of human behavior and applying this knowledge to enhance people’s online experience, safety, and productivity. His research group has received funding from the EU’s Horizon 2020 program, the Israeli Science Foundation (ISF), DARPA, and the Israel Ministry of Science. He also serves as an editor for the journal IMWUT and as a subcommittee chair for Security & Privacy at the ACM CHI conference. Previously, Professor Toch was a Visiting Associate Professor at Cornell University and a Fellow at Carnegie Mellon University. 

Lecture Held by Professor Eran Toch, McLab
Lecture Held by Professor Eran Toch, McLab

Details of the lecture are as follows:

  • Date & Time: Thursday, April 24, 2025, 15:10–16:40
  • Speaker: Eran Toch (Associate Professor, Head of the Industrial Engineering Department, Faculty of Engineering, Tel Aviv University)
  • Title: How User Privacy Behavior Shapes Machine Learning Models
  • Abstract: Machine learning models, including large language models, play a central role inthe computer systems we build today and in shaping our vision for the future.However, these models fundamentally depend on data collected and shared bypeople—data that often includes personal and sensitive information. In thistalk, I present several studies that examine how privacy concerns and userbehaviors impact the data used to train machine learning models and,consequently, their performance. Through a series of online experimentsinvolving 1,551 participants, we demonstrate that users’ sharing decisions cansignificantly degrade model performance. Paradoxically, we also show thatmechanisms such as differential privacy -- which adds noise to protectindividual data -- can increase user trust and improve data quality and modelaccuracy.


Recent Articles