NEMECYS Partner Presents Research at the 16th Asian Conference on Machine Learning

On 6th December 2024, Noam Koren (IBM Research and the Technion) and Abigail Goldsteen (IBM Research) participated in the 16th Asian Conference on Machine Learning (ACML). The conference provided a platform for researchers to present innovative developments in the field of machine learning, including work with significant implications for the medical domain.

During Oral Presentation Session #12, chaired by Rui Luo, Noam Koren presented the IBM Research team’s paper, titled:
Membership Inference Attacks Against Time-Series Models

Research Focus

The paper, authored by Noam Koren, Abigail Goldsteen, Guy Amit, and Ariel Farkash, addresses a pressing conflict in the medical domain: the increasing reliance on personal data for diagnostics and treatment versus the critical need to safeguard patient privacy. The study investigates vulnerabilities in time-series prediction models, a key component in medical machine learning, and demonstrates how these models are susceptible to membership inference attacks due to the unique characteristics of their prediction vectors.

The researchers proposed new features and combinations specific to time-series models to enhance the accuracy of these attacks, highlighting the need for rigorous privacy assessments when deploying machine learning models in sensitive applications such as healthcare.

This significant contribution from the NEMECYS partner IBM underscores the importance of ensuring privacy in machine learning applications and provides valuable insights for addressing security challenges in the healthcare sector.