Tanveer Khan from Tampere University (TUNI) presented HARPOCRATES research at the 29th ACM Symposium on Access Control Models and Technologies (SACMAT 2024), held on 15–17 May 2024 in San Antonio, Texas, USA. The presentation was delivered online.
SACMAT is the premier forum for the research and development of access control, security models, and policy technologies. It brings together the academic community and industry experts to address current challenges and future directions in designing secure and trustworthy systems.
Research Contribution
At the symposium, Tanveer Khan presented the paper:
“Make Split, not Hijack: Preventing Feature-Space Hijacking Attacks in Split Learning.”
The paper focuses on a new class of threats known as feature-space hijacking attacks in the context of split learning. Split learning enables distributed training by dividing computation between clients and servers, but it also introduces vulnerabilities where adversaries can manipulate intermediate representations. This research proposes novel defenses that mitigate such risks, ensuring that the split learning process can remain secure while preserving efficiency.
Connection to HARPOCRATES
The HARPOCRATES project aims to design privacy-preserving and secure machine learning frameworks that can be deployed in sensitive domains such as healthcare and public services. By addressing hijacking risks in split learning, this work directly supports HARPOCRATES’s mission of strengthening trust, confidentiality, and resilience in AI-driven systems.
Further Information
Details on the symposium can be found on the official website: SACMAT 2024