Dear CICI PI: A Letter from the OSDF

PI Yushun Dong discusses the MLaaS project: Secure Machine Learning as a Service for Collaborative Scientific Research. The project aims to establish a secure framework to protect the three stakeholders in collaborative scientific computing environments: the model owner, the data provider, and the model user.

PI Zhe Zhao discusses the project FoodGuard: Enabling a Safe and Directive Multi-modal Foundation Model Ecosystem for Food Science Research. The project aims to provide guardrails for safe usage in scientific collaborations, and to integrate with critical food science applications.

PI Qian Wang discusses the QURE project: Usable and Attack-Resistant Security Framework for Quantum Emulators. The project aims to ensure emulation platforms are trustworthy, reliable, and secure, in order to facilitate scientific research.

PI Justin Cappos discusses the GRISL project: Protecting and Hardening Scientific Use of Software Libraries with GRISL. The project aims to harden unsafe scientific libraries without changing scientists’ code, and to use lightweight userspace isolation to prevent crashes and data corruption, in order to improve the reliability of scientific workflows.

PI Michela Taufer discusses the SAFARI project: Scientific Analytics, Forensics, and Reproducibility for Workflows in Cyberinfrastructures. The project aims to integrate forensic data analytics into the Pegasus Workflow Management System to enhance the trustworthiness, reusability, and reproducibility of scientific workflows.

PI Mu Zhang discusses the HPCSafeChain project: Software Supply Chain Security in High-Performance Computing: Understanding, Evaluation and Transition. Scientific computing relies on open-source software vulnerable to software supply chain (SSC) attacks. This project aims to apply SSC security research to high-performance computing, an area that is currently under-explored.

PI Yanan Guo discusses the SecGPU4AI project: Securing GPU Computing for AI-Driven Scientific Workflows. The objective of the project is to bolster the security of scientific AI research against GPU memory safety vulnerabilities.

PI Wei Zhang discusses the SafeSci-TEE project: Advancing Security in TEE-Enabled Scientific Research Workflows: A Holistic Approach. Scientific data is increasingly the target of attackers. This project aims to develop comprehensive security for scientific workflows on untrusted HPC infrastructure, with the ultimate goal of safeguarding data, and promoting trust for multi-institutional collaborations.

PI Wajih Ul Hassan discusses the MLDL project: Multi-Layer Data Provenance and Federated Learning for Securing Scientific AI Pipelines. The project aims to build an end-to-end provenance infrastructure that tracks the full dataset lifecycle. This would enable transparency and accountability, and allow for trustworthy, reproducible AI-driven science.

Co-PI Eunsuk Kang discusses the CloudSec project: Collaborative Policy Alignment for Secure Scientific Computing Infrastructures. The project is positioned as a collaborative security policy analysis for research cyberinfrasturcture. It includes a novel interface designed to elicit policy requirements and explanations from project stakeholders.

PI Yuan Tian discusses the Dprov project: A Data Provenance Framework for Medical Machine Learning. In this talk Tian reviews the needs for data integrity, provence and authenticity. Integrity involves detecting public ML models trained on corrupt data, Provence involves establishing a Standardize efficient, reproducible dataset-model tracking. Authenticity involves removing patient or corrupt data from models effectively. Tian further discusses data set requirements.

PI Teodora Baluta discusses the SCRYPTS-AI project, whose focus is to design solutions for collaboration across disciplines and develop new insights and approaches for distributed, sensitive and private or noisy data sets using AI.

PI Ka Pui (Ricky) Mok discusses the CANIS project: Curated AI-ready Network telescope datasets for Internet Security. Mok shares that CANIS is a suite of modules to improve the USDS-NT infrastructure for acquisition, processing, & analytics of cybersecurity research workflows.

PI Francesco Restuccia discusses the REPAIRT project: Securing XApps in Open RANs with Reliable and Principled AI Red-Teaming. The approach involves algorithms for proactive and dynamic defense from adversarial AI.

PI Ying Wang discusses the WRAP project: Programmable Wireless Infrastructure with Formal Assurance for Cross-Campus Research. Wang talks about how formal assurance translates researcher goals into verifiable policies and how runtime anomaly detection ensures continuous, secure operation.

Brian Bockelman, PI of the Pelican Platform, describes how the OSDF can partner with CICI PIs (and other producers of research datasets) to share their data and build a community of users, using the OSDF as a platform that can control data access, track data access and distribute data to individuals and computing systems.

Ilkay Altintas, PI of the National Data Platform, shares how the NDP serves as a central organizing point for different datasets by providing cataloguing and search functionality.