North America Research Interest Group Meeting [November 2022]

View recording of meeting

Meeting Agenda:
Collaboratory Student Research Projects

Boston University students will present their 2022 Red Hat Collaboratory Student Research Projects. Learn more here. Additional student projects will be presented during the November 4 Boston University Systems Seminar.

Speaker: Shun Zhang 
Project Title: D-COLLECTIVE: Democratized Data Collection and Collaborative Training for Extreme-Scale Autonomous Systems
Mentor: Eshed Ohn-Bar, Assistant Professor, College of Engineering
Abstract: Learning and training from supervised and unsupervised datas are now trending, but privacy has become a problem with none open-source resources. We here provide federated learning ways to train both supervised and unsupervised datas with less decrease of the performance. Our goal is to enable everyday people to collaboratively develop robust, generalized, accessible and easily deployable decision-making policies for navigation. Through large-scale coverage, people-powered fleets enable collecting large amounts of rare edge-case data, which can also benefit various companies and organizations.

Speaker: Nengneng Yu
Project Title: Transformer-based advanced persistent threats (APT) detection system
Mentor: Dr. Alan Liu, Assistant Professor, College of Engineering
Abstract: An Advanced Persistent Threat, which we also call the APT attack, is a targeted attack frequently involving nation-state actors toward the government or other large occupations that take place over a long time scale. APT’s low and slow attack patterns fundamentally differ from conventional one-shot attacks. Consequently, traditional detection techniques are not well suited to APT. We are planning to build a Transformer-based detection system that can learn patterns of APT attacks generally and effectively in real-world scenarios and possibly deploy to servers to detect and prevent future APT attacks.

Speaker: Julia Hua
Project Title: Network-accelerated In-memory Key-Value Store Live Migration
Mentor: Dr. Alan Liu, Assistant Professor, College of Engineering
Abstract: Live migration—utilized to attain better spatial locality and load balancing—is a crucial component for cloud systems to achieve high-throughput and low-latency demands. More specifically, key-value store live migration migrates key-value pairs and client requests from a source machine, which is typically overloaded, to a destination machine, which is typically idle. There are three main existing approaches: source-based, destination-based, and hybrid approaches. In the source-based approach, the source machine takes client requests while incrementally migrating its key-value pairs to the destination machine. In the destination-based approach, the destination machine takes client requests while pulling key-value pairs from the source machine. In the hybrid approach, client requests are sent to both machines while key- value pairs migrate. With Prof. Alan Liu and PhD student Zeying Zhu at Boston University, we developed a migration system prototype, which used the hybrid approach and is implemented on the in-memory key-value store Redis.