Zongshun Zhang(张宗舜)

I am a fith-year CS Ph.D. student at Boston University, advised by Professor Abraham Matta. I am interested in researching cloud resource-orchestration methods, recently focusing on resource cost ($) analysis for machine learning tasks. I received my BS in CS from University of Minnesota, Twin City in 2019. Go Gophers!!

zhangzs AT bu DOT edu  /  CV  /  Google Scholar  /  LinkedIn  /  Github

• I have been studying general resource provisioning methods in the central cloud and edge environments. Recently the focus is on task placement of DL models with Early Exit.
• We are also developing novel architectures for deep neural networks while taking privacy into consideration.
  
• I have the details of my publications and projects in the following table.

(ing...) Container SBOM and vulnerability package detection are getting attention due to the increased cluster complexity.

We are researching a scalable ML-based method to construct an installation log to analyze anomaly caused by package installtions in a PaaS cluster.

(ing...) DL model can be splited over different services, but finding a good partitioning strategy is not a trivial problem, due to extra transmission delay, power consumption and training jobs competition.

Following on LIBRA, different services with different cost models can save money for DL training or inference given QoS guarantees.

(delayed...)Studying cloud resource provisioning techniques in general. Experimenting to enhance Docker container scheduling, scaling, and load balancing performance over k8s with machine learning techniques like NN or SVM to guide resource provisioning.

"It turns out that we don't really need a model that fits very well for systems..."

• (ing...)Researching novel architectures for efficient deep neural networks training considering model and data privacy.

• Our new architecture combines the efficiency of federated learning and source data privacy of split learning. Furthermore, our new system has lower resource demand and processing time comparing to the states of the arts. Our experiments show that this improvement has a linear relationship with the number of clients. This work has been published at IEEE CLOUD2021 and a poster at CoNEXT2020.

• Zongshun Zhang, Andrea Pinto, Valeria Turina, Flavio Esposito, and Ibrahim Matta. Privacy and Efficiency of Communications in Federated Split Learning.In IEEE Transactions on Big Data, Oct 2023.

• Valeria Turina, Zongshun Zhang, Flavio Esposito, and Ibrahim Matta. Federated or Split? A Performance and Privacy Analysis of Hybrid Split and Federated Learning Architectures.In IEEE International Conference on Cloud Computing (IEEE CLOUD), September 2021.

• Valeria Turina, Zongshun Zhang, Flavio Esposito, and Ibrahim Matta. Combining split and federated architectures for efficiency and privacy in deep learning. In Proceedings of the 16th International Conference on emerging Networking EXperiments and Technologies (CoNEXT 2020), Barcelona, Spain, December 2020.

• By load balancing between FaaS and IaaS platforms with a low overhead statistical model, we are able to reduce 85% SLA violations and 20% cost for any general cloud application comparing to the states of the arts. We got the best paper award at IEEE IC2E for this project.

• Ali Raza, Zongshun Zhang, Nabeel Akhtar, Vatche Isahagian, and Ibrahim Matta.2021. LIBRA: An Economical Hybrid Approach for Cloud Application with Strict SLAs (doi). In Proceedings of the 9th IEEE International Conference on Cloud Engineering (IEEE IC2E), October 2021. (Best Paper Award!)

CS 655
Graduate Introduction to Computer Networks.
Fall 2021.
Teaching Fellow.

CS 101
Introduction to Computing.
Fall 2019.
Teaching Fellow.

Check out the template by jonbarron !