Welcome! I am a researcher and quantitative engineer who enjoys creating novel solutions for big data and machine learning problems while exploring diverse research topics, transferring ideas across domains, and ensuring practical effectiveness.

I am currently on the 2025 job market. Please contact me if you would like to discuss potential openings or collaborations!

💼 Professional Experiences

• Graduate Researcher at UofT: I built personalized and group-wise differentially private federated learning algorithms to meet heterogeneous privacy needs, and engineered fault-tolerant distributed systems that mitigate stragglers.
• Privacy Research Scientist Intern at CISPA: I designed time-adaptive, personalized privacy budget mechanisms to enhance learning performance under anonymization constraints.
• Machine Learning Research Associate Intern at Huawei: I developed quantization error bounds to optimize hardware-aware training of deep neural networks.
• Information Theory Research Intern at CUHK: I explored fundamental problems in network information theory to model reliability trade-offs in communication channels.

🎓 Academic Background

I earned my Ph.D. and MASc. in Electrical and Computer Engineering at the University of Toronto, advised by Prof. Stark C. Draper, working on the intersection of distributed systems, machine learning, and data privacy. Before that, I received my B.Sc. in Electrical Engineering (digital systems) with a minor in Economics at the Sharif University of Technology.

🔬 Areas of Expertise

My work spans:

• Privacy in Machine Learning: I developed differentially private training methods and implemented them in Python using Opacus.
• Optimization: I tuned algorithm parameters and analyzed convergence of training algorithms under privacy and utility constraints.
• Federated Machine Learning: My research pioneered the design of private grouped federated learning under personalization constraints.
• Algorithm Design: I designed randomized, numerically stable, and approximate matrix multiplication algorithms for straggler-resilient distributed systems.
• Error-Correction Codes: I applied polynomial coding and interpolation to introduce efficient redundancy and address stragglers in distributed computing.
• Quantization in Machine Learning: I analyzed the error of using reduced-precision weights and activations to tradeoff accuracy for speed and memory efficiency.
• Implementation: I built differentially private ML in PyTorch with Opacus, and tested distributed computing on EC2 and SciNet HPC cluster using Python and mpi4py.

Selected Publications

ICLR 2025	Differentially Private Federated Learning with Time-Adaptive Privacy Spending. Shahrzad Kiani, Nupur Kulkarni, Adam Dziedzic, Stark Draper, Franziska Boenisch
JSAIT 2024	Controlled Privacy Leakage Propagation Throughout Overlapping Grouped Learning. Shahrzad Kiani, Franziska Boenisch, Stark Draper
JSAIT 2022	Successive Approximated Coded Matrix Multiplication. Shahrzad Kiani, Stark Draper
TIT 2021	Hierarchical Coded Matrix Multiplication. Shahrzad Kiani, Nuwan Ferdinand, Stark Draper

News

Personal

Outside of research, I enjoy practicing creativity through cooking and experimenting with new dishes, hiking new trails, and trying to win at backgammon or other strategic games. When I’m not lazy, I spend time gardening. Whenever possible, I travel to explore new cultures.