Ahmad Sajedi

I am Ahmad Sajedi, currently a Machine Learning Engineer II at Instacart. Before this role, I completed my Ph.D. at the University of Toronto, where I was fortunate to be advised by Prof. Konstantinos N. Platanioits and Prof. Yuri A. Lawryshyn. Prior to joining UofT, I received my M.Sc. degree from the Electrical and Computer Engineering Department at the University of Waterloo, mentored by Prof. En-Hui Yang.

Email: sajedi [dot] ah [at] gmail [dot] com

Email  |  Resume  |  LinkedIn  |  GitHub

News

Publications & Manuscripts

We present D2M, which embeds knowledge into a generative model, allowing for efficient and scalable training across various distillation ratios and architectures.

We introduce the ATOM module to leverage contextual and localization information from the channel and spatial attention mechanisms to improve dataset distillation efficiency.

We propose a simple yet effective probabilistic contrastive learning framework for multi-label image classification tasks using Gaussian mixture models.

We present an effective learning framework to distill informative knowledge from a large-scale training dataset into a small, synthetic one.

We propose a new probabilistic distance metric to compare two continuous probability density functions. This metric provides a closed-form expression for Gaussian Mixture Models.

We present an end-to-end kernel-based contrastive learning framework designed for multilabel datasets in computer vision and medical imaging.

We introduce a novel personalized federated learning algorithm that leverages the inherent graph-based relationships among clients.

We present subclass knowledge distillation, a method of transferring predicted subclass knowledge from a teacher to a smaller student model, designed for clinical applications.

We introduce a novel model distillation algorithm that leverages subclass labels' knowledge and quantifies the information the teacher can provide to the student through our framework.

By leveraging feature map sparsity, we introduce two novel convolution algorithms aiming to reduce memory usage, enhance inference speed, and maintain accuracy simultaneously.

Patents

Ahmad Sajedi, Ehsan Amjadian, Samir Khaki, Lucy Z. Liu, Yuri A. Lawryshyn, Konstantinos N. Plataniotis

Ahmad Sajedi, Ehsan Amjadian, Samir Khaki, Lucy Z. Liu, Yuri A. Lawryshyn, Konstantinos N. Plataniotis

Samir Khaki, Ahmad Sajedi, Lucy Z. Liu, Yuri A. Lawryshyn, Konstantinos N. Plataniotis

Education

Experiences

Professional Academic Activities

Selected Teaching

Source code from Jon Barron's lovely website.