About Me

I am a PhD student at Arizona State University advised by Hannah Kerner focusing on Machine Learning for Remote Sensing. I am part of the Rwanda Space Agency (RSA) Geospatial team, where I contribute on the development of machine learning systems that leverage satellite imagery for real-world applications.

Prior to RSA, I was a machine learning research engineer at INRIA Grenoble where my research focused on hyperspectral unmixing and sparse coding. I have a masters in Machine Intelligence from AMMI where I worked with Georgia Gkioxari on deep compression for edge computing and their application in conservation of the environment.

My CV/Resume

News

May 2023: Co-organized the Machine Learning for Remote Sensing workshop at ICLR 2023 [Workshop website]

Research and Publications


	Entropic Descent Archetypal Analysis for Blind Hyperspectral Unmixing: We introduced a new algorithm based on archetypal analysis for blind hyperspectral unmixing, assuming linear mixing of endmembers. /Github / IEEE Explore /
	Compressed Object Detection: In this work, we extended pruning, a compression technique which discards unnecessary model connections, and weight sharing techniques for the task of object detection. With our approach we are able to compress a state-of-the-art object detection model by 30.0% without a loss in performance. We also show that our compressed model can be easily initialized with existing pre-trained weights, and thus is able to fully utilize published state-of-the-art model zoos. /Github / arXiv /

Entropic Descent Archetypal Analysis for Blind Hyperspectral Unmixing: We introduced a new algorithm based on archetypal analysis for blind hyperspectral unmixing, assuming linear mixing of endmembers. /Github / IEEE Explore /

Compressed Object Detection: In this work, we extended pruning, a compression technique which discards unnecessary model connections, and weight sharing techniques for the task of object detection. With our approach we are able to compress a state-of-the-art object detection model by 30.0% without a loss in performance. We also show that our compressed model can be easily initialized with existing pre-trained weights, and thus is able to fully utilize published state-of-the-art model zoos. /Github / arXiv /