Two IC students win Qualcomm Fellowships
Kaicheng Yu (L) is a student in IC's Computer Vision Lab, while Drumond (R) studies parallel systems architecture. © EPFL
EPFL School of Computer and Communication Sciences (IC) PhD students Kaicheng Yu and Mario Paulo Drumond comprised two of the four winners of the 2019 Qualcomm Innovation Fellowship (QIF) Europe, administered by Qualcomm Technologies.
Kaicheng Yu, supervised by Mathieu Salzmann and Pascal Fua in the Computer Vision Laboratory (CVLab), was selected for his proposal, “Robust Neural Architecture Search with Soft Weight Sharing”. In this work, Yu proposed a new neural architecture search (NAS) training scheme with soft weight sharing that allows architectures to draw their weights from a larger parameter space.
Mario Paulo Drumond, supervised by Babak Falsafi in the Parallel Systems Architecture Laboratory (PARSA), was selected for his proposal, “ColTraIn: co-located deep learning training and inference”. Here, Drumond proposes an accelerator design for co-located training and inference in deep neural networks (DNN).
The annual Qualcomm fellowships, which bring $40,000 in academic funding plus mentoring by a Qualcomm Technologies researcher, reward innovative doctoral students doing engineering research in Europe, India and the United States.
“It’s a pleasure to discover the visionary research that students submit each year, research that could find its way in technology which will improve our transport, workplaces, homes and even health in the near future,” said Qualcomm Senior Vice President of Engineering and fellowship judge Rajesh Pankaj in a press statement.