Simla Burcu Harma

Ph.D. Student, Computer Science, EPFL

Contact me:
Email

simla[dot]harma[at]epfl.ch

Address

EPFL IC IINFCOM PARSA

INJ 236 (Bâtiment INJ), Station 14

CH-1015 Lausanne, Switzerland

Hello! I am a third year Ph.D. Student at the École Polytechnique Fédérale de Lausanne (EPFL), supervised by Prof. Babak Falsafi, and I am affiliated with the Parallel Systems Architecture Lab (PARSA).

My current research focuses on novel numerical encodings for time- and energy-efficient Deep Neural Network (DNN) training and DNN accelerator design.

I am one of the recipients of the Microsoft Research PhD Fellowship, 2022. (Announcement, EPFL, EcoCloud)

I am also one of the recipients of the Generation Google Scholarship (EMEA). (Announcement, EcoCloud)

You can find my CV (updated July 2022) here.

You can find my footprint on the internet on GitHub and LinkedIn.

Education

Doctor of Philosophy (Ph.D.), École Polytechnique Fédérale de Lausanne (EPFL), Feb. 2020 - Present

School of Computer and Communication Sciences
Supervisor: Prof. Babak Falsafi, Parallel Systems Architecture Lab (PARSA)

Master of Science (M.S.), TOBB University of Economics and Technology, Sept. 2018 - Jan. 2020

Computer Science
Thesis: An in-depth Performance Analysis of Neural Machine Translation Tasks
Supervisor: Prof. Oğuz Ergin

Bachelor of Science (B.S.), TOBB University of Economics and Technology, Sept. 2014 - Aug. 2018

Mathematics (Double Major)
G.P.A. 4.0/4.0

Bachelor of Science (B.S.), TOBB University of Economics and Technology, Sept. 2013 - Aug. 2018

Computer Science
G.P.A. 4.0/4.0

Publications

Numerical Encoding for DNN Accelerators

Simla Burcu Harma, Mario Drumond, Babak Falsafi

In Computer Architecture Today (blog). September 20, 2021.

An in-depth Study of Neural Machine Translation Performance

Simla Burcu Harma, Mario Drumond, Babak Falsafi, Oğuz Ergin.

In ASPLOS’20 Workshop on the Convergence of Machine Learning and High-Performance Computing (ML4HPC), 2020 (Poster).

In HiPEAC Workshop on Accelerated Machine Learning (AccML), 2020 (Full Paper).