NERSCPowering Scientific Discovery Since 1974

Amanda Dufek

photo 2020 06 30 18.34.42
Amanda Sabatini Dufek, Ph.D.
Application Performance Engineer
Application Performance Group

Biographical Sketch

Amanda Sabatini Dufek is a member of the Application Performance Group at the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory (LBNL) since January 2021. She is currently involved in three main projects: (1) Toolkit for Extreme Climate Analysis (TECA), a parallel toolkit for detecting extreme events in large climate datasets; (2) Learning to Grow (L2G) for LAMMPS, an evolutionary reinforcement learning of self-assembly protocols in LAMMPS simulations; and (3) some parallel implementations using SYCL programming model, such as SU3_Bench and MILC-Dslash benchmarks.

She has worked as a NERSC Exascale Science Applications Postdoctoral (NESAP) Fellow at LBNL in 2020. The main goal of her NESAP project for the simulation program was adding two new diagonal ceilings to the roofline model relative to NIC and PCI-e bandwidths to provide insights into how the communication operations are impacting the overall performance of some benchmark problems.

As an Institutional Training Program Fellow at the Brazilian National Laboratory for Scientific Computing (LNCC) (2016-2019), a multi- and many-threaded heterogeneous parallel version of the Grammatical Evolution algorithm was developed in order to achieve a fully parallel implementation, where both the breeding and evaluation are parallelized. A cooperative algorithmic-level parallel model was also implemented, in which many independent runs of the algorithm are launched simultaneously in a parallel cooperative way. It is a Free Software written in C/C++, OpenMP and OpenCL, freely available at She also developed a many-threaded parallel version of a bi-level optimization algorithm using an evolutionary method known as Differential Evolution, where both levels are parallelized. It has been written in C/C++ and OpenCL, whose implementation basically consists of a nested execution of two OpenCL kernels.

She received her Bachelor (2005) and Master (2008) degrees in Meteorology from the University of São Paulo located in Brazil, and her Doctor (2015) degree in Computational Modeling from the Brazilian National Laboratory for Scientific Computing. She was Assistant Professor of the Department of Atmospheric Sciences at University of São Paulo from 2008-2009. Her research interests are: high-performance computing, machine learning, evolutionary algorithms and meteorological problems.

Journal Articles

  1. Dufek, A. S., Augusto, D. A., Barbosa, H. J. C., Silva Dias, P. L. Data-driven symbolic ensemble models for wind speed forecasting through evolutionary algorithms, Applied Soft Computing, v.87, p.105976, 2020.
  2. Dufek, A. S., Augusto, D. A., Silva Dias, P. L., Barbosa, H. J. C. Application of evolutionary computation on ensemble forecast of quantitative precipitation, Computers & Geosciences, v.106, p.139-149, 2017.
  3. Dufek, A. S., Ambrizzi, T., Rocha, R. P. Are Reanalysis data useful to calculate climate indices over South America?, Annals of the New York Academy of Sciences, v.1146, p.87-104, 2008.
  4. Dufek, A. S., Ambrizzi, T. Precipitation variability in São Paulo State, Brazil, Theoretical and Applied Climatology, v.93, p.167-178, 2008.
  5. Dufek, A. S., Ambrizzi, T. Variabilidade climática da temperatura no Estado de São Paulo, Revista de Iniciação Cientíca (USP), v.7, p.23-29, 2005.

Conference Papers

  1. Dufek A. S., Gayatri, R., Mehta, N., Doerfler, D., Cook, B. G., Ghadar, Y., DeTar, C. Case Study of Using Kokkos and SYCL as Performance-Portable Frameworks for Milc-Dslash Benchmark on NVIDIA, AMD and Intel GPUs. International Workshop on Performance, Portability and Productivity in HPC (P3HPC), p.57-67, 2021.
  2. Dufek, A. S., Deslippe, J. R., Lin, P. T., Yang, C. J., Cook, B. G., Madsen, J. An Extended Roofline Performance Model with PCI-E and Network Ceilings. International Workshop on Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS), p.30-39, 2021.
  3. Dufek, A. S., Augusto, D. A., Barbosa, H. J. C., Silva Dias, P. L., Deslippe, J. R. An efficient fault-tolerant communication algorithm for population-based metaheuristics. In Proceedings of the Genetic and Evolutionary Computation Conference Companion (GECCO), p.1290-1298, 2021.
  4. Dufek, A. S., Augusto, D. A., Silva Dias, P. L., Barbosa, H. J. C. Evaluating the feasibility of grammar-based GP in combining meteorological forecast models, In IEEE Congress on Evolutionary Computation (CEC), Cancun, Mexico, 2013.
  5. Dufek, A. S., Augusto, D. A., Silva Dias, P. L., Barbosa, H. J. C. Grammatical evolution for data classication: Application to real meteorological data sets, In Proceedings of the XXXII Iberian Latin American Congress on Computational Methods in Engineering (CILAMCE), Ouro Preto, MG, Brazil, 2011.
  6. Dutra, L. M. M., Peres, J. R. R., Dufek, A. S., Camargo, R. Rastreamento dos bloqueios ocorridos próximos à América do Sul em julho de 2008 e 2009 e sua inuência sobre São Paulo e a Região Sul do Brasil, In XVI Congresso Brasileiro de Meteorologia, Belém, PA, Brasil, 2010.

Book Chapters

  1. Dufek, A. S., Augusto, D. A., Barbosa, H. J. C., Silva Dias, P. L. Handbook of Grammatical Evolution, chapter Multi- and many-threaded heterogeneous parallel grammatical evolution, Conor Ryan, Michael O'Neill, John J. Collins, editors, Springer International Publishing, p.219-244, 2018.