The 2023 Chicken-egg HPC/DL Workshop

Porto Alegre - RS, Brazil — October 17th, 2023.

About

Deep Learning (DL) models have become ubiquitous tools in several (if not all) science and industry areas. Large neural network models developed through DL research have been successful not only in learning from unstructured data - such as images, videos, natural language, and audio – but also from structured data – such as graphs. Deep neural networks are nowadays employed in healthcare, entertainment, marketing, scientific research, and many other fields. To efficiently train, adapt and deploy such large-scale neural models, it is primordial to rely on High-Performance Computing (HPC). HPC systems provide the necessary computational power to train these models in a reasonable amount of time, which is essential for many practical applications. On the other hand, deep learning models have shown great potential in enhancing the hardware and software optimization of HPC systems. By using DL models, it is possible to optimize HPC system parameters and improve their performance, reducing the time and cost required to manually optimize HPC systems. In this workshop, we will focus both on the use of HPC for learning deep learning models based on neural networks and on using deep learning models to enhance hardware and software optimization of HPC systems.
The workshop aims at facilitating knowledge sharing, networking, and collaboration among participants in the field of HPC and Deep Learning. The workshop will also provide a platform for researchers and practitioners to showcase their latest research and discuss potential applications of HPC and deep learning in industry and governments. We hope that the workshop will inspire new ideas, collaborations, and research directions in the intersection of HPC and DL.
We invite authors to submit papers with original research, practical application papers, Ph.D. and Master Thesis ongoing work, as well as to submit presentation proposals of papers recently published in top conferences and journals in the intersection of HPC and DL research areas.

Important Dates

Paper deadline: August 28thSeptember 11th, 2023 (Anywhere on earth!)

Author notification: September 22ndSeptember 29th, 2023

Camera-ready submission: September 29thOctober 6th, 2023

Workshop Date: October 17th, 2023

Topics of interest

Topics of interest include, but are not limited to:

  • HPC architectures and systems to learn deep neural networks from scratch;
  • HPC architectures and systems to adapt and tune deep neural networks;
  • HPC architectures and systems to learn deep neural networks in low-resource scenarios;
  • HPC architectures and systems to deploy deep learning models in low-cost hardware;
  • HPC architectures and systems to learn foundation models;
  • HPC architectures and systems to learn physics-informed models;
  • Distributed and federated learning of deep neural networks;
  • Parallel programming models and tools to enhance the learning of deep neural networks;
  • HPC for supporting performance analysis and tuning of deep neural networks;
  • HPC for supporting the practical use of deep neural networks;
  • Deep learning and its applications to improve HPC architectures and systems;
  • Hardware and software optimizations for HPC learning deep neural networks;
  • Deep learning for hardware and software optimization for HPC;
  • Benchmarking and performance evaluation of deep neural networks;
  • Use cases and applications of HPC and deep learning in diverse industries;
  • Use cases and applications of HPC and deep learning on texts;
  • Use cases and applications of HPC and deep learning on images;
  • Use cases and applications of HPC and deep learning on graphs;
  • Green HPC for learning deep models;
  • Using deep learning to enhance the sustainability of HPC Ecosystems;
  • Ethical aspects of HPC and deep learning.

Keynote speaker

Prof. Carla Osthoff - LNCC

Short-bio: Graduated in Electrical Engineering from the Pontifical Catholic University of Rio de Janeiro with a master's and a doctorate in Computer Engineering Systems from the Federal University of Rio de Janeiro. She is currently a researcher in the area of High-Performance Computing at the National Laboratory for Scientific Computing (LNCC), a professor at the Multidisciplinary Graduate Program at LNCC, a member of the Consultative Committee of the Supercomputer Santos Dumont, coordinates the LNCC National Center for High-Performance Processing (CENAPAD), and the High-Performance Processing Sector of LNCC, which has several collaborative projects in the area of High-Performance Computing. Her research focuses on High-Performance Computing, Parallel Programming, and scientific application performance optimizations.


Developing Efficient Scientific Gateways for Supercomputer Environments Supported by Deep Learning Models

The scientific gateway BioinfoPortal for bioinformatics applications is hosted in the National Laboratory for Scientific Computing (LNCC) and is coupled to the Santos Dumont (SDumont) supercomputer environment. BioinfoPortal offers a catalog of bioinformatics computational software that benefits from the parallel and distributed architecture offered by LNCC. Task submissions consume SDumont nodes shared by other supercomputer users; then, it is required to set the best configuration, defined by the best choice of the number of threads/nodes, to be allocated for every task submission. This talk presents research analysis using Deep Neural Networks to estimate the computational time required to execute bioinformatics software in several scenarios using a pre-configured number of nodes and threads. We aim to demonstrate the computational behavior of software in Bioinfoportal and which computational scenario can be chosen to execute software in SDumont efficiently. Results support that the neural networks can predict the most representative variable and identify the configuration with the lowest computational time. This way, BioinforPortal consuming time can lead to an efficient and green gateway, increasing Santos Dumont Supercomputing execution job throughput and decreasing job execution queue waiting time.

Panel

Towards developing HPC through ML and ML through HPC - the Chicken Egg dilemma

Machine Learning (ML) area have evolved in the last years due to the development of High Performance Computing (HPC). On the other hand, we also have seen many works for evolving HPC through ML. In this context, we could ask: If the community needs HPC to make huge ML-based models more effective, but this same community also needs ML to empower HPC, which area starts this run? Are they exclusive or can we consider both initiatives? This panel aims to discuss this chicken-egg dilemma.

Moderator

Profa. Flavia Bernardini - UFF

Participants

  • Profa. Lucia Drummond - UFF
  • Prof. Antônio Tadeu Azevedo Gomes - LNCC
  • Profa. Mariza Ferro - UFF
  • Dr. Fabio Alves de Oliveira - NVidia

Program

09:00-09:15 - Workshop opening

  • Workshop chairs

09:15-10:30 - Keynote talk

  • Developing Efficient Scientific Gateways for Supercomputer Environments Supported by Deep Learning Models.. Prof. Carla Osthoff

11:00-12:00 - Paper Session 1

  • Assessing the performance of an architecture-aware optimization tool for neural networks. Raúl Marichal, Ernesto Dufrechou, Pablo Ezzatti
  • An Exploratory Study of Deep Learning for Predicting Computational Tasks Behavior in HPC Systems. Alexandre Henrique Lopes Porto, Micaella Coelho, Kary Ocaña, Carla Osthoff, Francieli Boito, Douglas O. Cardoso

13:30-15:00 - Paper Session 2

  • PINNProv: Provenance for Physics-informed Neural Networks. Lyncoln S. de Oliveira, Liliane Kunstmann, Débora Pina, Daniel de Oliveira, Marta Mattoso
  • Exploring Federated Learning to Trace Depression in Social Media with Language Models. Arthur Vasconcelos, Lúcia Drummond, Rafaela Brum, Aline Paes
  • Computing seismic attributes with deep-learning models. Nícolas Hecker, Otávio O. Napoli, Carlos A. Astudillo, João Paulo Navarro, Alan Souza, Daniel Miranda, Leandro A. Villas, Edson Borin

15:30-17:00 - Panel

  • Towards developing HPC through ML and ML through HPC - the Chicken Egg dilemma.
    • Profa. Flavia Bernardini - UFF (Moderator)
    • Profa. Lucia Drummond - UFF
    • Prof. Antônio Tadeu Azevedo Gomes - LNCC
    • Profa. Mariza Ferro - UFF
    • Dr. Fabio Alves de Oliveira - NVidia

Submission Guidelines

Submissions must be in English, with 4-8 pages, following the IEEE conference formatting guidelines (submission). All accepted papers will be published at IEEE Xplore. Papers that do not meet these requirements may be rejected without a review.
Manuscript templates for IEEE conference proceedings can be found at the following link: https://www.ieee.org/conferences_events/conferences/publishing/

All paper submissions must be made through EasyChair.

Submission link: https://easychair.org/my/conference?conf=hpcdl23

Submit your work!

Organizing Committe

General chairs:

  • Flavia Bernardini (UFF)
  • Edson Borin (UNICAMP)

Program Committee Chairs:

  • Aline Paes (UFF)
  • Mariza Ferro (UFF)

Program Committee:

  • Alba Cristina M. A. Melo - UNB
  • Alvaro Fazenda - UNIFESP
  • Alvaro Coutinho - UFRJ/High Performance Computing Center
  • Antonio Tadeu Gomes - LNCC
  • Claude Tandoki - Mines ParisTech / CRI - Centre de Recherche en Informatique
  • Carla Osthoff - LNCC
  • Cristiana Bentes - UERJ
  • Daniel de Oliveira - UFF
  • Esteban Clua - UFF
  • Fabio Porto - LNCC
  • Leandro Santiago de Araújo - UFF
  • João Paulo Navarro - NVIDIA
  • Lúcia Drummond - UFF
  • Pedro Mario Silva - NVIDIA
  • Rafaelli C. Coutinho - CEFET-RJ
  • Roberto P. Souto - LNCC