About the Workshop

Deep learning is playing a transformative role in high-performance computing (HPC), enabling applications in scientific simulations, data-driven modeling, intelligent systems, and automated software engineering. However, the growing scale and complexity of models often exceed the available resources in HPC environments.

  • Explores techniques for efficient and scalable deep learning in HPC
  • Covers model distillation, quantization, pruning, sparse computation, and parameter-efficient tuning
  • Focuses on real-world integration in scientific and industrial HPC workflows

The LeanDL-HPC 2025 workshop offers a forum for researchers and practitioners to discuss these advances and share practical results. The program includes a technical session for research papers and experience reports, as well as the LLM Distillation Challenge: a poster session showcasing optimized LLMs evaluated on performance and efficiency under constrained computing conditions. Top submissions will be awarded based on a combined performance-efficiency score.

Workshop Scope

LeanDL-HPC 2025 invites high-quality submissions on both foundational and applied aspects of efficient deep learning for HPC environments. We welcome interdisciplinary work and practical contributions that explore new techniques, performance analyses, and deployment experiences.

Topics of interest include, but are not limited to:

  • Model distillation and compression techniques for scientific DL workloads
  • Quantization-aware training and low-precision inference in HPC systems
  • Pruning, sparsity, and structured compression for efficient model execution
  • Parameter-efficient tuning for domain-specific DL
  • Optimization of inference pipelines for large-scale simulations and data analysis
  • Memory- and compute-aware deployment strategies across HPC clusters
  • Efficient training and fine-tuning of LLMs and transformer-based models in HPC workflows
  • Lightweight GNNs, CNNs, and hybrid architectures for scientific applications
  • Benchmarks and performance metrics for evaluating DL efficiency in HPC settings
  • Energy-efficient and green AI techniques for deep learning in HPC
  • Use cases in climate modeling, materials science, physics-informed learning, bioinformatics, and other data-intensive scientific domains

Paper Submission

Main Track

The workshop will receive submissions of research papers and will carry out a single-blind peer review process to select high-quality papers to be presented in the workshop. Paper submissions must be in English, have from 6 up to 8 pages (including references), and follow the IEEE conference manuscript formatting guidelines for double-column text using a single-spaced 10-point font on 8.5 × 11-inch pages. Templates are available from http://www.ieee.org/conferences/publishing/templates.html. To be published in the conference proceedings and to be eligible for publication at the IEEE Xplore, one of the authors must register at the full rate and present his/her work at the conference.

Special Track – Challenge on LLM Distillation

In addition to the oral session, the workshop will host a special session dedicated to a challenge on efficient Large Language Model (LLM) distillation. Participants will be invited to submit lightweight versions of LLMs created using techniques such as distillation, pruning, quantization, or parameter-efficient tuning. Submitted solutions will be evaluated by the workshop organizers using a curated benchmark dataset.

Submission Link: will be available soon.

Important Dates (Anywhere on Earth)

  • Paper deadline: August 25th, 2025
  • Author notification: September 15th, 2025
  • Camera-ready submission: September 22th, 2025
  • Workshop date: October 28th–31th, 2025

Organizers

Technical Program Committee (in progress)

Below is an initial list of members for the Technical Program Committee (TPC).

  • Adenilso Simão - ICMC, University of São Paulo (USP)
  • Bruno Nogueira - Federal University of Mato Grosso do Sul (UFMS)
  • Diego F. Silva - ICMC, University of São Paulo (USP)
  • Edson Borin - University of Campinas (UNICAMP)
  • Edson Norberto Cáceres - Federal University of Mato Grosso do Sul (UFMS)
  • Edson Takashi Matsubura - Federal University of Mato Grosso do Sul (UFMS)
  • Jonathan de Andrade Silva - Federal University of Mato Grosso do Sul (UFMS)
  • Marcelo Augusto Santos Turine - Federal University of Mato Grosso do Sul (UFMS)
  • Mariana Caravanti de Souza - Federal University of Mato Grosso do Sul (UFMS)
  • Ricardo Marcacini - ICMC, University of São Paulo (USP)
  • Vitor Mesaque Alves de Lima - Federal University of Mato Grosso do Sul (UFMS)
  • Walfredo Cirne - Google, California