Liliane Kunstmann

ORCID: 0000-0003-2648-7059
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Scientific Computing and Data Management
  • Research Data Management Practices
  • Distributed and Parallel Computing Systems
  • Computational Physics and Python Applications
  • Reservoir Engineering and Simulation Methods
  • Machine Learning in Materials Science
  • Advanced Data Storage Technologies
  • Cloud Computing and Resource Management
  • Machine Learning and Data Classification
  • Neural Networks and Applications
  • Seismic Imaging and Inversion Techniques
  • Model Reduction and Neural Networks
  • Brain Tumor Detection and Classification
  • Seismology and Earthquake Studies
  • Big Data and Digital Economy

Universidade Federal do Rio de Janeiro
2019-2024

The Deep Learning (DL) workflow involves several steps of data transformation. Evaluating various configurations at each step the may be a complex task when it comes to selecting DL models. This decision-making process requires basing decisions on metrics and continuously monitoring progression workflow. With plethora framework options that manage execution workflows algorithms, such as accuracy loss, in addition hyperparameters, are no longer enough for choosing models deployed. need...

10.1145/3650203.3663337 article EN 2024-05-29

Diversos workflows produzem um grande volume de dados e requerem técnicas paralelismo ambientes distribuídos para reduzir o tempo execução. Esses são executados por Sistemas Workflow, que apoiam a execução eficiente, mas focam em específicos. A tecnologia contêineres surgiu como solução uma aplicação execute heterogêneos meio da virtualização do SO. Embora existam soluções gerenciamento orquestração contêineres, e.g., Kubernetes, elas não científicos. Neste artigo, propomos AkôFlow,...

10.5753/sbbd.2024.241126 article PT 2024-10-14

Many existing scientific workflows require High Performance Computing environments to produce results in a timely manner. These have several software library components and use different environments, making the deployment execution of stack not trivial. This complexity increases if user needs add provenance data capture services workflow. manuscript introduces ProvDeploy assist configuring containers for with integrated capture. was evaluated Scientific Machine Learning workflow, exploring...

10.48550/arxiv.2403.15324 preprint EN arXiv (Cornell University) 2024-03-22

Deploying scientific workflows in high-performance computing (HPC) environments is increasingly challenging due to diverse computational settings. Containers help deploy and reproduce workflows, but both require more than just accessing container images. Container provenance provides essential information about image usage, origins, recipes, crucial for deployment on various architectures or engines. Current support limited actions processes without workflow traceability. We propose...

10.5753/sbbd.2024.240194 article EN 2024-10-14
Rafael Ferreira da Silva Deborah Bard Kyle Chard de Witt Shaun Ian Foster and 95 more Tom Gibbs Carole Goble William F. Godoy Johan E. Gustafsson Utz‐Uwe Haus Stephen D. Hudson Shantenu Jha Laura de los Drew Paine Frédéric Suter Logan Ward Sean Wilkinson Marcos Amarís Yadu Babuji Jonathan Bader Riccardo Balin Daniel Balouek‐Thomert Sarah Beecroft Khalid Belhajjame Rajat Bhattarai Wesley Brewer Paul Brunk Silvina Caíno‐Lores Henri Casanova Daniela Cassol Jared Coleman Tainã Coleman Iacopo Colonnelli Anderson Andrei Da Silva Daniel de Oliveira Pascal Elahi Nabil El‐Faramawy Wael Elwasif Brian D. Etz Thomas Fahringer Weder N. Ferreira Rosa Filgueira Jacob Fosso Tande Luiz Gadelha Andy Gallo Daniel Garijo Yiannis Georgiou Philipp Gritsch Patricia Grubel Amal Gueroudji Quentin Guilloteau Carlo Hamalainen R Latorre Enriquez Lauren Huet Kevin Hunter Kesling Paula Iborra Shiva Jahangiri Jan Janßen Joanne L. Jordan Sehrish Kanwal Liliane Kunstmann Fabian Lehmann Ulf Leser Chen Li Peini Liu Jakob Luettgau Richard Lupat José M. Fernández Ketan Maheshwari Tanu Malik Jack Marquez Motohiko Matsuda Doriana Medić Somayeh Mohammadi Alberto Mulone John-Luke Navarro Kin Wai Ng Klaus Noelp Bruno P. Kinoshita Ryan Prout Michael R. Crusoe Sasko Ristov Stefan A. Robila Daniel Rosendo Billy Rowell Jedrzej Rybicki Hector Sanchez Lopez Nishant Saurabh Sumit Kumar Saurav Tom Scogland Dinindu Senanayake Woong Shin Raúl Sirvent Tyler J. Skluzacek Barry Sly-Delgado Stian Soiland‐Reyes Abel Souza Renan P. Souza Domenico Talia Nathan R. Tallent

The Workflows Community Summit gathered 111 participants from 18 countries to discuss emerging trends and challenges in scientific workflows, focusing on six key areas: time-sensitive AI-HPC convergence, multi-facility heterogeneous HPC environments, user experience, FAIR computational workflows. integration of AI exascale computing has revolutionized enabling higher-fidelity models complex, processes, while introducing managing environments data dependencies. rise large language is driving...

10.5281/zenodo.13844758 preprint EN cc-by 2024-10-18

Machine Learning is being used increasingly in different application areas. Physics-Informed Neural Networks (PINN) stand out, adapting neural networks to predict solutions Physics phenomena. Incorporating knowledge into the loss function of a network, PINNs revolutionize partial differential equations. Considering lack support for analytics and reproducibility trained models, this paper we propose capture use provenance data, aimed at analysis PINN models. We conducted experiments using...

10.1109/sbac-padw60351.2023.00013 article EN 2023-10-17

Training Deep Learning (DL) models require adjusting a series of hyperparameters. Although there are several tools to automatically choose the best hyperparameter configuration, user is still main actor take final decision. To decide whether training should continue or try different configurations, needs analyze online hyperparameters most adequate dataset, observing metrics such as accuracy and loss values. Provenance naturally represents data derivation relationships (i.e.,...

10.5753/jidm.2021.1924 article EN Journal of Information and Data Management 2021-11-19

As aplicações científicas demandam ambientes de Processamento Alto Desempenho (PAD). Essas possuem diversos componentes advindos bibliotecas e diferentes ambientes, tornando a pilha software ser gerenciada no momento da implantação execução nada trivial. Essa complexidade aumenta caso o usuário necessite acoplar serviços captura dados proveniência à sua aplicação. Este artigo apresenta ProvDeploy para auxiliar na configuração contêineres aplicação com proveniência. O foi avaliado uma...

10.5753/wscad.2022.226363 article PT 2022-10-05

As redes neurais guiadas pela Física (PINNs) vêm revolucionando a aplicação de métodos numéricos. Apesar da complexidade configuração e geração do modelo, uma vez treinado, o mesmo mostra um ganho significativo em relação ao tempo cálculo dos A incorporação no treinamento se dá por meio modelagem novos componentes na função perda rede neural. Tais aumentam as configurações hiperparâmetros. Mostramos como coleta dados proveniência pode ajudar avaliação hiperparâmetros PINNs. Apresentamos...

10.5753/sbbd.2022.225367 article PT 2022-09-19

Due to the exploratory nature of DNNs, DL specialists often need modify input dataset, change a filter when preprocessing data, or fine-tune models’ hyperparameters, while analyzing evolution training. However, specialist may lose track what hyperparameter configurations have been used and tuned if these data are not properly registered. Thus, must be tracked made available for user’s analysis. One way doing this is use provenance derivation traces help hyperparameter’s fine-tuning by...

10.5753/jidm.2022.2544 article EN Journal of Information and Data Management 2022-12-19

Summary Geophysical imaging faces challenges in seismic interpretations due to multiple sources of uncertainties related data measurements, pre-processing and velocity analysis procedures. An essential part the decision-making process is understanding how they influence outcomes. For this, we present a new scientific workflow built upon Bayesian tomography, Reverse Time Migration, image interpretation based on machine learning techniques. Our explores an efficient hybrid computational...

10.3997/2214-4609.201903295 article EN 2019-01-01

O tempo de duração do ciclo vida no aprendizado por meio redes neurais profundas depende acerto em decisões configuração dados que levem ao sucesso na obtenção modelos. A análise hiperparâmetros e da evolução rede permite adaptações diminuem o vida. No entanto, há desafios não apenas coleta hiperparâmetros, mas também modelagem dos relacionamentos entre esses dados. Este trabalho apresenta uma abordagem centrada proveniência para enfrentar desafios, propondo com flexibilidade escolha...

10.5753/sbbd.2020.13639 article PT 2020-09-28

O treinamento de redes neurais profundas requer o ajuste hiperparâmetros. Este processo é custoso e ainda que existam ferramentas para escolha automática da melhor configuração hiperparâmetros, usuário responsável pela decisão final. Para isso, necessário analisar impacto diferentes hiperparâmetros sobre métricas como acurácia perda. A proveniência uma forma representar as relações derivação dados, fornecem um suporte importante nesta análise dados. Observando dificuldades análises...

10.5753/sbbd_estendido.2021.18158 article PT 2021-10-04
Coming Soon ...