pSSim4AI
pSSim4AI project repository
Open source repository
In brief
The Probabilistic Symbolic Simulation For multi-core embedded Artificial Intelligence (pSSim4AI - 2020-2023) project aims at proposing a methodology to enable early performance and energy modeling for Artificial Neural Networks (ANNs) deployed on Multi-Processor Systems on Chip (MPSoCs).
This project is led in collaboration between the Institute Systems Engineering for Future Mobility (DLR-SE), Oldenburg, Germany and the Institut d'Électronique et des Technologies du numéRique (IETR) UMR CNRS 6164, Nantes Université, Nantes, France.
The considered ANNs are implemented using the LibFANN Deep Learning framework (https://github.com/libfann/fann) under GNU lesser general public license and CNN-cpp software (https://github.com/tranleanh/CNN-cpp).
Figure - Context: how to develop performance and energy models to enable fast-yet-accurate prediction of neural networks on multi-core platforms ahead of deployment phase?
Figure - Proposed modeling and Design Space Exploration (DSE) flow in the scope of the pSSim4AI project
List of folders
The pSSim4AI project open source Git repository is organized as follows:
- ./figures: Context and flow figures for this readme
- ./models: Modeling flow
- /MessageLevelModels: Simulable SystemC models,
- /DSE: Design Space Exploration flow including clusterizator module, pure analytical models and results.
- ./platform: Platform prototype (on Xilinx FPGA) and SW applications,
- ./project_management: Application documents regarding funding of the project
- ./publications: Sources, .pdf, slides and rebuttal letters for published papers,
- ./reports: Conference reports and manuscript high level of granularity plan
- ./scripts: Contains scripts including the transformation of LibFANN to C code executable on MicroBlaze microprocessor.
- ./workshops: Internal progress presentations regarding the project.
Publications
This project has led to several publications:
Conferences
-
RAPIDO'2023: Fast Yet Accurate Timing and Power Prediction of Artificial Neural Networks Deployed on Clock-Gated Multi-Core Platforms
- Authors: Dariol, Q.; Le Nours, S.; Helms, D.; Stemmer, R.; Pillement, S. & Grüttner, K.
- Conference: Workshop on System Engineering for constrained embedded systems (RAPIDO 2023), 2023, 8 pages
- DOI: 10.1145/3579170.3579263
- HAL_ID: hal-03932069
- E-LIB ID: 193755
-
SAMOS'2022: A Hybrid Performance Prediction Approach for Fully-Connected Artificial Neural Networks on Multi-core Platforms
- Authors: Dariol, Q.; Le Nours, S.; Pillement, S.; Stemmer, R.; Helms, D. & Grüttner, K.
- Editors: Orailoglu, A.; Reichenbach, M. & Jung, M.
- Conference: International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS 2022), Springer International Publishing, 2022, 250-263
- DOI: 10.1007/978-3-031-15074-6_16
- HAL_ID: hal-03758005
- E-LIB ID: 188200
Posters
- **GDRSOC'22: Hybrid Performance Prediction Models for Fully-Connected Neural Networks on MPSoC
- Authors: Quentin Dariol, Sébastien Le Nours, Sébastien Pillement, Ralf Stemmer, Domenik Helms, Kim Grüttner
- Conference: Colloque National du GDR SOC2
- Date: June 2022
- HAL_ID: hal-03758026
- E-LIB ID: https://elib.dlr.de/188199/
-
GDRSOC'21: A Measurement-based Performance Evaluation Framework for Neural Networks on MPSoCs
- Authors: Quentin Dariol, Sébastien Le Nours, Sébastien Pillement, Ralf Stemmer, Kim Gruettner, and Domenik Helms
- Conference: Colloque National du GDR SOC2
- Date: June 2021
- HAL_ID: hal-03248152
Technical report
-
Setup of an Experimental Framework for Performance Modeling and Prediction of Embedded Multicore AI Architectures
- Authors: Quentin Dariol, Sébastien Le Nours, Sébastien Pillement, Kim Grüttner, Domenik Helms, Ralf Stemmer
- Institute: Nantes Université, IETR UMR CNRS 6164, Nantes, France and German Aerospace Center - Institute Systems Engineering for Future Mobility (DLR-SE), Oldenburg, Germany
- Date: Jan 2022
- HAL_ID: hal-03546804
- E-LIB ID: 188380