DeepChip: Deep learning on resource-constrained systems
Summary: DeepChip gathers methods that support the implementation of deep convolutional networks for machine learning on embedded platforms. DeepChip heavily relies on extreme forms of quantization and related unsafe optimizations to match the computational and memory requirements of deep neural networks to available hardware resources. By integrating these techniques in existing tool stacks for machine learning, it allows specialists in machine learning to leverage the ubiquitous availability and high energy-efficiency of advanced embedded systems for an improved use of classification and regression methods. This D-A-CH (DFG/FWF) project is a collaboration with the group of Franz Pernkopf from Technical University of Graz. While we already started our efforts with the help of research assistants and master students earlier, the funded project has just started in December 2016. More on http://www.deepchip.org/.
Dissemination
2024
- Less Memory Means smaller GPUs: Backpropagation with Compressed ActivationsCoRR, abs/2409.11902, 2024
@article{barley2024, author = {Barley, Daniel and Fr{{\"o}}ning, Holger}, title = {Less Memory Means smaller GPUs: Backpropagation with Compressed Activations}, year = {2024}, volume = {abs/2409.11902}, journal = {CoRR}, url = {https://arxiv.org/abs/2409.11902}, }
- Resource-Efficient Neural Networks for Embedded SystemsJournal of Machine Learning Research, 25(50), 1–51, 2024
@article{JMLR:v25:18-566, author = {Roth, Wolfgang and Schindler, G{{\"u}}nther and Klein, Bernhard and Peharz, Robert and Tschiatschek, Sebastian and Fr{\"{o}}ning, Holger and Pernkopf, Franz and Ghahramani, Zoubin}, title = {Resource-Efficient Neural Networks for Embedded Systems}, journal = {Journal of Machine Learning Research}, year = {2024}, volume = {25}, number = {50}, pages = {1--51}, url = {http://jmlr.org/papers/v25/18-566.html}, }
- Walking Noise: On Layer-Specific Robustness of Neural Architectures against Noisy Computations and Associated Characteristic Learning DynamicsEuropean Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD), 2024
@inproceedings{borras2024, title = {Walking Noise: On Layer-Specific Robustness of Neural Architectures against Noisy Computations and Associated Characteristic Learning Dynamics}, author = {Borras, Hendrik and Klein, Bernhard and Fr{\"{o}}ning, Holger}, booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases}, year = {2024}, series = {ECML-PKDD}, url = {https://doi.org/10.1007/978-3-031-70359-1_3}, }
- Function Space Diversity for Uncertainty Prediction via Repulsive Last-Layer EnsemblesICML 2024 Workshop on Structured Probabilistic Inference & Generative Modeling, 2024
@inproceedings{steger2024function, title = {Function Space Diversity for Uncertainty Prediction via Repulsive Last-Layer Ensembles}, author = {Steger, Sophie and Knoll, Christian and Klein, Bernhard and Fr{\"o}ning, Holger and Pernkopf, Franz}, booktitle = {ICML 2024 Workshop on Structured Probabilistic Inference {\&} Generative Modeling}, year = {2024}, url = {https://openreview.net/forum?id=FbMN9HjgHI}, }
- Probabilistic Photonic Computing with Chaotic LightCoRR, abs/2401.17915, 2024
@article{brckerhoffplckelmann2024probabilistic, title = {Probabilistic Photonic Computing with Chaotic Light}, author = {Brückerhoff-Plückelmann, Frank and Borras, Hendrik and Klein, Bernhard and Varri, Akhil and Becker, Marlon and Dijkstra, Jelle and Brückerhoff, Martin and Wright, C. David and Salinga, Martin and Bhaskaran, Harish and Risse, Benjamin and Fr{\"o}ning, Holger and Pernice, Wolfram}, year = {2024}, volume = {abs/2401.17915}, journal = {CoRR}, url = {https://arxiv.org/abs/2401.17915}, }
- Implications of Noise in Resistive Memory on Deep Neural Networks for Image ClassificationCoRR, abs/2401.05820, 2024
@article{DBLP:journals/corr/abs-2401-05820, author = {Emonds, Yannick and Xi, Kai and Fröning, Holger}, title = {Implications of Noise in Resistive Memory on Deep Neural Networks for Image Classification}, journal = {CoRR}, volume = {abs/2401.05820}, year = {2024}, url = {https://arxiv.org/abs/2401.05820}, doi = {10.48550/ARXIV.2401.05820}, eprinttype = {arXiv}, eprint = {2401.05820}, timestamp = {Thu, 25 Jan 2024 00:00:00 +0100}, }
2023
- Reducing Memory Requirements for the IPU using Butterfly FactorizationsSC ’23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis, SC-W 2023, Denver, CO, USA, November 12-17, 2023, 1255–1263, ACM, 2023
@inproceedings{DBLP:conf/sc/ShekoftehAF23, author = {Shekofteh, S. Kazem and Alles, Christian and Fr{\"{o}}ning, Holger}, title = {Reducing Memory Requirements for the {IPU} using Butterfly Factorizations}, booktitle = {{SC} '23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis, {SC-W} 2023, Denver, CO, USA, November 12-17, 2023}, pages = {1255--1263}, publisher = {{ACM}}, year = {2023}, url = {https://doi.org/10.1145/3624062.3624196}, doi = {10.1145/3624062.3624196}, timestamp = {Tue, 28 Nov 2023 00:00:00 +0100}, }
- On the Non-Associativity of Analog ComputationsCoRR, abs/2309.14292, 2023
@article{DBLP:journals/corr/abs-2309-14292, author = {Kuhn, Lisa and Klein, Bernhard and Fr{\"{o}}ning, Holger}, title = {On the Non-Associativity of Analog Computations}, journal = {CoRR}, volume = {abs/2309.14292}, year = {2023}, url = {https://arxiv.org/abs/2309.14292}, doi = {10.48550/ARXIV.2309.14292}, eprinttype = {arXiv}, eprint = {2309.14292}, timestamp = {Wed, 27 Sep 2023 01:00:00 +0200}, }
2022
- QONNX: Representing Arbitrary-Precision Quantized Neural NetworksCoRR, abs/2206.07527, 2022
@article{DBLP:journals/corr/abs-2206-07527, author = {Pappalardo, Alessandro and Umuroglu, Yaman and Blott, Michaela and Mitrevski, Jovan and Hawks, Benjamin and Tran, Nhan and Loncar, Vladimir and Summers, Sioni and Borras, Hendrik and Muhizi, Jules and Trahms, Matthew and Hsu, Shih{-}Chieh and Hauck, Scott and Duarte, Javier M.}, title = {{QONNX:} Representing Arbitrary-Precision Quantized Neural Networks}, journal = {CoRR}, volume = {abs/2206.07527}, year = {2022}, url = {https://arxiv.org/abs/2206.07527}, doi = {10.48550/ARXIV.2206.07527}, eprinttype = {arXiv}, eprint = {2206.07527}, timestamp = {Tue, 21 Jun 2022 17:35:15 +0200}, }
- Open-source FPGA-ML codesign for the MLPerf Tiny BenchmarkCoRR, abs/2206.11791, 2022
@article{DBLP:journals/corr/abs-2206-11791, author = {Borras, Hendrik and Guglielmo, Giuseppe Di and Duarte, Javier M. and Ghielmetti, Nicol{\`{o}} and Hawks, Benjamin and Hauck, Scott and Hsu, Shih{-}Chieh and Kastner, Ryan and Liang, Jason and Meza, Andres and Muhizi, Jules and Nguyen, Tai and Roy, Rushil and Tran, Nhan and Umuroglu, Yaman and Weng, Olivia and Yokuda, Aidan and Blott, Michaela}, title = {Open-source {FPGA-ML} codesign for the MLPerf Tiny Benchmark}, journal = {CoRR}, volume = {abs/2206.11791}, year = {2022}, url = {https://arxiv.org/abs/2206.11791}, doi = {10.48550/ARXIV.2206.11791}, eprinttype = {arXiv}, eprint = {2206.11791}, timestamp = {Wed, 29 Jun 2022 11:10:54 +0200}, }
- Towards Hardware-Specific Automatic Compression of Neural NetworksCoRR, abs/2212.07818, 2022
@article{DBLP:journals/corr/abs-2212-07818, author = {Krieger, Torben and Klein, Bernhard and Fr{\"{o}}ning, Holger}, title = {Towards Hardware-Specific Automatic Compression of Neural Networks}, journal = {CoRR}, volume = {abs/2212.07818}, year = {2022}, url = {https://arxiv.org/abs/2212.07818}, doi = {10.48550/ARXIV.2212.07818}, eprinttype = {arXiv}, eprint = {2212.07818}, timestamp = {Mon, 02 Jan 2023 00:00:00 +0100}, }
2021
- Towards Addressing Noise and Static Variations of Analog Computations Using Efficient RetrainingMachine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of ECML PKDD 2021, Proceedings Part I (Communications in Computer and Information Science), 1524, 409–420, Springer, 2021
@inproceedings{DBLP:conf/pkdd/KleinKWESSF21, author = {Klein, Bernhard and Kuhn, Lisa and Weis, Johannes and Emmel, Arne and Stradmann, Yannik and Schemmel, Johannes and Fr{\"{o}}ning, Holger}, editor = {Kamp, Michael and Koprinska, Irena and Bibal, Adrien and Bouadi, Tassadit and Fr{\'{e}}nay, Beno{\^{\i}}t and Gal{\'{a}}rraga, Luis and Oramas, Jos{\'{e}} and Adilova, Linara and Krishnamurthy, Yamuna and Kang, Bo and Largeron, Christine and Lijffijt, Jefrey and Viard, Tiphaine and Welke, Pascal and Ruocco, Massimiliano and Aune, Erlend and Gallicchio, Claudio and Schiele, Gregor and Pernkopf, Franz and Blott, Michaela and Fr{\"{o}}ning, Holger and Schindler, G{\"{u}}nther and Guidotti, Riccardo and Monreale, Anna and Rinzivillo, Salvatore and Biecek, Przemyslaw and Ntoutsi, Eirini and Pechenizkiy, Mykola and Rosenhahn, Bodo and Buckley, Christopher L. and Cialfi, Daniela and Lanillos, Pablo and Ramstead, Maxwell and Verbelen, Tim and Ferreira, Pedro M. and Andresini, Giuseppina and Malerba, Donato and Medeiros, Ib{\'{e}}ria and Fournier{-}Viger, Philippe and Nawaz, M. Saqib and Ventura, Sebasti{\'{a}}n and Sun, Meng and Zhou, Min and Bitetta, Valerio and Bordino, Ilaria and Ferretti, Andrea and Gullo, Francesco and Ponti, Giovanni and Severini, Lorenzo and Ribeiro, Rita P. and Gama, Jo{\~{a}}o and Gavald{\`{a}}, Ricard and Cooper, Lee A. D. and Ghazaleh, Naghmeh and Richiardi, Jonas and Roqueiro, Damian and Miranda, Diego Saldana and Sechidis, Konstantinos and Gra{\c{c}}a, Guilherme}, title = {Towards Addressing Noise and Static Variations of Analog Computations Using Efficient Retraining}, booktitle = {Machine Learning and Principles and Practice of Knowledge Discovery in Databases - International Workshops of {ECML} {PKDD} 2021, Proceedings Part {I}}, series = {Communications in Computer and Information Science}, volume = {1524}, pages = {409--420}, publisher = {Springer}, year = {2021}, url = {https://doi.org/10.1007/978-3-030-93736-2_32}, doi = {10.1007/978-3-030-93736-2\_32}, }
2020
- Towards Real-Time Single-Channel Singing-Voice Separation with Pruned Multi-Scaled Densenets2020 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2020, Barcelona, Spain, May 4-8, 2020, 806–810, IEEE, 2020
@inproceedings{DBLP:conf/icassp/HuberSSRPF20, author = {Huber, Markus and Schindler, G{\"{u}}nther and Sch{\"{o}}rkhuber, Christian and Roth, Wolfgang and Pernkopf, Franz and Fr{\"{o}}ning, Holger}, title = {Towards Real-Time Single-Channel Singing-Voice Separation with Pruned Multi-Scaled Densenets}, booktitle = {2020 {IEEE} International Conference on Acoustics, Speech and Signal Processing, {ICASSP} 2020, Barcelona, Spain, May 4-8, 2020}, pages = {806--810}, publisher = {{IEEE}}, year = {2020}, url = {https://doi.org/10.1109/ICASSP40776.2020.9053542}, doi = {10.1109/ICASSP40776.2020.9053542}, timestamp = {Tue, 21 Mar 2023 00:00:00 +0100}, }
- On Resource-Efficient Bayesian Network Classifiers and Deep Neural Networks25th International Conference on Pattern Recognition, ICPR 2020, Virtual Event / Milan, Italy, January 10-15, 2021, 10297–10304, IEEE, 2020
@inproceedings{DBLP:conf/icpr/RothPSF20, author = {Roth, Wolfgang and Pernkopf, Franz and Schindler, G{\"{u}}nther and Fr{\"{o}}ning, Holger}, title = {On Resource-Efficient Bayesian Network Classifiers and Deep Neural Networks}, booktitle = {25th International Conference on Pattern Recognition, {ICPR} 2020, Virtual Event / Milan, Italy, January 10-15, 2021}, pages = {10297--10304}, publisher = {{IEEE}}, year = {2020}, url = {https://doi.org/10.1109/ICPR48806.2021.9413156}, doi = {10.1109/ICPR48806.2021.9413156}, timestamp = {Tue, 21 Mar 2023 00:00:00 +0100}, }
- Parameterized Structured Pruning for Deep Neural NetworksMachine Learning, Optimization, and Data Science - 6th International Conference, LOD 2020, Siena, Italy, July 19-23, 2020, Revised Selected Papers, Part II (Lecture Notes in Computer Science), 12566, 16–27, Springer, 2020
@inproceedings{DBLP:conf/mod/SchindlerRPF20, author = {Schindler, G{\"{u}}nther and Roth, Wolfgang and Pernkopf, Franz and Fr{\"{o}}ning, Holger}, editor = {Nicosia, Giuseppe and Ojha, Varun and Malfa, Emanuele La and Jansen, Giorgio and Sciacca, Vincenzo and Pardalos, Panos M. and Giuffrida, Giovanni and Umeton, Renato}, title = {Parameterized Structured Pruning for Deep Neural Networks}, booktitle = {Machine Learning, Optimization, and Data Science - 6th International Conference, {LOD} 2020, Siena, Italy, July 19-23, 2020, Revised Selected Papers, Part {II}}, series = {Lecture Notes in Computer Science}, volume = {12566}, pages = {16--27}, publisher = {Springer}, year = {2020}, url = {https://doi.org/10.1007/978-3-030-64580-9\_3}, doi = {10.1007/978-3-030-64580-9\_3}, timestamp = {Tue, 21 Mar 2023 00:00:00 +0100}, }
- On the Difficulty of Designing Processor Arrays for Deep Neural NetworksIoT Streams for Data-Driven Predictive Maintenance and IoT, Edge, and Mobile for Embedded Machine Learning - Second International Workshop, IoT Streams 2020, and First International Workshop, ITEM 2020, Co-located with ECML/PKDD 2020, Ghent, Belgium, September 14-18, 2020, Revised Selected Papers (Communications in Computer and Information Science), 1325, 229–240, Springer, 2020
@inproceedings{DBLP:conf/pkdd/StehleSF20, author = {Stehle, Kevin and Schindler, G{\"{u}}nther and Fr{\"{o}}ning, Holger}, editor = {Gama, Jo{\~{a}}o and Pashami, Sepideh and Bifet, Albert and {Sayed Mouchaweh}, Moamar and Fr{\"{o}}ning, Holger and Pernkopf, Franz and Schiele, Gregor and Blott, Michaela}, title = {On the Difficulty of Designing Processor Arrays for Deep Neural Networks}, booktitle = {IoT Streams for Data-Driven Predictive Maintenance and IoT, Edge, and Mobile for Embedded Machine Learning - Second International Workshop, IoT Streams 2020, and First International Workshop, {ITEM} 2020, Co-located with {ECML/PKDD} 2020, Ghent, Belgium, September 14-18, 2020, Revised Selected Papers}, series = {Communications in Computer and Information Science}, volume = {1325}, pages = {229--240}, publisher = {Springer}, year = {2020}, url = {https://doi.org/10.1007/978-3-030-66770-2\_17}, doi = {10.1007/978-3-030-66770-2\_17}, timestamp = {Mon, 15 Feb 2021 00:00:00 +0100}, }
- Resource-Efficient Neural Networks for Embedded SystemsCoRR, abs/2001.03048, 2020
@article{DBLP:journals/corr/abs-2001-03048, author = {Roth, Wolfgang and Schindler, G{\"{u}}nther and Z{\"{o}}hrer, Matthias and Pfeifenberger, Lukas and Peharz, Robert and Tschiatschek, Sebastian and Fr{\"{o}}ning, Holger and Pernkopf, Franz and Ghahramani, Zoubin}, title = {Resource-Efficient Neural Networks for Embedded Systems}, journal = {CoRR}, volume = {abs/2001.03048}, year = {2020}, url = {http://arxiv.org/abs/2001.03048}, eprinttype = {arXiv}, eprint = {2001.03048}, timestamp = {Mon, 13 Jan 2020 00:00:00 +0100}, }
- Resource-Efficient Speech Mask Estimation for Multi-Channel Speech EnhancementCoRR, abs/2007.11477, 2020
@article{DBLP:journals/corr/abs-2007-11477, author = {Pfeifenberger, Lukas and Z{\"{o}}hrer, Matthias and Schindler, G{\"{u}}nther and Roth, Wolfgang and Fr{\"{o}}ning, Holger and Pernkopf, Franz}, title = {Resource-Efficient Speech Mask Estimation for Multi-Channel Speech Enhancement}, journal = {CoRR}, volume = {abs/2007.11477}, year = {2020}, url = {https://arxiv.org/abs/2007.11477}, eprinttype = {arXiv}, eprint = {2007.11477}, timestamp = {Wed, 29 Jul 2020 01:00:00 +0200}, }
2019
- Training Discrete-Valued Neural Networks with Sign Activations Using Weight DistributionsMachine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2019, Würzburg, Germany, September 16-20, 2019, Proceedings, Part II (Lecture Notes in Computer Science), 11907, 382–398, Springer, 2019
@inproceedings{DBLP:conf/pkdd/RothSFP19, author = {Roth, Wolfgang and Schindler, G{\"{u}}nther and Fr{\"{o}}ning, Holger and Pernkopf, Franz}, editor = {Brefeld, Ulf and {\'{E}}lisa Fromont and Hotho, Andreas and Knobbe, Arno J. and Maathuis, Marloes H. and Robardet, C{\'{e}}line}, title = {Training Discrete-Valued Neural Networks with Sign Activations Using Weight Distributions}, booktitle = {Machine Learning and Knowledge Discovery in Databases - European Conference, {ECML} {PKDD} 2019, W{\"{u}}rzburg, Germany, September 16-20, 2019, Proceedings, Part {II}}, series = {Lecture Notes in Computer Science}, volume = {11907}, pages = {382--398}, publisher = {Springer}, year = {2019}, url = {https://doi.org/10.1007/978-3-030-46147-8\_23}, doi = {10.1007/978-3-030-46147-8\_23}, timestamp = {Tue, 21 Mar 2023 00:00:00 +0100}, }
2018
- Resource Efficient Deep Eigenvector Beamforming2018 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2018, Calgary, AB, Canada, April 15-20, 2018, 3354–3358, IEEE, 2018
@inproceedings{DBLP:conf/icassp/ZohrerPSFP18, author = {Z{\"{o}}hrer, Matthias and Pfeifenberger, Lukas and Schindler, G{\"{u}}nther and Fr{\"{o}}ning, Holger and Pernkopf, Franz}, title = {Resource Efficient Deep Eigenvector Beamforming}, booktitle = {2018 {IEEE} International Conference on Acoustics, Speech and Signal Processing, {ICASSP} 2018, Calgary, AB, Canada, April 15-20, 2018}, pages = {3354--3358}, publisher = {{IEEE}}, year = {2018}, url = {https://doi.org/10.1109/ICASSP.2018.8462503}, doi = {10.1109/ICASSP.2018.8462503}, timestamp = {Wed, 16 Oct 2019 14:14:52 +0200}, }
- Towards Efficient Forward Propagation on Resource-Constrained SystemsMachine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2018, Dublin, Ireland, September 10-14, 2018, Proceedings, Part I (Lecture Notes in Computer Science), 11051, 426–442, Springer, 2018
@inproceedings{DBLP:conf/pkdd/SchindlerZPF18, author = {Schindler, G{\"{u}}nther and Z{\"{o}}hrer, Matthias and Pernkopf, Franz and Fr{\"{o}}ning, Holger}, editor = {Berlingerio, Michele and Bonchi, Francesco and G{\"{a}}rtner, Thomas and Hurley, Neil and Ifrim, Georgiana}, title = {Towards Efficient Forward Propagation on Resource-Constrained Systems}, booktitle = {Machine Learning and Knowledge Discovery in Databases - European Conference, {ECML} {PKDD} 2018, Dublin, Ireland, September 10-14, 2018, Proceedings, Part {I}}, series = {Lecture Notes in Computer Science}, volume = {11051}, pages = {426--442}, publisher = {Springer}, year = {2018}, url = {https://doi.org/10.1007/978-3-030-10925-7\_26}, doi = {10.1007/978-3-030-10925-7\_26}, timestamp = {Tue, 21 Mar 2023 00:00:00 +0100}, }
- Efficient and Robust Machine Learning for Real-World SystemsCoRR, abs/1812.02240, 2018
@article{DBLP:journals/corr/abs-1812-02240, author = {Pernkopf, Franz and Roth, Wolfgang and Z{\"{o}}hrer, Matthias and Pfeifenberger, Lukas and Schindler, G{\"{u}}nther and Fr{\"{o}}ning, Holger and Tschiatschek, Sebastian and Peharz, Robert and Mattina, Matthew and Ghahramani, Zoubin}, title = {Efficient and Robust Machine Learning for Real-World Systems}, journal = {CoRR}, volume = {abs/1812.02240}, year = {2018}, url = {http://arxiv.org/abs/1812.02240}, eprinttype = {arXiv}, eprint = {1812.02240}, timestamp = {Tue, 01 Jan 2019 00:00:00 +0100}, }
2017
- Linking Application Description with Efficient SIMD Code Generation for Low-Precision Signed-Integer GEMMEuro-Par 2017: Parallel Processing Workshops - Euro-Par 2017 International Workshops, Santiago de Compostela, Spain, August 28-29, 2017, Revised Selected Papers (Lecture Notes in Computer Science), 10659, 688–699, Springer, 2017
@inproceedings{DBLP:conf/europar/SchindlerMF17, author = {Schindler, G{\"{u}}nther and M{\"{u}}cke, Manfred and Fr{\"{o}}ning, Holger}, editor = {Heras, Dora Blanco and Boug{\'{e}}, Luc and Mencagli, Gabriele and Jeannot, Emmanuel and Sakellariou, Rizos and Badia, Rosa M. and Barbosa, Jorge G. and Ricci, Laura and Scott, Stephen L. and Lankes, Stefan and Weidendorfer, Josef}, title = {Linking Application Description with Efficient {SIMD} Code Generation for Low-Precision Signed-Integer {GEMM}}, booktitle = {Euro-Par 2017: Parallel Processing Workshops - Euro-Par 2017 International Workshops, Santiago de Compostela, Spain, August 28-29, 2017, Revised Selected Papers}, series = {Lecture Notes in Computer Science}, volume = {10659}, pages = {688--699}, publisher = {Springer}, year = {2017}, url = {https://doi.org/10.1007/978-3-319-75178-8\_55}, doi = {10.1007/978-3-319-75178-8\_55}, timestamp = {Thu, 14 Oct 2021 10:28:38 +0200}, }