Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure
Deep learning models are defined in terms of a large number of hyperparameters, such as network architectures and optimiser settings. These hyperparameters must be determined separately from the model parameters such as network weights, and are often fixed by ad-hoc methods or by manual inspection o...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IOP Publishing
2025-01-01
|
| Series: | Machine Learning: Science and Technology |
| Subjects: | |
| Online Access: | https://doi.org/10.1088/2632-2153/adcd39 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849311506571722752 |
|---|---|
| author | Juan Cruz-Martinez Aron Jansen Gijs van Oord Tanjona R Rabemananjara Carlos M R Rocha Juan Rojo Roy Stegeman |
| author_facet | Juan Cruz-Martinez Aron Jansen Gijs van Oord Tanjona R Rabemananjara Carlos M R Rocha Juan Rojo Roy Stegeman |
| author_sort | Juan Cruz-Martinez |
| collection | DOAJ |
| description | Deep learning models are defined in terms of a large number of hyperparameters, such as network architectures and optimiser settings. These hyperparameters must be determined separately from the model parameters such as network weights, and are often fixed by ad-hoc methods or by manual inspection of the results. An algorithmic, objective determination of hyperparameters demands the introduction of dedicated target metrics, different from those adopted for the model training. Here we present a new approach to the automated determination of hyperparameters in deep learning models based on statistical estimators constructed from a ensemble of models sampling the underlying probability distribution in model space. This strategy requires the simultaneous parallel training of up to several hundreds of models and can be effectively implemented by deploying hardware accelerators such as graphical processing units (GPUs). As a proof-of-concept, we apply this method to the determination of the partonic substructure of the proton within the NNPDF framework and demonstrate the robustness of the resultant model uncertainty estimates. The new GPU-optimised NNPDF code results in a speed-up of up to two orders of magnitude, a stabilisation of the memory requirements, and a reduction in energy consumption of up to 90% as compared to sequential CPU-based model training. While focusing on proton structure, our method is fully general and is applicable to any deep learning problem relying on hyperparameter optimisation for an ensemble of models. |
| format | Article |
| id | doaj-art-fba0e9b8d0d446df8a59c826129e0055 |
| institution | Kabale University |
| issn | 2632-2153 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IOP Publishing |
| record_format | Article |
| series | Machine Learning: Science and Technology |
| spelling | doaj-art-fba0e9b8d0d446df8a59c826129e00552025-08-20T03:53:22ZengIOP PublishingMachine Learning: Science and Technology2632-21532025-01-016202502710.1088/2632-2153/adcd39Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structureJuan Cruz-Martinez0https://orcid.org/0000-0002-8061-1965Aron Jansen1Gijs van Oord2Tanjona R Rabemananjara3https://orcid.org/0000-0002-8395-8059Carlos M R Rocha4https://orcid.org/0000-0002-4118-8308Juan Rojo5https://orcid.org/0000-0003-4279-2192Roy Stegeman6https://orcid.org/0000-0002-3852-8009Theoretical Physics Department, CERN , CH-1211 Geneva 23, SwitzerlandNetherlands eScience Center , Science Park 140, 1098 XG Amsterdam, The NetherlandsNetherlands eScience Center , Science Park 140, 1098 XG Amsterdam, The NetherlandsDepartment of Physics and Astronomy, Vrije Universiteit , NL-1081 HV Amsterdam, The Netherlands; Nikhef Theory Group , Science Park 105, 1098 XG Amsterdam, The NetherlandsNetherlands eScience Center , Science Park 140, 1098 XG Amsterdam, The NetherlandsTheoretical Physics Department, CERN , CH-1211 Geneva 23, Switzerland; Department of Physics and Astronomy, Vrije Universiteit , NL-1081 HV Amsterdam, The Netherlands; Nikhef Theory Group , Science Park 105, 1098 XG Amsterdam, The NetherlandsThe Higgs Centre for Theoretical Physics, University of Edinburgh , JCMB, KB, Mayfield Rd, Edinburgh EH9 3FD, United KingdomDeep learning models are defined in terms of a large number of hyperparameters, such as network architectures and optimiser settings. These hyperparameters must be determined separately from the model parameters such as network weights, and are often fixed by ad-hoc methods or by manual inspection of the results. An algorithmic, objective determination of hyperparameters demands the introduction of dedicated target metrics, different from those adopted for the model training. Here we present a new approach to the automated determination of hyperparameters in deep learning models based on statistical estimators constructed from a ensemble of models sampling the underlying probability distribution in model space. This strategy requires the simultaneous parallel training of up to several hundreds of models and can be effectively implemented by deploying hardware accelerators such as graphical processing units (GPUs). As a proof-of-concept, we apply this method to the determination of the partonic substructure of the proton within the NNPDF framework and demonstrate the robustness of the resultant model uncertainty estimates. The new GPU-optimised NNPDF code results in a speed-up of up to two orders of magnitude, a stabilisation of the memory requirements, and a reduction in energy consumption of up to 90% as compared to sequential CPU-based model training. While focusing on proton structure, our method is fully general and is applicable to any deep learning problem relying on hyperparameter optimisation for an ensemble of models.https://doi.org/10.1088/2632-2153/adcd39hyperoptimisationmachine learninghardware accelerationproton structureparton distribution functionsGPU |
| spellingShingle | Juan Cruz-Martinez Aron Jansen Gijs van Oord Tanjona R Rabemananjara Carlos M R Rocha Juan Rojo Roy Stegeman Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure Machine Learning: Science and Technology hyperoptimisation machine learning hardware acceleration proton structure parton distribution functions GPU |
| title | Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure |
| title_full | Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure |
| title_fullStr | Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure |
| title_full_unstemmed | Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure |
| title_short | Hyperparameter optimisation in deep learning from ensemble methods: applications to proton structure |
| title_sort | hyperparameter optimisation in deep learning from ensemble methods applications to proton structure |
| topic | hyperoptimisation machine learning hardware acceleration proton structure parton distribution functions GPU |
| url | https://doi.org/10.1088/2632-2153/adcd39 |
| work_keys_str_mv | AT juancruzmartinez hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT aronjansen hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT gijsvanoord hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT tanjonarrabemananjara hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT carlosmrrocha hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT juanrojo hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure AT roystegeman hyperparameteroptimisationindeeplearningfromensemblemethodsapplicationstoprotonstructure |