pith. sign in

arxiv: 2508.12708 · v1 · submitted 2025-08-18 · ⚛️ physics.space-ph · astro-ph.IM· physics.ins-det

A Neural-Network Framework for Tracking and Identification of Cosmic-Ray Nuclei in the RadMap Telescope

Luise Meyer-Hetling , Martin J. Losekamm , Stephan Paul , Thomas P\"oschl This is my paper

Pith reviewed 2026-05-18 23:13 UTC · model grok-4.3

classification ⚛️ physics.space-ph astro-ph.IMphysics.ins-det
keywords neural networkcosmic-ray nucleitrajectory reconstructioncharge identificationenergy resolutionscintillating fibersspace radiation dosimetryGeant4 simulation
0
0 comments X

The pith

A neural network reconstructs cosmic-ray trajectories to better than 1.4 degrees and separates charges to 99.8 percent accuracy for hydrogen using simulated detector data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a neural-network framework to reconstruct the trajectories, charges, and energies of cosmic-ray nuclei as they pass through the scintillating-fiber tracking calorimeter of the RadMap Telescope. Training and testing rely entirely on events generated by a Geant4 simulation of a simplified detector model. The system delivers an angular resolution better than 1.4 degrees, charge identification above 95 percent for nuclei with atomic number up to 8, and energy resolution under 20 percent below 1 GeV per nucleon for elements up to iron. These performance figures are presented as sufficient to support accurate calculation of the biologically relevant radiation dose received by astronauts.

Core claim

A neural network trained on Geant4-simulated events from a simplified model of the RadMap Telescope's scintillating-fiber calorimeter reconstructs particle trajectories with angular resolution better than 1.4 degrees, separates charges with better than 95 percent accuracy for nuclei with Z less than or equal to 8 (reaching 99.8 percent for hydrogen), and achieves energy resolution below 20 percent for energies under 1 GeV per nucleon up to iron, thereby providing the spectroscopic information needed to determine the radiation dose astronauts experience in space.

What carries the argument

Neural network that ingests hit patterns from the scintillating fibers and outputs estimates of trajectory direction, nuclear charge, and kinetic energy per nucleon.

If this is right

  • The reported resolutions enable direct calculation of the biologically weighted radiation dose from measured cosmic-ray spectra.
  • The same network architecture can be retrained on improved detector models that include more hardware details.
  • Charge and energy estimates can be combined with trajectory information to separate primary cosmic rays from secondary particles produced in the spacecraft.
  • The framework supplies a concrete benchmark for comparing neural-network methods against traditional track-fitting algorithms in fiber calorimeters.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be adapted to other fiber-based trackers in future space missions without requiring new simulation campaigns for each detector geometry.
  • Combining the network outputs with real-time telemetry from the International Space Station would allow continuous monitoring of solar particle events.
  • The approach suggests that similar networks might identify heavier nuclei beyond iron if the training set is extended to higher-Z particles.
  • Validation against independent ground-based accelerator beams would strengthen confidence before flight deployment.

Load-bearing premise

The simplified Geant4 detector model used to generate all training and test data produces event topologies and light-yield distributions that are sufficiently close to those of the actual RadMap Telescope hardware.

What would settle it

Running the trained network on real flight data from the RadMap Telescope and finding that the charge-separation accuracy for hydrogen falls below 95 percent or the energy resolution exceeds 20 percent for nuclei below 1 GeV per nucleon would falsify the performance claims.

Figures

Figures reproduced from arXiv: 2508.12708 by Luise Meyer-Hetling, Martin J. Losekamm, Stephan Paul, Thomas P\"oschl.

Figure 1
Figure 1. Figure 1: Schematic representation of the RadMap Telescope’s main detector, which consists of a stack of 1024 scintillating plastic fibers. They are arranged in a stack of 32 layers of alternating orientation. Also shown are the coordinate system and the spherical coordinates ϕ and θ we use to parametrize the orientation of particle tracks. The red and blue arrows indicate the two-dimensional projections we use to v… view at source ↗
Figure 2
Figure 2. Figure 2: Simulated event signatures of a 540-MeV proton (top) and a 1.5-GeV/n iron nucleus (bot￾tom), shown in the yx- (left) and yz-projections (right) illustrated in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the parameters that can be determined from the energy-deposition profiles of cosmic-ray nuclei. The dashed curves show the mean energy deposition, ⟨ dE dx ⟩, of protons with differ￾ent energies in high-density polyethylenea . The energy-deposition profile of a stopping particle en￾codes its nuclear charge, Z, mass number, A, velocity, β, and kinetic energy, Ekin. For through-going particles… view at source ↗
Figure 4
Figure 4. Figure 4: Architectures of the neural networks we use. Each circle represents a network layer and the color indicates the layer type. Dense and convolutional layers are labeled with their total size and their filter size, respectively. The network for track reconstruction has about 2.8 million trainable parameters, the one for charge determination and energy measurement about 2.1 million. reconstruction steps depend… view at source ↗
Figure 5
Figure 5. Figure 5: Reconstruction performance for the track angles ϕ (left) and θ (right) for protons and for iron nuclei, shown as the differences (∆ϕ and ∆θ) between the reconstructed angles (θrec and ϕrec) and the true angles (θmc and ϕmc). The blue histograms show the performance using networks trained to reconstruct the tracks of stopping particles. The yellow histograms show the corresponding results for monoenergetic … view at source ↗
Figure 6
Figure 6. Figure 6: Dependence of σ on the number of fibers with signal per event for stopping particles (blue) and minimum-ionizing particles (red). Values of σ∆ϕ are multiples of the neural network’s output resolution (0.2 ◦ ); σ∆θ is calculated via Equation 1. The number of traversed fibers can be larger than the depth of the ADU (32 fibers/layers) due to the emission of secondary particles (see [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 7
Figure 7. Figure 7: Confusion matrix of the charge determination via the consecutive application of two neural networks, trained to identify light nuclei through oxygen (Z = 8) and heavy nuclei through iron (Z = 26), respectively. The second network was trained (but not tested) on a data set containing cobalt (Z = 27). Each column shows the distribution of reconstructed nuclear charge, Zrec, for all events with true nuclear c… view at source ↗
Figure 8
Figure 8. Figure 8: The accuracy (left panel) and purity (bottom right panel) of the charge determination as a function of Z. We define the accuracy as the fraction of correctly identified events for a given Zmc and the purity as the fraction of events in a Zrec class for which Zrec = Zmc. The top right panel shows the standard deviation, σZ, of the reconstructed-charge distributions (the columns in [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 9
Figure 9. Figure 9: Bin-wise energy resolution for protons (hydrogen) with energies from 20 MeV to 1 GeV. The left panel shows the results if we apply no selection cuts to our test data, the right panel those for events with N yx sig ≥ 3 and N yz sig ≥ 3. The blue histograms give the performance of a single network trained over the full data set of stopping and through-going hydrogen nuclei; the yellow and red histograms that… view at source ↗
Figure 10
Figure 10. Figure 10: Energy resolution for hydrogen (H), helium (He), carbon (C), and iron (Fe) nuclei with energies from 20 MeV/n to 1 GeV/n (10 GeV/n for iron). The yellow histograms show the per￾formance for events of the target charge which we manually selected, the blue ones the result for particles that were selected by our charge-determination network from a test data set containing all elements up to cobalt. For heliu… view at source ↗
read the original abstract

We present a neural-network framework designed to reconstruct the properties of cosmic-ray nuclei traversing the scintillating-fiber tracking calorimeter of the RadMap Telescope. Employing the Geant4 simulation toolkit and a simplified model of the detector to generate training and test data, we achieve the spectroscopic capabilities required for an accurate determination of the biologically relevant dose that astronauts receive in space. We can reconstruct a particle's trajectory with an angular resolution of better than $1.4^\circ$ and achieve a charge separation of better than $95\%$ for nuclei with $Z\leq8$; specifically, we reach an accuracy of $99.8\%$ for hydrogen. The energy resolution is $<20\%$ for energies below 1 GeV/n and elements up to iron. We also discuss the limitations of our detector, the reconstruction framework, and this feasibility study, as well as possible improvements.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript presents a neural-network framework for reconstructing trajectory, charge, and energy of cosmic-ray nuclei in the RadMap Telescope's scintillating-fiber tracking calorimeter. All training and test data are generated via Geant4 using a simplified detector model; the authors report angular resolution better than 1.4°, charge separation >95% for Z≤8 (99.8% for hydrogen), and energy resolution <20% below 1 GeV/n up to iron. The work is framed as a feasibility study discussing limitations and possible improvements.

Significance. If the simulation faithfully reproduces the physical detector response, the multi-task neural network could provide an efficient reconstruction tool for space-based cosmic-ray dosimetry. The approach of jointly optimizing tracking and particle identification is a reasonable application of modern ML to this domain. However, the complete absence of any quantitative sim-to-data comparison substantially reduces the immediate significance for actual RadMap Telescope operations.

major comments (1)
  1. [Methods (Geant4 simulation)] Methods section on Geant4 simulation and data generation: all quoted performance metrics (angular resolution <1.4°, charge accuracy >95% for Z≤8, energy resolution <20%) are evaluated exclusively on held-out events from the simplified detector model. No quantitative comparison is presented between simulated light-yield distributions, hit multiplicities, or track residuals and any calibration or flight data from the real RadMap Telescope hardware. This gap directly affects the transferability of the claimed resolutions to the instrument.
minor comments (2)
  1. [Abstract] Abstract and introduction: the claim that the framework achieves 'the spectroscopic capabilities required for an accurate determination of the biologically relevant dose' would benefit from a brief statement of the target dose precision or reference to the relevant radiation-protection standards.
  2. [Results] Results section: the manuscript could clarify whether the reported resolutions include uncertainty quantification (e.g., bootstrap or ensemble estimates) or are point estimates from a single network training run.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their constructive review of our manuscript. We agree that the lack of real-data validation is a limitation for immediate operational transferability and have revised the text to better emphasize this point while defending the value of the simulation-based feasibility study.

read point-by-point responses

  1. Referee: [Methods (Geant4 simulation)] Methods section on Geant4 simulation and data generation: all quoted performance metrics (angular resolution <1.4°, charge accuracy >95% for Z≤8, energy resolution <20%) are evaluated exclusively on held-out events from the simplified detector model. No quantitative comparison is presented between simulated light-yield distributions, hit multiplicities, or track residuals and any calibration or flight data from the real RadMap Telescope hardware. This gap directly affects the transferability of the claimed resolutions to the instrument.

    Authors: We agree that all performance metrics are obtained from held-out simulated events generated with a simplified Geant4 model and that no quantitative sim-to-data comparison with real RadMap Telescope calibration or flight data is presented. This is inherent to the current feasibility study, as the instrument is still in development and no such experimental datasets exist for direct comparison at this time. The reported figures therefore represent performance under idealized conditions and should be interpreted as an upper bound. In the revised manuscript we have added a dedicated paragraph in the Discussion section that qualitatively addresses expected differences between simulation and hardware (e.g., fiber attenuation, PMT non-uniformity, electronic noise, and hit-multiplicity variations) and explicitly cautions readers about transferability. We also outline a path for future validation once real data become available. We believe these changes directly respond to the concern while preserving the manuscript’s focus on demonstrating the neural-network framework. revision: partial

standing simulated objections not resolved
  • Quantitative comparison of simulated light-yield distributions, hit multiplicities, or track residuals against actual calibration or flight data from the RadMap Telescope hardware, as no such real detector data are currently available.

Circularity Check

0 steps flagged

No circularity detected in the derivation chain

full rationale

The paper generates synthetic training and test events via a Geant4 simulation of the detector geometry and response, trains a neural network to regress trajectory, charge, and energy, then reports resolution and accuracy figures by comparing network outputs against the known ground-truth labels of the held-out simulated events. These performance numbers are therefore independent measurements rather than quantities defined by the network itself or recovered by construction from the training procedure. No self-definitional equations, fitted parameters relabeled as predictions, load-bearing self-citations, or ansatzes imported from prior author work appear in the reported chain. The framework is a self-contained simulation-based feasibility study whose internal logic does not collapse to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

All performance claims rest on the assumption that a simplified Geant4 model captures the dominant detector response; no independent experimental calibration data are invoked.

axioms (1)
  • domain assumption Geant4 simulation with simplified detector model produces training data representative of real RadMap Telescope response
    Explicitly stated as the source of all training and test events

pith-pipeline@v0.9.0 · 5700 in / 1262 out tokens · 53599 ms · 2026-05-18T23:13:11.680200+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Geant4–a simulation toolkit. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,506(3), 250–303.http://dx.doi.org/10.1016/s0168-9002(03) 01368-8. 3.4 Alaghband, Y ., P. M. Klein, E. A. Kramar, M. N. Cranston, B. C. Perry, et al.,

    work page doi:10.1016/s0168-9002(03

  2. [2]

    1 21 Meyer-Hetling et al.: A Neural-Network Framework for the RadMap Telescope Allison, J., K

    Galactic cosmic radiation exposure causes multifaceted neurocognitive impairments.Cellular and Molecular Life Sciences,80(1), 29.http://dx.doi.org/10.1007/s00018-022-04666-8. 1 21 Meyer-Hetling et al.: A Neural-Network Framework for the RadMap Telescope Allison, J., K. Amako, J. Apostolakis, P. Arce, M. Asai, et al.,

    work page doi:10.1007/s00018-022-04666-8

  3. [3]

    Recent developments in Geant4. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,835, 186–225.http://dx.doi.org/10.1016/j.nima.2016.06.125. 3.4 Barthel, J., and N. Sarigul-Klijn,

    work page doi:10.1016/j.nima.2016.06.125 2016

  4. [4]

    1016/j.paerosci.2019.100553

    A review of radiation shielding needs and concepts for space voyages beyond Earth’s magnetic influence.Progress in Aerospace Sciences,110.http://dx.doi.org/10. 1016/j.paerosci.2019.100553. 1 Caldwell, A., D. Koll ´ar, and K. Kr ¨oninger,

    work page arXiv 2019

  5. [5]

    3.3 Chancellor, J

    BAT – The Bayesian analysis toolkit.Computer Physics Communications,180(11), 2197–2209.http://dx.doi.org/10.1016/j.cpc.2009.06.026. 3.3 Chancellor, J. C., R. S. Blue, K. A. Cengel, S. M. Au ˜n´on-Chancellor, K. H. Rubins, H. G. Katzgraber, and A. R. Kennedy,

    work page doi:10.1016/j.cpc.2009.06.026 2009

  6. [6]

    npj Microgravity,4, 8.http://dx.doi.org/10.1038/s41526-018-0043-2

    Limitations in predicting the space radiation health risk for exploration astronauts. npj Microgravity,4, 8.http://dx.doi.org/10.1038/s41526-018-0043-2. 1, 2 Chancellor, J. C., G. Scott, and J. Sutton,

    work page doi:10.1038/s41526-018-0043-2

  7. [7]

    1 Cucinotta, F

    Space Radiation: The Number One Risk to Astronaut Health beyond Low Earth Orbit.Life,4, 491–510.http://dx.doi.org/10.3390/life4030491. 1 Cucinotta, F. A.,

    work page doi:10.3390/life4030491

  8. [8]

    2 Cucinotta, F

    A New Approach to Reduce Uncertainties in Space Radiation Cancer Risk Predictions.PLoS ONE,10, e0120,717.http://dx.doi.org/10.1371/journal.pone.0120717. 2 Cucinotta, F. A., M. Alp, B. Rowedder, and M.-H. Y . Kim,

    work page doi:10.1371/journal.pone.0120717

  9. [9]

    1 Cucinotta, F

    Safe days in space with accept- able uncertainty from space radiation exposure.Life Sciences in Space Research,5, 31–38.http: //dx.doi.org/10.1016/j.lssr.2015.04.002. 1 Cucinotta, F. A., K. To, and E. Cacao,

    work page doi:10.1016/j.lssr.2015.04.002 2015

  10. [10]

    Predictions of space radiation fatality risk for exploration missions.Life Sciences in Space Research,13, 1–11.http://dx.doi.org/10.1016/j.lssr.2017. 01.005. 2 Delp, M. D., J. M. Charvat, C. L. Limoli, R. K. Globus, and P. Ghosh,

    work page doi:10.1016/j.lssr.2017 2017

  11. [11]

    1 Durante, M., and F

    Apollo Lunar Astronauts Show Higher Cardiovascular Disease Mortality: Possible Deep Space Radiation Effects on the Vascular Endothelium.Scientific Reports,6, 29,901.http://dx.doi.org/10.1038/srep29901. 1 Durante, M., and F. A. Cucinotta,

    work page doi:10.1038/srep29901

  12. [12]

    1 Fogtman, A., S

    Physical basis of radiation protection in space travel.Reviews of Modern Physics,83, 1245–1281.http://dx.doi.org/10.1103/RevModPhys.83.1245. 1 Fogtman, A., S. Baatout, B. Baselet, T. Berger, C. E. Hellweg, et al.,

    work page doi:10.1103/revmodphys.83.1245

  13. [13]

    1 Gruhn, C., M

    Towards sustainable human space exploration–priorities for radiation research to quantify and mitigate radiation risks.npj Microgravity, 9(1).http://dx.doi.org/10.1038/s41526-023-00262-7. 1 Gruhn, C., M. Binimi, R. Legrain, R. Loveman, W. Pang, et al.,

    work page doi:10.1038/s41526-023-00262-7

  14. [14]

    Neoplasia,32, 100,828.http://dx.doi.org/10.1016/j.neo.2022.100828

    Carcinogenesis induced by space radiation: A systematic review. Neoplasia,32, 100,828.http://dx.doi.org/10.1016/j.neo.2022.100828. 1 Hellweg, C. E., T. Berger, D. Matthi¨a, and C. Baumstark-Khan,

    work page doi:10.1016/j.neo.2022.100828 2022

  15. [15]

    SpringerBriefs in Space Life Sciences

    Radiation in Space: Relevance and Risk for Human Missions. SpringerBriefs in Space Life Sciences. Springer, Cham. ISBN 978-3-030-46743-2. http://dx.doi.org/10.1007/978-3-030-46744-9. 2 22 Meyer-Hetling et al.: A Neural-Network Framework for the RadMap Telescope Hollender, L.,

    work page doi:10.1007/978-3-030-46744-9

  16. [16]

    1 Koontz, S

    Detrimental impacts of mixed- ion radiation on nervous system function.Neurobiology of Disease,151, 105,252.http://dx.doi.org/ 10.1016/j.nbd.2021.105252. 1 Koontz, S. L., P. A. Boeder, C. Pankop, and B. Reddell,

    work page doi:10.1016/j.nbd.2021.105252 2021

  17. [17]

    expectations for avionics and materials

    The ionizing radiation environment on the International Space Station: performance vs. expectations for avionics and materials. In IEEE Radiation Effects Data Workshop, 110–116. IEEE, Seattle, W A.http://dx.doi.org/10.1109/redw.2005. 1532675. 3.4 Lecun, Y ., L. Bottou, Y . Bengio, and P. Haffner,

    work page doi:10.1109/redw.2005 2005

  18. [18]

    Gradient-based learning applied to document recognition

    Gradient-based learning applied to document recog- nition.Proceedings of the IEEE,86(11), 2278–2324.http://dx.doi.org/10.1109/5.726791. 3.3 Longair, M. S.,

    work page doi:10.1109/5.726791

  19. [19]

    Cambridge University Press, Cambridge, 3rd edn

    High Energy Astrophysics. Cambridge University Press, Cambridge, 3rd edn. ISBN 9780521756181.http://dx.doi.org/10.1017/cbo9780511778346. 2 Losekamm, M. J.,

    work page doi:10.1017/cbo9780511778346

  20. [20]

    In Proceedings of 38th International Cosmic Ray Conference – PoS(ICRC2023).http://dx.doi.org/10.22323/1.444.0099

    Measuring Cosmic Rays with the RadMap Telescope on the International Space Station. In Proceedings of 38th International Cosmic Ray Conference – PoS(ICRC2023).http://dx.doi.org/10.22323/1.444.0099. 1 Losekamm, M. J., M. Milde, T. P ¨oschl, D. Greenwald, and S. Paul,

    work page doi:10.22323/1.444.0099

  21. [21]

    A new analysis method using Bragg curve spectroscopy for a Multi-purpose Active-target Particle Telescope for radiation monitoring. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,845, 520–523.http://dx.doi.org/10.1016/j.nima.2016.05.029. 3.3 Losekamm, M. J., S. Paul, T. P ¨osc...

    work page doi:10.1016/j.nima.2016.05.029 2016

  22. [22]

    In 2021 IEEE Aerospace Conference (50100)

    The RadMap Telescope on the International Space Station. In 2021 IEEE Aerospace Conference (50100). IEEE, Big Sky, MT.http://dx.doi.org/ 10.1109/AERO50100.2021.9438435. 1 Loshchilov, I., and F. Hutter,

    work page doi:10.1109/aero50100.2021.9438435 2021

  23. [23]

    Decoupled Weight Decay Regularization

    Decoupled Weight Decay Regularization. 10.48550/arXiv.1711.05101, 1711.05101. 4 Mao, X. W., M. J. Pecaut, and D. S. Gridley. Acute Risks of Space Radiation, chap. 27, 263–276. Springer, Cham,

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.05101

  24. [24]

    2 Milde, M.,

    ISBN 978-3-319-12190-1.http://dx.doi.org/10.1007/978-3-319-12191-8_27. 2 Milde, M.,

    work page doi:10.1007/978-3-319-12191-8_27

  25. [25]

    org/10.3390/life11080829

    Space Radiation Protection Countermeasures in Microgravity and Planetary Exploration.Life,11(8).http://dx.doi. org/10.3390/life11080829. 1 Naito, M., and S. Kodaira,

    work page doi:10.3390/life11080829

  26. [26]

    The National Academies Press, Washington, DC

    Space Radiation and Astronaut Health: Managing and Communicating Cancer Risks. The National Academies Press, Washington, DC. ISBN 978-0-309-47966-0.http://dx.doi.org/10.17226/26155. 1 Parihar, V . K., B. Allen, K. K. Tran, T. G. Macaraeg, E. M. Chu, et al.,

    work page doi:10.17226/26155

  27. [27]

    1 Prechelt, L

    What happens to your brain on the way to Mars.Science Advances,1, e1400,256.http://dx.doi.org/10.1126/sciadv.1400256. 1 Prechelt, L. Early Stopping - But When?, chap. 3, 55–69. Lecture Notes in Computer Science. Springer, Berlin, Heidelberg,

    work page doi:10.1126/sciadv.1400256

  28. [28]

    3.2 Raissi, M., P

    Measurement of ionization quenching in plastic scintillators.Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,988, 164,865.http://dx.doi.org/10.1016/ j.nima.2020.164865. 3.2 Raissi, M., P. Perdikaris, and G. Karniadakis,

    work page arXiv 2020

  29. [29]

    Journal of Computational Physics,378, 686–707.http://dx.doi.org/10.1016/j.jcp.2018.10

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics,378, 686–707.http://dx.doi.org/10.1016/j.jcp.2018.10

    work page doi:10.1016/j.jcp.2018.10 2018

  30. [30]

    Lecture Notes in Physics

    Solar Energetic Particles. Lecture Notes in Physics. Springer, Cham, 2nd edn. ISBN 978-3-030-66401-5.http://dx.doi.org/10.1007/978-3-030-66402-2. 2 Reames, D. V .,

    work page doi:10.1007/978-3-030-66402-2

  31. [31]

    1 Scarselli, F., M

    Radiation induced cardiovascular disease: An odyssey of bedside-bench-bedside approach.Life Sciences in Space Research, 27, 49–55.http://dx.doi.org/10.1016/j.lssr.2020.07.005. 1 Scarselli, F., M. Gori, A. C. Tsoi, M. Hagenbuchner, and G. Monfardini,

    work page doi:10.1016/j.lssr.2020.07.005 2020

  32. [32]

    2008.2005605

    The Graph Neural Network Model.IEEE Transactions on Neural Networks,20(1), 61–80.http://dx.doi.org/10.1109/TNN. 2008.2005605. 7 Slaba, T. C., A. A. Bahadori, B. D. Reddell, R. C. Singleterry, M. S. Clowdsley, and S. R. Blattnig,

    work page doi:10.1109/tnn 2008

  33. [33]

    http://dx.doi.org/10.1016/j.lssr.2016.12.003

    Optimal shielding thickness for galactic cosmic ray environments.Life Sciences in Space Research,12. http://dx.doi.org/10.1016/j.lssr.2016.12.003. 7 24 Meyer-Hetling et al.: A Neural-Network Framework for the RadMap Telescope Szegedy, C., W. Liu, Y . Jia, P. Sermanet, S. Reed, and D. Anguelov,

    work page doi:10.1016/j.lssr.2016.12.003 2016

  34. [34]

    In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Going deeper with convolutions. In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, Boston, MA. 3.3 van Laarhoven, P. J. M., and E. H. L. Aarts,

    work page 2015

  35. [35]

    Mathematics and Its Applications

    Simulated Annealing: Theory and Applications. Mathematics and Its Applications. Springer, Dordrecht. ISBN 978-90-481-8438-5.http://dx.doi. org/10.1007/978-94-015-7744-1. 3.3 Vaswani, A., N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin,

    work page doi:10.1007/978-94-015-7744-1

  36. [36]

    1 Zeitlin, C., and C

    Research plans in Europe for radiation health hazard assessment in exploratory space missions.Life Sciences in Space Research,21, 73–82.http://dx.doi.org/10.1016/j.lssr.2019.04.002. 1 Zeitlin, C., and C. La Tessa,

    work page doi:10.1016/j.lssr.2019.04.002 2019

  37. [37]

    The Role of Nuclear Fragmentation in Particle Therapy and Space Radiation Protection.Frontiers in Oncology,6, 1–13.http://dx.doi.org/10.3389/fonc.2016. 00065. 3.4 25

    work page doi:10.3389/fonc.2016 2016