pith. sign in

arxiv: 2603.01655 · v2 · pith:ILG4YYBQnew · submitted 2026-03-02 · 💻 cs.LG · eess.SP

Transform-Invariant Generative Ray Path Sampling for Efficient Radio Propagation Modeling

J\'erome Eertmans , Enrico M. Vitucci , Vittorio Degli-Esposti , Nicola Di Cicco , Laurent Jacques , Claude Oestges This is my paper

Pith reviewed 2026-05-25 06:33 UTC · model grok-4.3

classification 💻 cs.LG eess.SP
keywords ray tracinggenerative flow networksradio propagationpath samplingmachine learningurban modelingcomputational efficiencysparse rewards
0
0 comments X

The pith

Generative Flow Networks sample valid radio ray paths instead of exhaustive enumeration, delivering up to 100x speedups on CPU while retaining high coverage accuracy.

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

The paper shows how Generative Flow Networks can be trained to sample ray paths for radio propagation modeling, replacing the exponential enumeration of all possible interaction sequences with learned intelligent sampling. Three additions—an experience replay buffer that stores rare valid paths, a uniform exploratory policy that prevents collapse to simple geometries, and physics-based action masking that eliminates impossible actions—enable the model to handle the sparse-reward problem that otherwise causes training failure at high interaction orders. On idealized street-canyon test cases the resulting sampler runs substantially faster than exhaustive search yet still recovers the important propagation paths. The same model, however, exhibits clear degradation when evaluated on real Manhattan geometries that differ in layout statistics from the training distribution.

Core claim

Generative Flow Networks equipped with an experience replay buffer, uniform exploratory policy, and physics-based action masking can learn to sample physically valid radio propagation paths, achieving up to 10× GPU and 100× CPU speedups over exhaustive search on street-canyon scenarios while maintaining high coverage accuracy.

What carries the argument

Generative Flow Networks for path sampling, augmented by experience replay, uniform exploration, and physics-based masking to manage sparse rewards.

If this is right

  • Radio-propagation tools can move from heuristic pruning to learned sampling without sacrificing coverage of high-order paths.
  • Real-time or city-scale ray-tracing simulations become computationally practical on commodity hardware.
  • The same sampling framework can be applied to any wave-propagation or particle-interaction problem whose candidate sequences grow exponentially with interaction order.

Where Pith is reading between the lines

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

  • The out-of-distribution drop on Manhattan geometry suggests that future work could test whether curriculum training on progressively more varied city layouts closes the gap.
  • The approach may transfer to analogous exponential-search tasks such as neutron transport or acoustic ray tracing once the masking and replay mechanisms are adapted to the new physics.
  • Hybrid systems that first use the learned sampler to propose candidate paths and then verify them with exact ray-tracing could further reduce wall-clock time while preserving deterministic correctness.

Load-bearing premise

The three added components—experience replay buffer, uniform exploratory policy, and physics-based action masking—are together sufficient to produce stable learning and useful generalization despite extremely sparse valid-path rewards.

What would settle it

Training the same GFlowNet architecture on the reported street-canyon scenarios without any one of the three components and observing either convergence to trivial solutions or failure to recover the known valid paths would falsify the claim that the combination suffices.

read the original abstract

Ray tracing has become a standard for accurate radio propagation modeling, but suffers from exponential computational complexity, as the number of candidate paths scales with the number of objects raised to the interaction order. This bottleneck limits its use in large-scale or real-time applications, forcing traditional tools to rely on heuristics that reduce path candidates at the cost of potentially reduced accuracy. To overcome this limitation, we propose a machine-learning-assisted framework that replaces exhaustive path searching with intelligent sampling via Generative Flow Networks. Applying these generative models to this domain presents challenges, particularly sparse rewards due to the rarity of valid paths, which can lead to convergence failures and trivial solutions when evaluating high-order interactions in complex environments. To ensure robust learning and efficient exploration, our framework incorporates three key components. First, an \emph{experience replay buffer} captures and retains rare valid paths. Second, a uniform exploratory policy improves generalization and prevents overfitting to simple geometries. Third, a physics-based action masking strategy filters out physically impossible paths before the model considers them. Validated on idealized street-canyon scenarios, our model achieves substantial speedups over exhaustive search -- up to $10\times$ faster on GPU and $100\times$ faster on CPU -- while maintaining high coverage accuracy and successfully uncovering complex propagation paths. However, out-of-distribution evaluations on real-world Manhattan street geometries reveal that generalizing to substantially different urban morphologies requires further advancement in model capacity or alternative training strategies. Source code, tests, and a tutorial are available at https://github.com/jeertmans/sampling-paths.

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

0 major / 3 minor

Summary. The manuscript introduces a Generative Flow Networks framework for sampling ray paths in radio propagation modeling, replacing exhaustive search to address exponential complexity in ray tracing. It incorporates an experience replay buffer, uniform exploratory policy, and physics-based action masking to mitigate sparse rewards and convergence issues in high-order interactions. Validation is reported on idealized street-canyon scenarios, claiming up to 10× GPU and 100× CPU speedups over exhaustive search with high coverage accuracy and discovery of complex paths; out-of-distribution tests on Manhattan geometries show generalization failures attributed to model capacity. Public code, tests, and a tutorial are provided.

Significance. If the reported speedups and accuracy hold on the scoped idealized cases, the approach could enable more efficient radio propagation modeling for targeted urban geometries, with the open-source release supporting reproducibility and extension. The explicit scoping to idealized scenarios and disclosure of OOD limitations strengthen the paper's internal consistency, though broader applicability would require further model advances.

minor comments (3)
  1. The title references 'Transform-Invariant' sampling, but neither the abstract nor the provided description explains the transform-invariance mechanism, its implementation, or its contribution to the results; this should be clarified in the methods section.
  2. The abstract claims 'high coverage accuracy' and 'successfully uncovering complex propagation paths' on idealized scenarios; specific quantitative metrics (e.g., coverage percentages, path counts) and comparisons to baselines should be highlighted earlier to support the speedup claims.
  3. The GitHub repository is mentioned but should be formally cited in the main text (e.g., in the contributions or experimental setup) with a stable reference.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their constructive review and recommendation of minor revision. The report accurately summarizes our contributions, correctly identifies the scope to idealized street-canyon scenarios, and notes the disclosed OOD generalization limitations on Manhattan geometries. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an empirical ML framework (GFlowNets with replay buffer, uniform policy, and physics masking) trained on simulation data and validated via independent metrics (speedup, coverage accuracy) on held-out scenarios. No equations or claims reduce a prediction to its own fitted inputs by construction, no self-citation chain bears the central result, and no ansatz or uniqueness theorem is smuggled in. The derivation is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no information on specific free parameters, axioms, or invented entities used in the work.

pith-pipeline@v0.9.0 · 5833 in / 1129 out tokens · 66152 ms · 2026-05-25T06:33:00.246108+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

40 extracted references · 40 canonical work pages

  1. [1]

    Prentice Hall PTR, USA (2001)

    Rappaport, T.: Wireless Communications: Principles and Practice, 2nd edn. Prentice Hall PTR, USA (2001)

    work page 2001

  2. [2]

    IEEE Vehicular Technology Magazine15(4), 22–32 (2020) https://doi.org/10.1109/MVT

    Wang, C.-X., Huang, J., Wang, H., Gao, X., You, X., Hao, Y .: 6g wireless channel measurements and models: Trends and challenges. IEEE Vehicular Technology Magazine15(4), 22–32 (2020) https://doi.org/10.1109/MVT. 2020.3018436

    work page doi:10.1109/mvt 2020

  3. [3]

    IEEE Transactions on Vehicular Technology29(3), 317–325 (1980) https://doi.org/10.1109/T-VT.1980.23859

    Hata, M.: Empirical formula for propagation loss in land mobile radio services. IEEE Transactions on Vehicular Technology29(3), 317–325 (1980) https://doi.org/10.1109/T-VT.1980.23859

    work page doi:10.1109/t-vt.1980.23859 1980

  4. [4]

    IEEE Antennas and Propagation Magazine45(3), 51–82 (2003) https://doi.org/10.1109/MAP

    Sarkar, T.K., Ji, Z., Kim, K., Medouri, A., Salazar-Palma, M.: A survey of various propagation models for mobile communication. IEEE Antennas and Propagation Magazine45(3), 51–82 (2003) https://doi.org/10.1109/MAP. 2003.1232163

    work page doi:10.1109/map 2003

  5. [5]

    IEEE Communications Surveys & Tutorials21(1), 10–27 (2019) https://doi.org/10.1109/COMST.2018.2865724

    He, D., Ai, B., Guan, K., Wang, L., Zhong, Z., K ¨urner, T.: The design and applications of high-performance ray-tracing simulation platform for 5g and beyond wireless communications: A tutorial. IEEE Communications Surveys & Tutorials21(1), 10–27 (2019) https://doi.org/10.1109/COMST.2018.2865724

    work page doi:10.1109/comst.2018.2865724 2019

  6. [6]

    IEEE Access 3, 1089–1100 (2015) https://doi.org/10.1109/ACCESS.2015.2453991

    Yun, Z., Iskander, M.F.: Ray tracing for radio propagation modeling: Principles and applications. IEEE Access 3, 1089–1100 (2015) https://doi.org/10.1109/ACCESS.2015.2453991

    work page doi:10.1109/access.2015.2453991 2015

  7. [8]

    Wireless Personal Communications106(1), 41–70 (2019) https://doi.org/10.1007/s11277-019-06275-4

    Aldossari, S.M., Chen, K.-C.: Machine learning for wireless communication channel modeling: An overview. Wireless Personal Communications106(1), 41–70 (2019) https://doi.org/10.1007/s11277-019-06275-4

    work page doi:10.1007/s11277-019-06275-4 2019

  8. [9]

    IEEE Open Journal of the Communications Society5, 5123–5153 (2024) https://doi.org/10.1109/OJCOMS.2024

    Vasudevan, M., Yuksel, M.: Machine learning for radio propagation modeling: A comprehensive survey. IEEE Open Journal of the Communications Society5, 5123–5153 (2024) https://doi.org/10.1109/OJCOMS.2024. 3446457

    work page doi:10.1109/ojcoms.2024 2024

  9. [10]

    In: Proceedings of the 29th Annual International Conference on Mobile Computing and Networking

    Zhao, X., An, Z., Pan, Q., Yang, L.: NeRF2: Neural radio-frequency radiance fields. In: Proceedings of the 29th Annual International Conference on Mobile Computing and Networking. ACM MobiCom ’23. Association for Computing Machinery, New York, NY , USA (2023). https://doi.org/10.1145/3570361.3592527

    work page doi:10.1145/3570361.3592527 2023

  10. [11]

    arXiv (2024)

    Yang, H., Jin, Z., Wu, C., Xiong, R., Qiu, R.C., Ling, Z.: R-NeRF: Neural Radiance Fields for Modeling RIS- enabled Wireless Environments. arXiv (2024). https://arxiv.org/abs/2405.11541

    work page arXiv 2024

  11. [12]

    In: IEEE INFOCOM 2025 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), pp

    Shen, J., Zhao, T., Wu, Y ., Wang, X.: NeRF-APT: A new NeRF framework for wireless channel prediction. In: IEEE INFOCOM 2025 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), pp. 1–6 (2025). https://doi.org/10.1109/INFOCOMWKSHPS65812.2025.11152993

    work page doi:10.1109/infocomwkshps65812.2025.11152993 2025

  12. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Chen, X., Feng, Z., Qian, K., Zhang, X.: Radio frequency ray tracing with neural object representation for enhanced RF modeling. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 21339–21348 (2025)

    work page 2025

  13. [14]

    IEEE Access8, 199523–199538 (2020) https://doi.org/10.1109/ACCESS

    Wu, L., He, D., Ai, B., Wang, J., Qi, H., Guan, K., Zhong, Z.: Artificial neural network based path loss prediction for wireless communication network. IEEE Access8, 199523–199538 (2020) https://doi.org/10.1109/ACCESS. 2020.3035209

    work page doi:10.1109/access 2020

  14. [15]

    IEEE Access8, 7925–7936 (2020) https://doi.org/10.1109/ACCESS.2020

    Thrane, J., Zibar, D., Christiansen, H.L.: Model-aided deep learning method for path loss prediction in mobile communication systems at 2.6 GHz. IEEE Access8, 7925–7936 (2020) https://doi.org/10.1109/ACCESS.2020. 2964103

    work page doi:10.1109/access.2020 2020

  15. [16]

    In: ICML 2024 Workshop on Geometry-grounded Representation Learning and Generative Modeling (2024).https://openreview.net/forum?id=dlz5lzhpE7

    Hehn, T., Peschl, M., Orekondy, T., Behboodi, A., Brehmer, J.: Geometric wireless simulation with equivari- ant transformers. In: ICML 2024 Workshop on Geometry-grounded Representation Learning and Generative Modeling (2024).https://openreview.net/forum?id=dlz5lzhpE7

    work page 2024

  16. [17]

    In: 2024 International Conference on Comput- ing, Networking and Communications (ICNC), pp

    Zhang, L., Sun, H., Sun, J., Hu, R.Q.: WiSegRT: Dataset for site-specific indoor radio propagation modeling with 24 PREPRINTSUBMITTED TO NPJWIRELESSTECHNOLOGY 3d segmentation and differentiable ray-tracing: (invited paper). In: 2024 International Conference on Comput- ing, Networking and Communications (ICNC), pp. 744–748 (2024). https://doi.org/10.1109/I...

    work page doi:10.1109/icnc59896.2024 2024

  17. [19]

    In: 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), pp

    Jin, Y ., Maatouk, A., Girdzijauskas, S., Xu, S., Tassiulas, L., Ying, R.: SANDWICH: Towards an offline, dif- ferentiable, fully-trainable wireless neural ray-tracing surrogate. In: 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), pp. 1–7 (2025). https://doi.org/10.1109/ ICMLCN64995.2025.11139897

    work page arXiv 2025

  18. [20]

    In: 2023 IEEE Globecom Workshops (GC Wkshps), pp

    Hoydis, J., Aoudia, F.A., Cammerer, S., Nimier-David, M., Binder, N., Marcus, G., Keller, A.: Sionna RT: Dif- ferentiable ray tracing for radio propagation modeling. In: 2023 IEEE Globecom Workshops (GC Wkshps), pp. 317–321 (2023). https://doi.org/10.1109/GCWkshps58843.2023.10465179

    work page doi:10.1109/gcwkshps58843.2023.10465179 2023

  19. [22]

    Software available from tensorflow.org (2015)

    Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G.S., Davis, A., Dean, J., Devin, M., Ghemawat, S., Goodfellow, I., Harp, A., Irving, G., Isard, M., Jia, Y ., Jozefowicz, R., Kaiser, L., Kudlur, M., Levenberg, J., Man ´e, D., Monga, R., Moore, S., Murray, D., Olah, C., Schuster, M., Shlens, J., Steiner, B., Sutskever, I., Tal...

    work page 2015

  20. [23]

    In: Proceedings of the 33rd International Conference on Neural Information Processing Systems (2019)

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., Desmaison, A., K ¨opf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., Chintala, S.: Pytorch: an imperative style, high-performance deep learning library. In: Proceedings of the 33rd I...

    work page 2019

  21. [24]

    http://github.com/jax-ml/jax

    Bradbury, J., Frostig, R., Hawkins, P., Johnson, M.J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., Van- derPlas, J., Wanderman-Milne, S., Zhang, Q.: JAX: composable transformations of Python+NumPy programs (2018). http://github.com/jax-ml/jax

    work page 2018

  22. [25]

    In: Proceedings of the 15th ACM Multimedia Systems Conference

    Testolina, P., Polese, M., Johari, P., Melodia, T.: Boston twin: the boston digital twin for ray-tracing in 6g net- works. In: Proceedings of the 15th ACM Multimedia Systems Conference. MMSys ’24, pp. 441–447. Association for Computing Machinery, New York, NY , USA (2024). https://doi.org/10.1145/3625468.3652190

    work page doi:10.1145/3625468.3652190 2024

  23. [26]

    IEEE Communications Surveys & Tutorials15(1), 255–270 (2013) https://doi.org/10.1109/SURV .2012.022412

    Phillips, C., Sicker, D., Grunwald, D.: A survey of wireless path loss prediction and coverage mapping methods. IEEE Communications Surveys & Tutorials15(1), 255–270 (2013) https://doi.org/10.1109/SURV .2012.022412. 00172

    work page doi:10.1109/surv 2013

  24. [27]

    In: The Eleventh International Conference on Learning Representations (2023).https://openreview.net/forum?id=tPKKXeW33YU

    Orekondy, T., Kumar, P., Kadambi, S., Ye, H., Soriaga, J., Behboodi, A.: WiNeRT: Towards neural ray tracing for wireless channel modelling and differentiable simulations. In: The Eleventh International Conference on Learning Representations (2023).https://openreview.net/forum?id=tPKKXeW33YU

    work page 2023

  25. [28]

    IEEE Journal on Multiscale and Multiphysics Computational Techniques9, 330–340 (2024) https: //doi.org/10.1109/JMMCT.2024.3464373

    Cao, G., Peng, Z.: RayProNet: A neural point field framework for radio propagation modeling in 3d envi- ronments. IEEE Journal on Multiscale and Multiphysics Computational Techniques9, 330–340 (2024) https: //doi.org/10.1109/JMMCT.2024.3464373

    work page doi:10.1109/jmmct.2024.3464373 2024

  26. [29]

    In: 2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall), pp

    Li, Y ., Wang, Y ., Huang, C.: NeRA: Neural reflectance and attenuation fields for radio map reconstruction. In: 2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall), pp. 1–5 (2024). https://doi.org/10.1109/ VTC2024-Fall63153.2024.10757803

    work page arXiv 2024

  27. [30]

    IEEE Access7, 150462–150483 (2019) https://doi.org/10.1109/ACCESS.2019.2947009

    Popoola, S.I., Jefia, A., Atayero, A.A., Kingsley, O., Faruk, N., Oseni, O.F., Abolade, R.O.: Determination of 25 PREPRINTSUBMITTED TO NPJWIRELESSTECHNOLOGY neural network parameters for path loss prediction in very high frequency wireless channel. IEEE Access7, 150462–150483 (2019) https://doi.org/10.1109/ACCESS.2019.2947009

    work page doi:10.1109/access.2019.2947009 2019

  28. [31]

    https://arxiv.org/abs/2104.13562

    Knodt, J., Bartusek, J., Baek, S.-H., Heide, F.: Neural Ray-Tracing: Learning Surfaces and Reflectance for Relighting and View Synthesis (2021). https://arxiv.org/abs/2104.13562

    work page arXiv 2021

  29. [32]

    IEEE Transactions on Wireless Communications 24(7), 5811–5824 (2025) https://doi.org/10.1109/TWC.2025.3549498

    Wang, S., Gao, S., Yang, W., Zhang, Q., Loh, T.-H., Yang, Y ., Qin, F.: A physics-informed deep ray tracing network for regional channel impulse response estimation. IEEE Transactions on Wireless Communications 24(7), 5811–5824 (2025) https://doi.org/10.1109/TWC.2025.3549498

    work page doi:10.1109/twc.2025.3549498 2025

  30. [33]

    In: 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), pp

    Wiesmayr, R., Cammerer, S., Aoudia, F.A., Hoydis, J., Zakrzewski, J., Keller, A.: Design of a standard- compliant real-time neural receiver for 5G NR. In: 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), pp. 1–6 (2025). https://doi.org/10.1109/ICMLCN64995.2025. 11140048

    work page doi:10.1109/icmlcn64995.2025 2025

  31. [34]

    IEEE Transactions on Antennas and Propagation67(2), 1180–1192 (2019) https://doi.org/10.1109/TAP.2018.2880036

    Lu, J.S., Vitucci, E.M., Degli-Esposti, V ., Fuschini, F., Barbiroli, M., Blaha, J.A., Bertoni, H.L.: A discrete environment-driven GPU-based ray launching algorithm. IEEE Transactions on Antennas and Propagation67(2), 1180–1192 (2019) https://doi.org/10.1109/TAP.2018.2880036

    work page doi:10.1109/tap.2018.2880036 2019

  32. [35]

    IEEE Transactions on Antennas and Propagation70(10), 9977–9982 (2022) https://doi.org/10.1109/TAP

    Hussain, S., Brennan, C.: A visibility matching technique for efficient millimeter-wave vehicular channel mod- eling. IEEE Transactions on Antennas and Propagation70(10), 9977–9982 (2022) https://doi.org/10.1109/TAP. 2022.3178130

    work page doi:10.1109/tap 2022

  33. [36]

    Journal of Machine Learning Research24(210), 1–55 (2023)

    Bengio, Y ., Lahlou, S., Deleu, T., Hu, E.J., Tiwari, M., Bengio, E.: GFlowNet foundations. Journal of Machine Learning Research24(210), 1–55 (2023)

    work page 2023

  34. [37]

    In: Advances in Neural Information Processing Systems, vol

    Zaheer, M., Kottur, S., Ravanbakhsh, S., Poczos, B., Salakhutdinov, R.R., Smola, A.J.: Deep sets. In: Advances in Neural Information Processing Systems, vol. 30. Curran Associates, Inc., Red Hook, NY , USA (2017)

    work page 2017

  35. [38]

    The Journal of the Acoustical Society of America 75(6), 1827–1836 (1984) https://doi.org/10.1121/1.390983

    Borish, J.: Extension of the image model to arbitrary polyhedra. The Journal of the Acoustical Society of America 75(6), 1827–1836 (1984) https://doi.org/10.1121/1.390983

    work page doi:10.1121/1.390983 1984

  36. [39]

    IEEE Transactions on Antennas and Propagation62(8), 4336–4341 (2014) https://doi.org/10.1109/TAP.2014.2323961

    Puggelli, F., Carluccio, G., Albani, M.: A novel ray tracing algorithm for scenarios comprising pre-ordered multiple planar reflectors, straight wedges, and vertexes. IEEE Transactions on Antennas and Propagation62(8), 4336–4341 (2014) https://doi.org/10.1109/TAP.2014.2323961

    work page doi:10.1109/tap.2014.2323961 2014

  37. [40]

    In: 2023 17th European Conference on Antennas and Propagation (EuCAP), pp

    Eertmans, J., Oestges, C., Jacques, L.: Min-Path-Tracing: A diffraction aware alternative to image method in ray tracing. In: 2023 17th European Conference on Antennas and Propagation (EuCAP), pp. 1–5 (2023). https: //doi.org/10.23919/EuCAP57121.2023.10132934

    work page doi:10.23919/eucap57121.2023.10132934 2023

  38. [41]

    Differentiable Programming workshop at Neural Information Processing Systems 2021 (2021)

    Kidger, P., Garcia, C.: Equinox: neural networks in JAX via callable PyTrees and filtered transformations. Differentiable Programming workshop at Neural Information Processing Systems 2021 (2021)

    work page 2021

  39. [42]

    http://github.com/google-deepmind

    DeepMind, Babuschkin, I., Baumli, K., Bell, A., Bhupatiraju, S., Bruce, J., Buchlovsky, P., Budden, D., Cai, T., Clark, A., Danihelka, I., Dedieu, A., Fantacci, C., Godwin, J., Jones, C., Hemsley, R., Hennigan, T., Hessel, M., Hou, S., Kapturowski, S., Keck, T., Kemaev, I., King, M., Kunesch, M., Martens, L., Merzic, H., Mikulik, V ., Norman, T., Papamaka...

    work page 2020

  40. [43]

    https://kellerjordan.github.io/posts/muon/ 26

    Jordan, K., Jin, Y ., Boza, V ., Jiacheng, Y ., Cesista, F., Newhouse, L., Bernstein, J.: Muon: An optimizer for hidden layers in neural networks (2024). https://kellerjordan.github.io/posts/muon/ 26

    work page 2024