Cosmo3DFlow: Wavelet Flow Matching for Spatial-to-Spectral Compression in Reconstructing the Early Universe. Islam, M. K., Xia, Z., Goudjil, R., Wang, J., Farahi, A., & Fox, J. In Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V2, Jeju Island, Korea, August, 2026. Association for Computing Machinery(ACM). arXiv:2602.10172 [astro-ph]![Cosmo3DFlow: Wavelet Flow Matching for Spatial-to-Spectral Compression in Reconstructing the Early Universe [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex Reconstructing the early Universe from the evolved present-day Universe is a challenging and computationally demanding problem in modern astrophysics. We devise a novel generative framework, Cosmo3DFlow, designed to address dimensionality and sparsity, the critical bottlenecks inherent in current state-of-the-art methods for cosmological inference. By integrating 3D Discrete Wavelet Transform (DWT) with flow matching, we effectively represent high-dimensional cosmological structures. The Wavelet Transform addresses the “void problem” by translating spatial emptiness into spectral sparsity. It decouples high-frequency details from lowfrequency structures through spatial compression, and waveletspace velocity fields facilitate stable ordinary differential equation (ODE) solvers with large step sizes. Using large-scale cosmological 𝑁 -body simulations, at 1283 resolution, we achieve up to 50× faster sampling than diffusion models, combining a 10× reduction in integration steps with lower per-step computational cost from wavelet compression. Our results enable initial conditions to be sampled in seconds, compared to minutes for previous methods.
@inproceedings{islam_cosmo3dflow_2026,
address = {Jeju Island, Korea},
title = {{Cosmo3DFlow}: {Wavelet} {Flow} {Matching} for {Spatial}-to-{Spectral} {Compression} in {Reconstructing} the {Early} {Universe}},
shorttitle = {{Cosmo3DFlow}},
url = {http://arxiv.org/abs/2602.10172},
doi = {doi.org/10.1145/3770855.3818994},
abstract = {Reconstructing the early Universe from the evolved present-day Universe is a challenging and computationally demanding problem in modern astrophysics. We devise a novel generative framework, Cosmo3DFlow, designed to address dimensionality and sparsity, the critical bottlenecks inherent in current state-of-the-art methods for cosmological inference. By integrating 3D Discrete Wavelet Transform (DWT) with flow matching, we effectively represent high-dimensional cosmological structures. The Wavelet Transform addresses the “void problem” by translating spatial emptiness into spectral sparsity. It decouples high-frequency details from lowfrequency structures through spatial compression, and waveletspace velocity fields facilitate stable ordinary differential equation (ODE) solvers with large step sizes. Using large-scale cosmological 𝑁 -body simulations, at 1283 resolution, we achieve up to 50× faster sampling than diffusion models, combining a 10× reduction in integration steps with lower per-step computational cost from wavelet compression. Our results enable initial conditions to be sampled in seconds, compared to minutes for previous methods.},
language = {en},
urldate = {2026-02-17},
booktitle = {Proceedings of the 32nd {ACM} {SIGKDD} {Conference} on {Knowledge} {Discovery} and {Data} {Mining} {V2}},
publisher = {Association for Computing Machinery(ACM)},
author = {Islam, Md Khairul and Xia, Zeyu and Goudjil, Ryan and Wang, Jialu and Farahi, Arya and Fox, Judy},
month = aug,
year = {2026},
note = {arXiv:2602.10172 [astro-ph]},
keywords = {Astrophysics - Instrumentation and Methods for Astrophysics, Computer Science - Artificial Intelligence, SYS: CosmicAI Contact Author, WG: Explainable},
}
Downloads: 0
{"_id":"CMSctPHZDvm7A9Pym","bibbaseid":"islam-xia-goudjil-wang-farahi-fox-cosmo3dflowwaveletflowmatchingforspatialtospectralcompressioninreconstructingtheearlyuniverse-2026","author_short":["Islam, M. K.","Xia, Z.","Goudjil, R.","Wang, J.","Farahi, A.","Fox, J."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","address":"Jeju Island, Korea","title":"Cosmo3DFlow: Wavelet Flow Matching for Spatial-to-Spectral Compression in Reconstructing the Early Universe","shorttitle":"Cosmo3DFlow","url":"http://arxiv.org/abs/2602.10172","doi":"doi.org/10.1145/3770855.3818994","abstract":"Reconstructing the early Universe from the evolved present-day Universe is a challenging and computationally demanding problem in modern astrophysics. We devise a novel generative framework, Cosmo3DFlow, designed to address dimensionality and sparsity, the critical bottlenecks inherent in current state-of-the-art methods for cosmological inference. By integrating 3D Discrete Wavelet Transform (DWT) with flow matching, we effectively represent high-dimensional cosmological structures. The Wavelet Transform addresses the “void problem” by translating spatial emptiness into spectral sparsity. It decouples high-frequency details from lowfrequency structures through spatial compression, and waveletspace velocity fields facilitate stable ordinary differential equation (ODE) solvers with large step sizes. Using large-scale cosmological 𝑁 -body simulations, at 1283 resolution, we achieve up to 50× faster sampling than diffusion models, combining a 10× reduction in integration steps with lower per-step computational cost from wavelet compression. Our results enable initial conditions to be sampled in seconds, compared to minutes for previous methods.","language":"en","urldate":"2026-02-17","booktitle":"Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V2","publisher":"Association for Computing Machinery(ACM)","author":[{"propositions":[],"lastnames":["Islam"],"firstnames":["Md","Khairul"],"suffixes":[]},{"propositions":[],"lastnames":["Xia"],"firstnames":["Zeyu"],"suffixes":[]},{"propositions":[],"lastnames":["Goudjil"],"firstnames":["Ryan"],"suffixes":[]},{"propositions":[],"lastnames":["Wang"],"firstnames":["Jialu"],"suffixes":[]},{"propositions":[],"lastnames":["Farahi"],"firstnames":["Arya"],"suffixes":[]},{"propositions":[],"lastnames":["Fox"],"firstnames":["Judy"],"suffixes":[]}],"month":"August","year":"2026","note":"arXiv:2602.10172 [astro-ph]","keywords":"Astrophysics - Instrumentation and Methods for Astrophysics, Computer Science - Artificial Intelligence, SYS: CosmicAI Contact Author, WG: Explainable","bibtex":"@inproceedings{islam_cosmo3dflow_2026,\n\taddress = {Jeju Island, Korea},\n\ttitle = {{Cosmo3DFlow}: {Wavelet} {Flow} {Matching} for {Spatial}-to-{Spectral} {Compression} in {Reconstructing} the {Early} {Universe}},\n\tshorttitle = {{Cosmo3DFlow}},\n\turl = {http://arxiv.org/abs/2602.10172},\n\tdoi = {doi.org/10.1145/3770855.3818994},\n\tabstract = {Reconstructing the early Universe from the evolved present-day Universe is a challenging and computationally demanding problem in modern astrophysics. We devise a novel generative framework, Cosmo3DFlow, designed to address dimensionality and sparsity, the critical bottlenecks inherent in current state-of-the-art methods for cosmological inference. By integrating 3D Discrete Wavelet Transform (DWT) with flow matching, we effectively represent high-dimensional cosmological structures. The Wavelet Transform addresses the “void problem” by translating spatial emptiness into spectral sparsity. It decouples high-frequency details from lowfrequency structures through spatial compression, and waveletspace velocity fields facilitate stable ordinary differential equation (ODE) solvers with large step sizes. Using large-scale cosmological 𝑁 -body simulations, at 1283 resolution, we achieve up to 50× faster sampling than diffusion models, combining a 10× reduction in integration steps with lower per-step computational cost from wavelet compression. Our results enable initial conditions to be sampled in seconds, compared to minutes for previous methods.},\n\tlanguage = {en},\n\turldate = {2026-02-17},\n\tbooktitle = {Proceedings of the 32nd {ACM} {SIGKDD} {Conference} on {Knowledge} {Discovery} and {Data} {Mining} {V2}},\n\tpublisher = {Association for Computing Machinery(ACM)},\n\tauthor = {Islam, Md Khairul and Xia, Zeyu and Goudjil, Ryan and Wang, Jialu and Farahi, Arya and Fox, Judy},\n\tmonth = aug,\n\tyear = {2026},\n\tnote = {arXiv:2602.10172 [astro-ph]},\n\tkeywords = {Astrophysics - Instrumentation and Methods for Astrophysics, Computer Science - Artificial Intelligence, SYS: CosmicAI Contact Author, WG: Explainable},\n}\n\n\n\n","author_short":["Islam, M. K.","Xia, Z.","Goudjil, R.","Wang, J.","Farahi, A.","Fox, J."],"key":"islam_cosmo3dflow_2026","id":"islam_cosmo3dflow_2026","bibbaseid":"islam-xia-goudjil-wang-farahi-fox-cosmo3dflowwaveletflowmatchingforspatialtospectralcompressioninreconstructingtheearlyuniverse-2026","role":"author","urls":{"Paper":"http://arxiv.org/abs/2602.10172"},"keyword":["Astrophysics - Instrumentation and Methods for Astrophysics","Computer Science - Artificial Intelligence","SYS: CosmicAI Contact Author","WG: Explainable"],"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://bibbase.org/zotero-group/pratikmhatre/5933976","dataSources":["yJr5AAtJ5Sz3Q4WT4"],"keywords":["astrophysics - instrumentation and methods for astrophysics","computer science - artificial intelligence","sys: cosmicai contact author","wg: explainable"],"search_terms":["cosmo3dflow","wavelet","flow","matching","spatial","spectral","compression","reconstructing","early","universe","islam","xia","goudjil","wang","farahi","fox"],"title":"Cosmo3DFlow: Wavelet Flow Matching for Spatial-to-Spectral Compression in Reconstructing the Early Universe","year":2026}