Abstract Understanding how deep learning architectures work is a central scientific problem. Recently, a correspondence between neural networks (NNs) and Euclidean quantum field theories has been proposed. This work investigates this correspondence in the framework of p-adic statistical field theories (SFTs) and neural networks. In this case, the fields are real-valued functions defined on an infinite regular rooted tree with valence p, a fixed prime number. This infinite tree provides the topology for a continuous deep Boltzmann machine (DBM), which is identified with a statistical field theory on this infinite tree. In the p-adic framework, there is a natural method to discretize SFTs. Each discrete SFT corresponds to a Boltzmann machine with a tree-like topology. This method allows us to recover the standard DBMs and gives new convolutional DBMs. The new networks use O(N) parameters while the classical ones use O(N$^{2}$) parameters.
{ "_oai": { "updated": "2023-06-15T15:30:24Z", "id": "oai:repo.scoap3.org:78326", "sets": [ "PTEP" ] }, "authors": [ { "affiliations": [ { "country": "USA", "value": "University of Texas Rio Grande Valley, School of Mathematical and Statistical Sciences, , , One West University Blvd., Brownsville, TX 78520, , , USA" } ], "surname": "Z\u00fa\u00f1iga-Galindo", "email": "wilson.zunigagalindo@utrgv.edu", "full_name": "Z\u00fa\u00f1iga-Galindo, W A", "given_names": "W A" }, { "affiliations": [ { "country": "USA", "value": "Oklahoma State University, Department of Mathematics, , , MSCS 425, Stillwater, OK, , , USA" } ], "surname": "He", "given_names": "C", "full_name": "He, C" }, { "affiliations": [ { "country": "USA", "value": "University of Texas Rio Grande Valley, School of Mathematical and Statistical Sciences, , , One West University Blvd., Brownsville, TX 78520, , , USA" } ], "surname": "Zambrano-Luna", "given_names": "B A", "full_name": "Zambrano-Luna, B A" } ], "titles": [ { "source": "OUP", "title": "-Adic statistical field theory and convolutional deep Boltzmann machines An introduction to restricted Boltzmann machines" } ], "dois": [ { "value": "10.1093/ptep/ptad061" } ], "publication_info": [ { "journal_title": "Progress of Theoretical and Experimental Physics", "material": "article", "journal_volume": "2023", "artid": "063A01", "year": 2023, "journal_issue": "6" } ], "$schema": "http://repo.scoap3.org/schemas/hep.json", "acquisition_source": { "date": "2023-06-15T15:30:22.751982", "source": "OUP", "method": "OUP", "submission_number": "7d98f3dc0b9111ee8e62c23b20d38f61" }, "page_nr": [ 17 ], "license": [ { "url": "https://creativecommons.org/licenses/by/4.0/", "license": "CC-BY-4.0" } ], "copyright": [ { "statement": "\u00a9 The Author(s) 2023. Published by Oxford University Press on behalf of the Physical Society of Japan.", "year": "2023" } ], "control_number": "78326", "record_creation_date": "2023-06-15T15:30:22.752001", "_files": [ { "checksum": "md5:b9d21bc3c3a9b13e9550d39708dd8362", "filetype": "xml", "bucket": "c7ad9561-c448-4a7b-893d-8105179d5045", "version_id": "8d917da8-70f6-4885-9edf-ac1ca9e33800", "key": "10.1093/ptep/ptad061.xml", "size": 117894 }, { "checksum": "md5:ec9d0ac8f1db4449c81cf93c37215bfc", "filetype": "pdf", "bucket": "c7ad9561-c448-4a7b-893d-8105179d5045", "version_id": "9ec23467-c318-487e-af00-2e517bc92b27", "key": "10.1093/ptep/ptad061.pdf", "size": 3425452 }, { "checksum": "md5:3a476bd9a657bae45c0aced05e6c60dc", "filetype": "pdf/a", "bucket": "c7ad9561-c448-4a7b-893d-8105179d5045", "version_id": "ce6c288f-359d-4543-a15d-b3269d1551a7", "key": "10.1093/ptep/ptad061_a.pdf", "size": 3379039 } ], "collections": [ { "primary": "Progress of Theoretical and Experimental Physics" } ], "arxiv_eprints": [ { "categories": [ "hep-th", "cond-mat.dis-nn" ], "value": "2302.03817" } ], "abstracts": [ { "source": "OUP", "value": "Abstract Understanding how deep learning architectures work is a central scientific problem. Recently, a correspondence between neural networks (NNs) and Euclidean quantum field theories has been proposed. This work investigates this correspondence in the framework of p-adic statistical field theories (SFTs) and neural networks. In this case, the fields are real-valued functions defined on an infinite regular rooted tree with valence p, a fixed prime number. This infinite tree provides the topology for a continuous deep Boltzmann machine (DBM), which is identified with a statistical field theory on this infinite tree. In the p-adic framework, there is a natural method to discretize SFTs. Each discrete SFT corresponds to a Boltzmann machine with a tree-like topology. This method allows us to recover the standard DBMs and gives new convolutional DBMs. The new networks use O(N) parameters while the classical ones use O(N$^{2}$) parameters." } ], "imprints": [ { "date": "2023-05-10", "publisher": "OUP" } ] }