Bayesian RG flow in neural network field theories
The Neural Network Field Theory correspondence (NNFT) is a mapping from neural network (NN) architectures into the space of statistical field theories (SFTs). The Bayesian renormalization group (BRG) is an information-theoretic coarse graining scheme that generalizes the principles of the exact reno...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SciPost
2025-03-01
|
| Series: | SciPost Physics Core |
| Online Access: | https://scipost.org/SciPostPhysCore.8.1.027 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849706929747656704 |
|---|---|
| author | Jessica N. Howard, Marc S. Klinger, Anindita Maiti, Alexander G. Stapleton |
| author_facet | Jessica N. Howard, Marc S. Klinger, Anindita Maiti, Alexander G. Stapleton |
| author_sort | Jessica N. Howard, Marc S. Klinger, Anindita Maiti, Alexander G. Stapleton |
| collection | DOAJ |
| description | The Neural Network Field Theory correspondence (NNFT) is a mapping from neural network (NN) architectures into the space of statistical field theories (SFTs). The Bayesian renormalization group (BRG) is an information-theoretic coarse graining scheme that generalizes the principles of the exact renormalization group (ERG) to arbitrarily parameterized probability distributions, including those of NNs. In BRG, coarse graining is performed in parameter space with respect to an information-theoretic distinguishability scale set by the Fisher information metric. In this paper, we unify NNFT and BRG to form a powerful new framework for exploring the space of NNs and SFTs, which we coin BRG-NNFT. With BRG-NNFT, NN training dynamics can be interpreted as inducing a flow in the space of SFTs from the information-theoretic 'IR' $→$ 'UV'. Conversely, applying an information-shell coarse graining to the trained network's parameters induces a flow in the space of SFTs from the information-theoretic 'UV' $→$ 'IR'. When the information-theoretic cutoff scale coincides with a standard momentum scale, BRG is equivalent to ERG. We demonstrate the BRG-NNFT correspondence on two analytically tractable examples. First, we construct BRG flows for trained, infinite-width NNs, of arbitrary depth, with generic activation functions. As a special case, we then restrict to architectures with a single infinitely-wide layer, scalar outputs, and generalized cos-net activations. In this case, we show that BRG coarse-graining corresponds exactly to the momentum-shell ERG flow of a free scalar SFT. Our analytic results are corroborated by a numerical experiment in which an ensemble of asymptotically wide NNs are trained and subsequently renormalized using an information-shell BRG scheme. |
| format | Article |
| id | doaj-art-d542071ba9f643d5b64a9f488e5c09d3 |
| institution | DOAJ |
| issn | 2666-9366 |
| language | English |
| publishDate | 2025-03-01 |
| publisher | SciPost |
| record_format | Article |
| series | SciPost Physics Core |
| spelling | doaj-art-d542071ba9f643d5b64a9f488e5c09d32025-08-20T03:16:04ZengSciPostSciPost Physics Core2666-93662025-03-018102710.21468/SciPostPhysCore.8.1.027Bayesian RG flow in neural network field theoriesJessica N. Howard, Marc S. Klinger, Anindita Maiti, Alexander G. StapletonThe Neural Network Field Theory correspondence (NNFT) is a mapping from neural network (NN) architectures into the space of statistical field theories (SFTs). The Bayesian renormalization group (BRG) is an information-theoretic coarse graining scheme that generalizes the principles of the exact renormalization group (ERG) to arbitrarily parameterized probability distributions, including those of NNs. In BRG, coarse graining is performed in parameter space with respect to an information-theoretic distinguishability scale set by the Fisher information metric. In this paper, we unify NNFT and BRG to form a powerful new framework for exploring the space of NNs and SFTs, which we coin BRG-NNFT. With BRG-NNFT, NN training dynamics can be interpreted as inducing a flow in the space of SFTs from the information-theoretic 'IR' $→$ 'UV'. Conversely, applying an information-shell coarse graining to the trained network's parameters induces a flow in the space of SFTs from the information-theoretic 'UV' $→$ 'IR'. When the information-theoretic cutoff scale coincides with a standard momentum scale, BRG is equivalent to ERG. We demonstrate the BRG-NNFT correspondence on two analytically tractable examples. First, we construct BRG flows for trained, infinite-width NNs, of arbitrary depth, with generic activation functions. As a special case, we then restrict to architectures with a single infinitely-wide layer, scalar outputs, and generalized cos-net activations. In this case, we show that BRG coarse-graining corresponds exactly to the momentum-shell ERG flow of a free scalar SFT. Our analytic results are corroborated by a numerical experiment in which an ensemble of asymptotically wide NNs are trained and subsequently renormalized using an information-shell BRG scheme.https://scipost.org/SciPostPhysCore.8.1.027 |
| spellingShingle | Jessica N. Howard, Marc S. Klinger, Anindita Maiti, Alexander G. Stapleton Bayesian RG flow in neural network field theories SciPost Physics Core |
| title | Bayesian RG flow in neural network field theories |
| title_full | Bayesian RG flow in neural network field theories |
| title_fullStr | Bayesian RG flow in neural network field theories |
| title_full_unstemmed | Bayesian RG flow in neural network field theories |
| title_short | Bayesian RG flow in neural network field theories |
| title_sort | bayesian rg flow in neural network field theories |
| url | https://scipost.org/SciPostPhysCore.8.1.027 |
| work_keys_str_mv | AT jessicanhowardmarcsklingeraninditamaitialexandergstapleton bayesianrgflowinneuralnetworkfieldtheories |