SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.18512 · 2023

A Rainbow in Deep Network Black Boxes

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 14 functions out of this paper's own repositories and ran 2 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
florentinguth/rainbow canonical 2 of 14
FunctionStatusWhere it lives
conv1x1 Ran florentinguth/rainbow/models/resnet.py
code served (permissive licence) · get_code("d9def42110729a85")
conv3x3 Ran florentinguth/rainbow/models/resnet.py
code served (permissive licence) · get_code("fac5364e2f53c6db")
DCT_I Not yet run florentinguth/rainbow/models/DCT.py
code served (permissive licence) · get_code("8dafc755add82efd")
DCT_II Not yet run florentinguth/rainbow/models/DCT.py
code served (permissive licence) · get_code("240c75536744ea04")
DCT_III Not yet run florentinguth/rainbow/models/DCT.py
code served (permissive licence) · get_code("1dbdb66cd1152c11")
conv2d Not yet run florentinguth/rainbow/models/LinearProj.py
code served (permissive licence) · get_code("214f33cf948da458")
gaussian_window Not yet run florentinguth/rainbow/models/STFT.py
code served (permissive licence) · get_code("de6dcf943e71f6fb")
get_datasets Not yet run florentinguth/rainbow/datasets.py
code served (permissive licence) · get_code("e6fb903521a8cee4")
hanning_window Not yet run florentinguth/rainbow/models/STFT.py
code served (permissive licence) · get_code("7331f2a3d29fc235")
modulus Not yet run florentinguth/rainbow/models/Analysis.py
code served (permissive licence) · get_code("92d96ed40e7d0a99")
new_logfile Not yet run florentinguth/rainbow/main_block.py
code served (permissive licence) · get_code("31e6ebc3a900f3de")
relu Not yet run florentinguth/rainbow/models/Analysis.py
code served (permissive licence) · get_code("1a2d7d179b6fcef3")
resnet Not yet run florentinguth/rainbow/models/resnet.py
code served (permissive licence) · get_code("ec4c52396ad68e3e")
softshrink Not yet run florentinguth/rainbow/models/Analysis.py
code served (permissive licence) · get_code("f2a6aad1f81a2872")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

A central question in deep learning is to understand the functions learned by deep networks. What is their approximation class? Do the learned weights and representations depend on initialization? Previous empirical work has evidenced that kernels defined by network activations are similar across initializations. For shallow networks, this has been theoretically studied with random feature models, but an extension to deep networks has remained elusive. Here, we provide a deep extension of such random feature models, which we call the rainbow model. We prove that rainbow networks define deterministic (hierarchical) kernels in the infinite-width limit. The resulting functions thus belong to a data-dependent RKHS which does not depend on the weight randomness. We also verify numerically our modeling assumptions on deep CNNs trained on image classification tasks, and show that the trained networks approximately satisfy the rainbow hypothesis. In particular, rainbow networks sampled from the corresponding random feature model achieve similar performance as the trained networks. Our results highlight the central role played by the covariances of network weights at each layer, which are observed to be low-rank as a result of feature learning.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2305.18512")
get_code_for_paper("2305.18512")
have("2305.18512")

Connect an agent — have() is free.