abstracts[] |
{'sha1': '630d849e232dfcf13974d4f59dc022ba4f31675a', 'content': 'Inspired by the adaptation phenomenon of neuronal firing, we propose an\nunsupervised attention mechanism (UAM) which computes the statistical\nregularity in the implicit space of neural networks under the Minimum\nDescription Length (MDL) principle. Treating the neural network optimization\nprocess as a partially observable model selection problem, UAM constrained the\nimplicit space by a normalization factor, the universal code length. We compute\nthis universal code incrementally across neural network layers and demonstrated\nthe flexibility to include data priors such as top-down attention and other\noracle information. Empirically, our approach outperforms existing\nnormalization methods in tackling limited, imbalanced and nonstationary input\ndistribution in computer vision and reinforcement learning tasks. Lastly, UAM\ntracks dependency and critical learning stages across layers and recurrent time\nsteps of deep networks.', 'mimetype': 'text/plain', 'lang': 'en'}
|
container |
|
container_id |
|
contribs[] |
{'index': 0, 'creator_id': None, 'creator': None, 'raw_name': 'Baihan Lin', 'given_name': None, 'surname': None, 'role': 'author', 'raw_affiliation': None, 'extra': None}
|
ext_ids |
{'doi': None, 'wikidata_qid': None, 'isbn13': None, 'pmid': None, 'pmcid': None, 'core': None, 'arxiv': '1902.10658v7', 'jstor': None, 'ark': None, 'mag': None, 'doaj': None, 'dblp': None, 'oai': None, 'hdl': None}
|
files[] |
{'state': 'active', 'ident': 'svifzvf67nb3zebxjkakd6zbl4', 'revision': 'f180741a-3ed5-45d0-9573-1acbdfeff4e1', 'redirect': None, 'extra': None, 'edit_extra': None, 'size': 653954, 'md5': '9b738954e382fef47c2cbfb5531ab5f8', 'sha1': '41260d5435bcd592f16da0bf5f594f6353e78823', 'sha256': 'a7bbe856e3a69b10e989af1871b280af758518566d5b327c4e788224db76a8d0', 'urls': [{'url': 'https://arxiv.org/pdf/1902.10658v7.pdf', 'rel': 'repository'}, {'url': 'https://web.archive.org/web/20200827152947/https://arxiv.org/pdf/1902.10658v7.pdf', 'rel': 'webarchive'}], 'mimetype': 'application/pdf', 'content_scope': None, 'release_ids': ['ydic3m3wczedlcqktcvcc7fw3m'], 'releases': None}
|
filesets |
[]
|
issue |
|
language |
en
|
license_slug |
ARXIV-1.0
|
number |
|
original_title |
|
pages |
|
publisher |
|
refs |
[]
|
release_date |
2019-07-31
|
release_stage |
submitted
|
release_type |
article
|
release_year |
2019
|
subtitle |
|
title |
Unsupervised Attention Mechanism across Neural Network Layers
|
version |
v7
|
volume |
|
webcaptures |
[]
|
withdrawn_date |
|
withdrawn_status |
|
withdrawn_year |
|
work_id |
2ulnzgzuyzelrhbls23xmvmyt4
|