Hierarchical abstraction of information in Deep Neural Networks

Hierarchical abstraction of information in Deep Neural Networks. Dehmamy, N., Rohani, N., & Katsaggelos, A. In APS March Meeting Abstracts, volume 2017, pages T1—-371, 2017.

Paper abstract bibtex

We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.

@inproceedings{Nima2017a,
abstract = {We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.},
author = {Dehmamy, Nima and Rohani, Neda and Katsaggelos, Aggelos},
booktitle = {APS March Meeting Abstracts},
pages = {T1----371},
title = {{Hierarchical abstraction of information in Deep Neural Networks}},
url = {https://meetings.aps.org/Meeting/MAR17/Event/299460},
volume = {2017},
year = {2017}
}

Downloads: 0

{"_id":"do3whd9vsKBARdL4G","bibbaseid":"dehmamy-rohani-katsaggelos-hierarchicalabstractionofinformationindeepneuralnetworks-2017","author_short":["Dehmamy, N.","Rohani, N.","Katsaggelos, A."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.","author":[{"propositions":[],"lastnames":["Dehmamy"],"firstnames":["Nima"],"suffixes":[]},{"propositions":[],"lastnames":["Rohani"],"firstnames":["Neda"],"suffixes":[]},{"propositions":[],"lastnames":["Katsaggelos"],"firstnames":["Aggelos"],"suffixes":[]}],"booktitle":"APS March Meeting Abstracts","pages":"T1—-371","title":"Hierarchical abstraction of information in Deep Neural Networks","url":"https://meetings.aps.org/Meeting/MAR17/Event/299460","volume":"2017","year":"2017","bibtex":"@inproceedings{Nima2017a,\nabstract = {We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.},\nauthor = {Dehmamy, Nima and Rohani, Neda and Katsaggelos, Aggelos},\nbooktitle = {APS March Meeting Abstracts},\npages = {T1----371},\ntitle = {{Hierarchical abstraction of information in Deep Neural Networks}},\nurl = {https://meetings.aps.org/Meeting/MAR17/Event/299460},\nvolume = {2017},\nyear = {2017}\n}\n","author_short":["Dehmamy, N.","Rohani, N.","Katsaggelos, A."],"key":"Nima2017a","id":"Nima2017a","bibbaseid":"dehmamy-rohani-katsaggelos-hierarchicalabstractionofinformationindeepneuralnetworks-2017","role":"author","urls":{"Paper":"https://meetings.aps.org/Meeting/MAR17/Event/299460"},"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://sites.northwestern.edu/ivpl/files/2023/06/IVPL_Updated_publications-1.bib","dataSources":["ePKPjG8C6yvpk4mEK","ya2CyA73rpZseyrZ8","D8k2SxfC5dKNRFgro","7Dwzbxq93HWrJEhT6","qhF8zxmGcJfvtdeAg","fvDEHD49E2ZRwE3fb","H7crv8NWhZup4d4by","DHqokWsryttGh7pJE","vRJd4wNg9HpoZSMHD","sYxQ6pxFgA59JRhxi","w2WahSbYrbcCKBDsC","XasdXLL99y5rygCmq","3gkSihZQRfAD2KBo3","t5XMbyZbtPBo4wBGS","bEpHM2CtrwW2qE8FP","teJzFLHexaz5AQW5z"],"keywords":[],"search_terms":["hierarchical","abstraction","information","deep","neural","networks","dehmamy","rohani","katsaggelos"],"title":"Hierarchical abstraction of information in Deep Neural Networks","year":2017}