@article{Balduzzi_The_2017,
  year={2017},
  abstract={A long-standing obstacle to progress in deep learning is the problem of vanishing and exploding gradients. The problem has largely been overcome through the introduction of carefully constructed initializations and batch normalization. Nevertheless, architectures incorporating skip-connections such as resnets perform much better than standard feedforward architectures despite well-chosen initialization and batch normalization. In this paper, we identify the shattered gradients problem. Specifically, we show that the correlation between gradients in standard feedforward networks decays exponentially with depth resulting in gradients that resemble white noise. In contrast, the gradients in architectures with skip-connections are far more resistant to shattering decaying sublinearly. Detailed empirical evidence is presented in support of the analysis, on both fully-connected networks and convnets. Finally, we present a new "looks linear" {(LL)} initialization that prevents shattering. Preliminary experiments show the new initialization allows to train very deep networks without the addition of skip-connections.},
  title={The Shattered Gradients Problem: If resnets are the answer, then what is the question?},
  author={Balduzzi, David and Frean, Marcus and Leary, Lennox and Lewis, {JP} and Ma, Kurt {Wan-Duo} and {McWilliams,} Brian}
}

{"_id":"kuM7XLd9gGcKabjRf","bibbaseid":"balduzzi-frean-leary-lewis-ma-brian-theshatteredgradientsproblemifresnetsaretheanswerthenwhatisthequestion-2017","downloads":0,"creationDate":"2017-05-10T16:03:41.186Z","title":"The Shattered Gradients Problem: If resnets are the answer, then what is the question?","author_short":["Balduzzi, D.","Frean, M.","Leary, L.","Lewis, J.","Ma, K. W.","Brian, M."],"year":2017,"bibtype":"article","biburl":"http://tilke.github.io/bib/deep-learning.bib","bibdata":{"bibtype":"article","type":"article","year":"2017","abstract":"A long-standing obstacle to progress in deep learning is the problem of vanishing and exploding gradients. The problem has largely been overcome through the introduction of carefully constructed initializations and batch normalization. Nevertheless, architectures incorporating skip-connections such as resnets perform much better than standard feedforward architectures despite well-chosen initialization and batch normalization. In this paper, we identify the shattered gradients problem. Specifically, we show that the correlation between gradients in standard feedforward networks decays exponentially with depth resulting in gradients that resemble white noise. In contrast, the gradients in architectures with skip-connections are far more resistant to shattering decaying sublinearly. Detailed empirical evidence is presented in support of the analysis, on both fully-connected networks and convnets. Finally, we present a new \"looks linear\" (LL) initialization that prevents shattering. Preliminary experiments show the new initialization allows to train very deep networks without the addition of skip-connections.","title":"The Shattered Gradients Problem: If resnets are the answer, then what is the question?","author":[{"propositions":[],"lastnames":["Balduzzi"],"firstnames":["David"],"suffixes":[]},{"propositions":[],"lastnames":["Frean"],"firstnames":["Marcus"],"suffixes":[]},{"propositions":[],"lastnames":["Leary"],"firstnames":["Lennox"],"suffixes":[]},{"propositions":[],"lastnames":["Lewis"],"firstnames":["JP"],"suffixes":[]},{"propositions":[],"lastnames":["Ma"],"firstnames":["Kurt","Wan-Duo"],"suffixes":[]},{"firstnames":["McWilliams,"],"propositions":[],"lastnames":["Brian"],"suffixes":[]}],"bibtex":"@article{Balduzzi_The_2017,\n year={2017},\n abstract={A long-standing obstacle to progress in deep learning is the problem of vanishing and exploding gradients. The problem has largely been overcome through the introduction of carefully constructed initializations and batch normalization. Nevertheless, architectures incorporating skip-connections such as resnets perform much better than standard feedforward architectures despite well-chosen initialization and batch normalization. In this paper, we identify the shattered gradients problem. Specifically, we show that the correlation between gradients in standard feedforward networks decays exponentially with depth resulting in gradients that resemble white noise. In contrast, the gradients in architectures with skip-connections are far more resistant to shattering decaying sublinearly. Detailed empirical evidence is presented in support of the analysis, on both fully-connected networks and convnets. Finally, we present a new \"looks linear\" {(LL)} initialization that prevents shattering. Preliminary experiments show the new initialization allows to train very deep networks without the addition of skip-connections.},\n title={The Shattered Gradients Problem: If resnets are the answer, then what is the question?},\n author={Balduzzi, David and Frean, Marcus and Leary, Lennox and Lewis, {JP} and Ma, Kurt {Wan-Duo} and {McWilliams,} Brian}\n}\n","author_short":["Balduzzi, D.","Frean, M.","Leary, L.","Lewis, J.","Ma, K. W.","Brian, M."],"key":"Balduzzi_The_2017","id":"Balduzzi_The_2017","bibbaseid":"balduzzi-frean-leary-lewis-ma-brian-theshatteredgradientsproblemifresnetsaretheanswerthenwhatisthequestion-2017","role":"author","urls":{},"downloads":0},"search_terms":["shattered","gradients","problem","resnets","answer","question","balduzzi","frean","leary","lewis","ma","brian"],"keywords":[],"authorIDs":[],"dataSources":["NScDFpTBCtWwAQ6eq"]}