@article{Karpathy2015,
abstract = {Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study.},
archivePrefix = {arXiv},
arxivId = {1506.02078},
author = {Karpathy, Andrej and Johnson, Justin and Fei-Fei, Li},
eprint = {1506.02078},
file = {:Users/shanest/Documents/Library/Karpathy, Johnson, Fei-Fei/Unknown/Karpathy, Johnson, Fei-Fei - 2015 - Visualizing and Understanding Recurrent Networks.pdf:pdf},
keywords = {method: individual neurons,method: visualization},
pages = {1--12},
title = {{Visualizing and Understanding Recurrent Networks}},
url = {http://arxiv.org/abs/1506.02078},
year = {2015}
}

{"_id":"Xxjyi96ND2TbBNTke","bibbaseid":"karpathy-johnson-feifei-visualizingandunderstandingrecurrentnetworks-2015","authorIDs":[],"author_short":["Karpathy, A.","Johnson, J.","Fei-Fei, L."],"bibdata":{"bibtype":"article","type":"article","abstract":"Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study.","archiveprefix":"arXiv","arxivid":"1506.02078","author":[{"propositions":[],"lastnames":["Karpathy"],"firstnames":["Andrej"],"suffixes":[]},{"propositions":[],"lastnames":["Johnson"],"firstnames":["Justin"],"suffixes":[]},{"propositions":[],"lastnames":["Fei-Fei"],"firstnames":["Li"],"suffixes":[]}],"eprint":"1506.02078","file":":Users/shanest/Documents/Library/Karpathy, Johnson, Fei-Fei/Unknown/Karpathy, Johnson, Fei-Fei - 2015 - Visualizing and Understanding Recurrent Networks.pdf:pdf","keywords":"method: individual neurons,method: visualization","pages":"1–12","title":"Visualizing and Understanding Recurrent Networks","url":"http://arxiv.org/abs/1506.02078","year":"2015","bibtex":"@article{Karpathy2015,\nabstract = {Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study.},\narchivePrefix = {arXiv},\narxivId = {1506.02078},\nauthor = {Karpathy, Andrej and Johnson, Justin and Fei-Fei, Li},\neprint = {1506.02078},\nfile = {:Users/shanest/Documents/Library/Karpathy, Johnson, Fei-Fei/Unknown/Karpathy, Johnson, Fei-Fei - 2015 - Visualizing and Understanding Recurrent Networks.pdf:pdf},\nkeywords = {method: individual neurons,method: visualization},\npages = {1--12},\ntitle = {{Visualizing and Understanding Recurrent Networks}},\nurl = {http://arxiv.org/abs/1506.02078},\nyear = {2015}\n}\n","author_short":["Karpathy, A.","Johnson, J.","Fei-Fei, L."],"key":"Karpathy2015","id":"Karpathy2015","bibbaseid":"karpathy-johnson-feifei-visualizingandunderstandingrecurrentnetworks-2015","role":"author","urls":{"Paper":"http://arxiv.org/abs/1506.02078"},"keyword":["method: individual neurons","method: visualization"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://www.shane.st/teaching/575/win20/MachineLearning-interpretability.bib","creationDate":"2020-01-06T00:27:01.054Z","downloads":0,"keywords":["method: individual neurons","method: visualization"],"search_terms":["visualizing","understanding","recurrent","networks","karpathy","johnson","fei-fei"],"title":"Visualizing and Understanding Recurrent Networks","year":2015,"dataSources":["okYcdTpf4JJ2zkj7A"]}