In this summary I like to provide a rough overview of Sequence-To-Sequence neural network architectures and what purposes these serve.<\/p>\n
Motivation<\/strong> Ingredients<\/strong> Steps<\/strong> For this reasons, altered neural networks architectures where proposed and pursued. Only within recent years, such tuned recurrent neural networks were used as encoders and decoders in the framework of Sequence-To-Sequence architectures. The Sequence-To-Sequence architectures as every machine learning system has to undergo a certain training process. As for every deep neural network the amount of available training data is crucial for achieving a good performance. Latest results show that such architectures are able to model dialogs of the previous form well. With focus on Musio and it’s emotional abilities it is of importance to generate specific datasets that capture proper emotional responses.<\/p>\n Resources In this summary I like to provide a rough overview of Sequence-To-Sequence neural network architectures and wh […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_exactmetrics_skip_tracking":false,"_exactmetrics_sitenote_active":false,"_exactmetrics_sitenote_note":"","_exactmetrics_sitenote_category":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[3642,3640],"tags":[3650,3652,3760,4304,3656,3762,3658,3700,3664,4002,4306,4006,4308,4274,4278,3710,4310,4312,3932,4010,3840,3906,3854,3712,3714,4314,4266],"class_list":["post-997","post","type-post","status-publish","format-standard","hentry","category-ai-en","category-all-en","tag-ai-ja-en","tag-aka-ja-en","tag-artificial-intelligence-en","tag-back-propagation-en","tag-baggage-en","tag-children-book-ja-en","tag-christmas-en","tag-cmos-en","tag-crowd-funding-en","tag-decoder-en","tag-deep-neural-network-en","tag-encoder-en","tag-hidden-layer-en","tag-hinoki-en","tag-lstm-en","tag-musio-en","tag-n-gram-en","tag-naive-bayesian-classifier-en","tag-natural-conversation-ja-en","tag-natural-interaction-en","tag-natural-language-processing-ja-en","tag-negative-sampling-en","tag-neural-network-en","tag-neural-networks-en","tag-recurrent-neural-network-en","tag-sentence-embedding-en","tag-word-embedding-en"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t
\nA key observation when dealing with neural networks is that these can only handle objects of a fixed size.
\nThis means that the architecture has to be adopted if sequences like sentences should be process-able.
\nThe same problems with objects of variable length also appear on the dialog level, where a certain number of utterances and responses string together.\u00a0Besides dialog modeling, speech recognition and machine translation demand for advanced neural networks.<\/p>\n
\nDeep neural network, hidden layer, recurrent neural network, encoder, decoder, LSTM, back-propagation, word embedding, sentence embedding<\/p>\n
\nAs already stated standard neural networks can not deal with sequences of variable lengths.
\nMoreover they have no knowledge of the previous input.
\nHowever this is of big importance for understanding sentences for example.<\/p>\n
\nA first step are recurrent neural networks which can process variable sequences of fixed size objects, as words in sentences.
\nThey also solve the problem of keeping knowledge about the previous input by passing a state in the hidden layer.
\nFor certain mathematical shortcomings of standard neurons, this does not allow for endless stretching back memories about previous inputs.
\nA proposal to deal with these memory issues are Long-Short-Term Memory and GRU cells replacing neurons.<\/p>\n
\nThe encoder part maps an input, e.g. a sequence of words to a fixed-size vector by processing word by word.
\nThe vector output can then be considered as an sentence embedding, which abstractly stores the meaning.
\nIn a second step the decoder maps the abstract vector into an output sequence, by spitting out word by word.
\nIn this way the network architecture is able to respond to an utterance with an response.
\nLast year this concept was generalized to including a dialog encoder layer on top of the standard encoder.
\nThis might further enhance the architecture to keep track of previous utterances in a full dialog.<\/p>\n
\nHere, the encoder and the decoder are trained together by presenting corresponding sequence pairs to them.
\nOptimization methods using back-propagation algorithms can be borrowed from standard neural networks.<\/p>\n
\nInteresting data sets for dialog modeling exist as Subtitles of Movie corpora or Scripts for theater plays and TV series.<\/p>\n
\nFor specific data sets from IT help-desk discussions a given technical problem can sometimes be addressed properly.<\/p>\n
\n<\/strong>“Sequence to Sequence Learning with Neural Networks<\/a>“(PDF). Sequence to Sequence Learning with Neural Networks. <\/em>December 2014.\u00a0Retrieved Feburary 26, 2016.
\n“Learning Phrase Representations using RNN Encoder\u2013Decoder for Statistical Machine Translation<\/a>“(PDF). Learning Phrase Representations using RNN Encoder\u2013Decoder for Statistical Machine Translation. September 2014.\u00a0Retrieved Feburary\u00a026, 2016.
\n“Building End-To-End Dialogue Systems:\u00a0Using Generative Hierarchical Neural Network Models<\/a>“(PDF).\u00a0Building End-To-End Dialogue Systems:\u00a0Using Generative Hierarchical Neural Network Models<\/em>. November 2015.\u00a0Retrieved Feburary\u00a026, 2016.
\n“Recurrent Continuous Translation Models<\/a>” (PDF). Recurrent Continuous Translation Models<\/em>. October 2013.\u00a0Retrieved Feburary\u00a026, 2016.
\n“Generating Sequences With Recurrent Neural Networks<\/a>” (PDF). Generating Sequences With Recurrent Neural Networks.\u00a0<\/em>June\u00a02014.\u00a0Retrieved Feburary\u00a026, 2016.
\n“NEURAL MACHINE TRANSLATION\u00a0BY JOINTLY LEARNING TO ALIGN AND TRANSLATE<\/a>” (PDF).\u00a0NEURAL MACHINE TRANSLATION\u00a0BY JOINTLY LEARNING TO ALIGN AND TRANSLATE. <\/em>April 2015.\u00a0Retrieved Feburary\u00a026, 2016.
\n“Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks<\/a>” (PDF). Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks. 2006.\u00a0Retrieved Feburary\u00a026, 2016.
\n“Hierarchical Encoder Decoder for Dialog Modelling<\/a>” (GIT).\u00a0Hierarchical Encoder Decoder for Dialog Modelling<\/em>.\u00a0Retrieved February 26, 2016.<\/p>\n","protected":false},"excerpt":{"rendered":"