Applications Deep learning

1 applications

1.1 automatic speech recognition
1.2 image recognition
1.3 visual art processing
1.4 natural language processing
1.5 drug discovery , toxicology
1.6 customer relationship management
1.7 recommendation systems
1.8 bioinformatics
1.9 mobile advertising

applications
automatic speech recognition

large-scale automatic speech recognition first , convincing successful case of deep learning. lstm rnns can learn deep learning tasks involve multi-second intervals containing speech events separated thousands of discrete time steps, 1 time step corresponds 10 ms. lstm forget gates competitive traditional speech recognizers on tasks.

the initial success in speech recognition based on small-scale recognition tasks based on timit. data set contains 630 speakers 8 major dialects of american english, each speaker reads 10 sentences. small size allows many configurations tried. more importantly, timit task concerns phone-sequence recognition, which, unlike word-sequence recognition, allows weak language models (without strong grammar). allows weaknesses in acoustic modeling aspects of speech recognition more analyzed. error rates listed below, including these results , measured percent phone error rates (per), have been summarized on past 20 years:

the debut of dnns speaker recognition in late 1990s , speech recognition around 2009-2011 , of lstm around 2003-2007, accelerated progress in 8 major areas:

scale-up/out , acclerated dnn training , decoding
sequence discriminative training
feature processing deep models solid understanding of underlying mechanisms
adaptation of dnns , related deep models
multi-task , transfer learning dnns , related deep models
cnns , how design them best exploit domain knowledge of speech
rnn , rich lstm variants
other types of deep models including tensor-based models , integrated deep generative/discriminative models.

all major commercial speech recognition systems (e.g., microsoft cortana, xbox, skype translator, amazon alexa, google now, apple siri, baidu , iflytek voice search, , range of nuance speech products, etc.) based on deep learning.

image recognition

a common evaluation set image classification mnist database data set. mnist composed of handwritten digits , includes 60,000 training examples , 10,000 test examples. timit, small size allows multiple configurations tested. comprehensive list of results on set available.

deep learning-based image recognition has become superhuman , producing more accurate results human contestants. first occurred in 2011.

deep learning-trained vehicles interpret 360° camera views. example facial dysmorphology novel analysis (fdna) used analyze cases of human malformation connected large database of genetic syndromes.

visual art processing

closely related progress has been made in image recognition increasing application of deep learning techniques various visual art tasks. dnns have proven capable, example, of a) identifying style period of given painting, b) capturing style of given painting , applying in visually pleasing manner arbitrary photograph, , c) generating striking imagery based on random visual input fields.

natural language processing

neural networks have been used implementing language models since 2000s. lstm helped improve machine translation , language modeling.

other key techniques in field negative sampling , word embedding. word embedding, such word2vec, can thought of representational layer in deep learning architecture transforms atomic word positional representation of word relative other words in dataset; position represented point in vector space. using word embedding rnn input layer allows network parse sentences , phrases using effective compositional vector grammar. compositional vector grammar can thought of probabilistic context free grammar (pcfg) implemented rnn. recursive auto-encoders built atop word embeddings can assess sentence similarity , detect paraphrasing. deep neural architectures provide best results constituency parsing, sentiment analysis, information retrieval, spoken language understanding, machine translation, contextual entity linking, writing style recognition, categorizing clinical narratives, , others.

google translate (gt) uses large end-to-end long short-term memory network. gnmt uses example-based machine translation method in system learns millions of examples. translates whole sentences @ time, rather pieces. google translate supports on 1 hundred languages. network encodes semantics of sentence rather memorizing phrase-to-phrase translations . gt can translate directly 1 language another, rather using english intermediate.

drug discovery , toxicology

a large percentage of candidate drugs fail win regulatory approval. these failures caused insufficient efficacy (on-target effect), undesired interactions (off-target effects), or unanticipated toxic effects. research has explored use of deep learning predict biomolecular target, off-target , toxic effects of environmental chemicals in nutrients, household products , drugs.

atomnet deep learning system structure-based rational drug design. atomnet used predict novel candidate biomolecules disease targets such ebola virus , multiple sclerosis.

customer relationship management

deep reinforcement learning has been used approximate value of possible direct marketing actions, defined in terms of rfm variables. estimated value function shown have natural interpretation customer lifetime value.

recommendation systems

recommendation systems have used deep learning extract meaningful features latent factor model content-based music recommendations. multiview deep learning has been applied learning user preferences multiple domains. model uses hybrid collaborative , content-based approach , enhances recommendations in multiple tasks.

bioinformatics

an autoencoder ann used in bioinformatics, predict gene ontology annotations , gene-function relationships.

in medical informatics, deep learning used predict sleep quality based on data wearables , predictions of health complications electronic health record data.

an extension of word2vec used create semantic labels radiological images exploiting clinical narratives.

mobile advertising

finding appropriate mobile audience mobile advertising challenging since there many data points need considered , assimilated before target segment can created , used in ad serving ad server. deep learning has been used interpret large, many-dimensioned advertising datasets. many data points collected during request/serve/click internet advertising cycle. information can form basis of machine learning improve ad selection.