Early forms of neural networks were inspired by information processing and distributed communication nodes inbiological systems, particularly thehuman brain. However, current neural networks do not intend to model the brain function of organisms, and are generally seen as low-quality models for that purpose.[6]
Fundamentally, deep learning refers to a class ofmachine learningalgorithms in which a hierarchy of layers is used to transform input data into a progressively more abstract and composite representation. For example, in animage recognition model, the raw input may be animage (represented as atensor ofpixels). The first representational layer may attempt to identify basic shapes such as lines and circles, the second layer may compose and encode arrangements of edges, the third layer may encode a nose and eyes, and the fourth layer may recognize that the image contains a face.
Importantly, a deep learning process can learn which features to optimally place at which levelon its own. Prior to deep learning, machine learning techniques often involved hand-craftedfeature engineering to transform the data into a more suitable representation for a classification algorithm to operate on. In the deep learning approach, features are not hand-crafted and the modeldiscovers useful feature representations from the data automatically. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction.[8][2]
The word "deep" in "deep learning" refers to the number of layers through which the data is transformed. More precisely, deep learning systems have a substantialcredit assignment path (CAP) depth. The CAP is the chain of transformations from input to output. CAPs describe potentially causal connections between input and output. For afeedforward neural network, the depth of the CAPs is that of the network and is the number of hidden layers plus one (as the output layer is also parameterized). Forrecurrent neural networks, in which a signal may propagate through a layer more than once, the CAP depth is potentially unlimited.[9] No universally agreed-upon threshold of depth divides shallow learning from deep learning, but most researchers agree that deep learning involves CAP depth higher than two. CAP of depth two has been shown to be a universal approximator in the sense that it can emulate any function.[10] Beyond that, more layers do not add to the function approximator ability of the network. Deep models (CAP > two) are able to extract better features than shallow models and hence, extra layers help in learning the features effectively.
Deep learning architectures can be constructed with agreedy layer-by-layer method.[11] Deep learning helps to disentangle these abstractions and pick out which features improve performance.[8]
Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data are more abundant than the labeled data. Examples of deep structures that can be trained in an unsupervised manner aredeep belief networks.[8][12]
The termDeep Learning was introduced to the machine learning community byRina Dechter in 1986,[13] and to artificial neural networks by Igor Aizenberg and colleagues in 2000, in the context ofBoolean threshold neurons.[14][15] Although the history of its appearance is apparently more complicated.[16]
The universal approximation theorem fordeep neural networks concerns the capacity of networks with bounded width but the depth is allowed to grow. Lu et al.[21] proved that if the width of a deep neural network withReLU activation is strictly larger than the input dimension, then the network can approximate anyLebesgue integrable function; if the width is smaller or equal to the input dimension, then a deep neural network is not a universal approximator.
Frank Rosenblatt (1958)[36] proposed the perceptron, an MLP with 3 layers: an input layer, a hidden layer with randomized weights that did not learn, and an output layer. He later published a 1962 book that also introduced variants and computer experiments, including a version with four-layer perceptrons "with adaptive preterminal networks" where the last two layers have learned weights (here he credits H. D. Block and B. W. Knight).[37]: section 16 The book cites an earlier network by R. D. Joseph (1960)[38] "functionally equivalent to a variation of" this four-layer system (the book mentions Joseph over 30 times). Should Joseph therefore be considered the originator of proper adaptivemultilayer perceptrons with learning hidden units? Unfortunately, the learning algorithm was not a functional one, and fell into oblivion.
The first working deep learning algorithm was theGroup method of data handling, a method to train arbitrarily deep neural networks, published byAlexey Ivakhnenko and Lapa in 1965. They regarded it as a form of polynomial regression,[39] or a generalization of Rosenblatt's perceptron.[40] A 1971 paper described a deep network with eight layers trained by this method,[41] which is based on layer by layer training through regression analysis. Superfluous hidden units are pruned using a separate validation set. Since the activation functions of the nodes are Kolmogorov-Gabor polynomials, these were also the first deep networks with multiplicative units or "gates".[31]
Backpropagation is an efficient application of thechain rule derived byGottfried Wilhelm Leibniz in 1673[47] to networks of differentiable nodes. The terminology "back-propagating errors" was actually introduced in 1962 by Rosenblatt,[37] but he did not know how to implement this, althoughHenry J. Kelley had a continuous precursor of backpropagation in 1960 in the context ofcontrol theory.[48] The modern form of backpropagation was first published inSeppo Linnainmaa's master thesis (1970).[49][50][31] G.M. Ostrovski et al. republished it in 1971.[51][52]Paul Werbos applied backpropagation to neural networks in 1982[53] (his 1974 PhD thesis, reprinted in a 1994 book,[54] did not yet describe the algorithm[52]). In 1986,David E. Rumelhart et al. popularised backpropagation but did not cite the original work.[55][56]
Thetime delay neural network (TDNN) was introduced in 1987 byAlex Waibel to apply CNN to phoneme recognition. It used convolutions, weight sharing, and backpropagation.[57][58] In 1988, Wei Zhang applied a backpropagation-trained CNN to alphabet recognition.[59] In 1989,Yann LeCun et al. created a CNN calledLeNet forrecognizing handwritten ZIP codes on mail. Training required 3 days.[60] In 1990, Wei Zhang implemented a CNN onoptical computing hardware.[61] In 1991, a CNN was applied to medical image object segmentation[62] and breast cancer detection in mammograms.[63]LeNet-5 (1998), a 7-level CNN byYann LeCun et al., that classifies digits, was applied by several banks to recognize hand-written numbers on checks digitized in 32x32 pixel images.[64]
Recurrent neural networks (RNN)[28][30] were further developed in the 1980s. Recurrence is used for sequence processing, and when a recurrent network is unrolled, it mathematically resembles a deep feedforward layer. Consequently, they have similar properties and issues, and their developments had mutual influences. In RNN, two early influential works were theJordan network (1986)[65] and theElman network (1990),[66] which applied RNN to study problems incognitive psychology.
In the 1980s, backpropagation did not work well for deep learning with long credit assignment paths. To overcome this problem, in 1991,Jürgen Schmidhuber proposed a hierarchy of RNNs pre-trained one level at a time byself-supervised learning where each RNN tries to predict its own next input, which is the next unexpected input of the RNN below.[67][68] This "neural history compressor" usespredictive coding to learninternal representations at multiple self-organizing time scales. This can substantially facilitate downstream deep learning. The RNN hierarchy can becollapsed into a single RNN, bydistilling a higher levelchunker network into a lower levelautomatizer network.[67][68][31] In 1993, a neural history compressor solved a "Very Deep Learning" task that required more than 1000 subsequentlayers in an RNN unfolded in time.[69] The "P" inChatGPT refers to such pre-training.
Sepp Hochreiter's diploma thesis (1991)[70] implemented the neural history compressor,[67] and identified and analyzed thevanishing gradient problem.[70][71] Hochreiter proposed recurrentresidual connections to solve the vanishing gradient problem. This led to thelong short-term memory (LSTM), published in 1995.[72] LSTM can learn "very deep learning" tasks[9] with long credit assignment paths that require memories of events that happened thousands of discrete time steps before. That LSTM was not yet the modern architecture, which required a "forget gate", introduced in 1999,[73] which became the standard RNN architecture.
During 1985–1995, inspired by statistical mechanics, several architectures and methods were developed byTerry Sejnowski,Peter Dayan,Geoffrey Hinton, etc., including theBoltzmann machine,[77]restricted Boltzmann machine,[78]Helmholtz machine,[79] and thewake-sleep algorithm.[80] These were designed for unsupervised learning of deep generative models. However, those were more computationally expensive compared to backpropagation. Boltzmann machine learning algorithm, published in 1985, was briefly popular before being eclipsed by the backpropagation algorithm in 1986. (p. 112[81]). A 1988 network became state of the art inprotein structure prediction, an early application of deep learning to bioinformatics.[82]
Both shallow and deep learning (e.g., recurrent nets) of ANNs forspeech recognition have been explored for many years.[83][84][85] These methods never outperformed non-uniform internal-handcrafting Gaussianmixture model/Hidden Markov model (GMM-HMM) technology based on generative models of speech trained discriminatively.[86] Key difficulties have been analyzed, including gradient diminishing[70] and weak temporal correlation structure in neural predictive models.[87][88] Additional difficulties were the lack of training data and limited computing power.
Mostspeech recognition researchers moved away from neural nets to pursue generative modeling. An exception was atSRI International in the late 1990s. Funded by the US government'sNSA andDARPA, SRI researched in speech andspeaker recognition. The speaker recognition team led byLarry Heck reported significant success with deep neural networks in speech processing in the 1998NIST Speaker Recognition benchmark.[89][90] It was deployed in the Nuance Verifier, representing the first major industrial application of deep learning.[91]
The principle of elevating "raw" features over hand-crafted optimization was first explored successfully in the architecture of deep autoencoder on the "raw" spectrogram or linearfilter-bank features in the late 1990s,[90] showing its superiority over theMel-Cepstral features that contain stages of fixed transformation from spectrograms. The raw features of speech,waveforms, later produced excellent larger-scale results.[92]
Neural networks entered a null, and simpler models that use task-specific handcrafted features such asGabor filters andsupport vector machines (SVMs) became the preferred choices in the 1990s and 2000s, because of artificial neural networks' computational cost and a lack of understanding of how the brain wires its biological networks.[citation needed]
In 2006, publications byGeoff Hinton,Ruslan Salakhutdinov, Osindero andTeh[97][98]deep belief networks were developed for generative modeling. They are trained by training one restricted Boltzmann machine, then freezing it and training another one on top of the first one, and so on, then optionallyfine-tuned using supervised backpropagation.[99] They could model high-dimensional probability distributions, such as the distribution ofMNIST images, but convergence was slow.[100][101][102]
The impact of deep learning in industry began in the early 2000s, when CNNs already processed an estimated 10% to 20% of all the checks written in the US, according to Yann LeCun.[103] Industrial applications of deep learning to large-scale speech recognition started around 2010.
The 2009 NIPS Workshop on Deep Learning for Speech Recognition was motivated by the limitations of deep generative models of speech, and the possibility that given more capable hardware and large-scale data sets that deep neural nets might become practical. It was believed that pre-training DNNs using generative models of deep belief nets (DBN) would overcome the main difficulties of neural nets. However, it was discovered that replacing pre-training with large amounts of training data for straightforward backpropagation when using DNNs with large, context-dependent output layers produced error rates dramatically lower than then-state-of-the-art Gaussian mixture model (GMM)/Hidden Markov Model (HMM) and also than more-advanced generative model-based systems.[104] The nature of the recognition errors produced by the two types of systems was characteristically different,[105] offering technical insights into how to integrate deep learning into the existing highly efficient, run-time speech decoding system deployed by all major speech recognition systems.[23][106][107] Analysis around 2009–2010, contrasting the GMM (and other generative speech models) vs. DNN models, stimulated early industrial investment in deep learning for speech recognition.[105] That analysis was done with comparable performance (less than 1.5% in error rate) between discriminative DNNs and generative models.[104][105][108]In 2010, researchers extended deep learning fromTIMIT to large vocabulary speech recognition, by adopting large output layers of the DNN based on context-dependent HMM states constructed bydecision trees.[109][110][111][106]
How deep learning is a subset of machine learning and how machine learning is a subset of artificial intelligence (AI)
The deep learning revolution started around CNN- and GPU-based computer vision.
Although CNNs trained by backpropagation had been around for decades and GPU implementations of NNs for years,[112] including CNNs,[113] faster implementations of CNNs on GPUs were needed to progress on computer vision. Later, as deep learning becomes widespread, specialized hardware and algorithm optimizations were developed specifically for deep learning.[114]
A key advance for the deep learning revolution was hardware advances, especially GPU. Some early work dated back to 2004.[112][113] In 2009, Raina, Madhavan, andAndrew Ng reported a 100M deep belief network trained on 30 NvidiaGeForce GTX 280 GPUs, an early demonstration of GPU-based deep learning. They reported up to 70 times faster training.[115]
In 2011, a CNN namedDanNet[116][117] by Dan Ciresan, Ueli Meier, Jonathan Masci,Luca Maria Gambardella, andJürgen Schmidhuber achieved for the first time superhuman performance in a visual pattern recognition contest, outperforming traditional methods by a factor of 3.[9] It then won more contests.[118][119] They also showed howmax-pooling CNNs on GPU improved performance significantly.[3]
In 2012,Andrew Ng andJeff Dean created an FNN that learned to recognize higher-level concepts, such as cats, only from watching unlabeled images taken fromYouTube videos.[120]
The success in image classification was then extended to the more challenging task ofgenerating descriptions (captions) for images, often as a combination of CNNs and LSTMs.[123][124][125]
In 2014, the state of the art was training “very deep neural network” with 20 to 30 layers.[126] Stacking too many layers led to a steep reduction intraining accuracy,[127] known as the "degradation" problem.[128] In 2015, two techniques were developed to train very deep networks: the Highway Network was published in May 2015, and theresidual neural network (ResNet)[129] in Dec 2015. ResNet behaves like an open-gated Highway Net.
Around the same time, deep learning started impacting the field of art. Early examples includedGoogle DeepDream (2015), andneural style transfer (2015),[130] both of which were based on pretrained image classification neural networks, such asVGG-19.
Generative adversarial network (GAN) by (Ian Goodfellow et al., 2014)[131] (based onJürgen Schmidhuber's principle of artificial curiosity[74][76])became state of the art in generative modeling during 2014-2018 period. Excellent image quality is achieved byNvidia'sStyleGAN (2018)[132] based on the Progressive GAN by Tero Karras et al.[133] Here the GAN generator is grown from small to large scale in a pyramidal fashion. Image generation by GAN reached popular success, and provoked discussions concerningdeepfakes.[134]Diffusion models (2015)[135] eclipsed GANs in generative modeling since then, with systems such asDALL·E 2 (2022) andStable Diffusion (2022).
Deep learning is part of state-of-the-art systems in various disciplines, particularly computer vision andautomatic speech recognition (ASR). Results on commonly used evaluation sets such asTIMIT (ASR) andMNIST (image classification), as well as a range of large-vocabulary speech recognition tasks have steadily improved.[104][138] Convolutional neural networks were superseded for ASR byLSTM.[137][139][140][141] but are more successful in computer vision.
Simplified example of training a neural network in object detection: The network is trained by multiple images that are known to depictstarfish andsea urchins, which are correlated with "nodes" that represent visualfeatures. The starfish match with a ringed texture and a star outline, whereas most sea urchins match with a striped texture and oval shape. However, the instance of a ring textured sea urchin creates a weakly weighted association between them.
Subsequent run of the network on an input image (left):[143] The network correctly detects the starfish. However, the weakly weighted association between ringed texture and sea urchin also confers a weak signal to the latter from one of two intermediate nodes. In addition, ashell that was not included in the training gives a weak signal for the oval shape, also resulting in a weak signal for the sea urchin output. These weak signals may result in afalse positive result for sea urchin. In reality, textures and outlines would not be represented by single nodes, but rather by associated weight patterns of multiple nodes.
Artificial neural networks (ANNs) orconnectionist systems are computing systems inspired by thebiological neural networks that constitute animal brains. Such systems learn (progressively improve their ability) to do tasks by considering examples, generally without task-specific programming. For example, in image recognition, they might learn to identify images that contain cats by analyzing example images that have been manuallylabeled as "cat" or "no cat" and using the analytic results to identify cats in other images. They have found most use in applications difficult to express with a traditional computer algorithm usingrule-based programming.
An ANN is based on a collection of connected units calledartificial neurons, (analogous to biologicalneurons in abiological brain). Each connection (synapse) between neurons can transmit a signal to another neuron. The receiving (postsynaptic) neuron can process the signal(s) and then signal downstream neurons connected to it. Neurons may have state, generally represented byreal numbers, typically between 0 and 1. Neurons and synapses may also have a weight that varies as learning proceeds, which can increase or decrease the strength of the signal that it sends downstream.
Typically, neurons are organized in layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first (input), to the last (output) layer, possibly after traversing the layers multiple times.
The original goal of the neural network approach was to solve problems in the same way that a human brain would. Over time, attention focused on matching specific mental abilities, leading to deviations from biology such asbackpropagation, or passing information in the reverse direction and adjusting the network to reflect that information.
As of 2017, neural networks typically have a few thousand to a few million units and millions of connections. Despite this number being several order of magnitude less than the number of neurons on a human brain, these networks can perform many tasks at a level beyond that of humans (e.g., recognizing faces, or playing "Go"[144]).
A deep neural network (DNN) is an artificial neural network with multiple layers between the input and output layers.[7][9] There are different types of neural networks but they always consist of the same components: neurons, synapses, weights, biases, and functions.[145] These components as a whole function in a way that mimics functions of the human brain, and can be trained like any other ML algorithm.[citation needed]
For example, a DNN that is trained to recognize dog breeds will go over the given image and calculate the probability that the dog in the image is a certain breed. The user can review the results and select which probabilities the network should display (above a certain threshold, etc.) and return the proposed label. Each mathematical manipulation as such is considered a layer,[146] and complex DNN have many layers, hence the name "deep" networks.
DNNs can model complex non-linear relationships. DNN architectures generate compositional models where the object is expressed as a layered composition ofprimitives.[147] The extra layers enable composition of features from lower layers, potentially modeling complex data with fewer units than a similarly performing shallow network.[7] For instance, it was proved that sparsemultivariate polynomials are exponentially easier to approximate with DNNs than with shallow networks.[148]
Deep architectures include many variants of a few basic approaches. Each architecture has found success in specific domains. It is not always possible to compare the performance of multiple architectures, unless they have been evaluated on the same data sets.[146]
DNNs are typically feedforward networks in which data flows from the input layer to the output layer without looping back. At first, the DNN creates a map of virtual neurons and assigns random numerical values, or "weights", to connections between them. The weights and inputs are multiplied and return an output between 0 and 1. If the network did not accurately recognize a particular pattern, an algorithm would adjust the weights.[149] That way the algorithm can make certain parameters more influential, until it determines the correct mathematical manipulation to fully process the data.
As with ANNs, many issues can arise with naively trained DNNs. Two common issues areoverfitting and computation time.
DNNs are prone to overfitting because of the added layers of abstraction, which allow them to model rare dependencies in the training data.Regularization methods such as Ivakhnenko's unit pruning[41] orweight decay (-regularization) orsparsity (-regularization) can be applied during training to combat overfitting.[159] Alternativelydropout regularization randomly omits units from the hidden layers during training. This helps to exclude rare dependencies.[160] Another interesting recent development is research into models of just enough complexity through an estimation of the intrinsic complexity of the task being modelled. This approach has been successfully applied for multivariate time series prediction tasks such as traffic prediction.[161] Finally, data can be augmented via methods such as cropping and rotating such that smaller training sets can be increased in size to reduce the chances of overfitting.[162]
DNNs must consider many training parameters, such as the size (number of layers and number of units per layer), thelearning rate, and initial weights.Sweeping through the parameter space for optimal parameters may not be feasible due to the cost in time and computational resources. Various tricks, such asbatching (computing the gradient on several training examples at once rather than individual examples)[163] speed up computation. Large processing capabilities of many-core architectures (such as GPUs or the Intel Xeon Phi) have produced significant speedups in training, because of the suitability of such processing architectures for the matrix and vector computations.[164][165]
Alternatively, engineers may look for other types of neural networks with more straightforward and convergent training algorithms. CMAC (cerebellar model articulation controller) is one such kind of neural network. It doesn't require learning rates or randomized initial weights. The training process can be guaranteed to converge in one step with a new batch of data, and the computational complexity of the training algorithm is linear with respect to the number of neurons involved.[166][167]
Since the 2010s, advances in both machine learning algorithms andcomputer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[168] By 2019, graphics processing units (GPUs), often with AI-specific enhancements, had displaced CPUs as the dominant method for training large-scale commercial cloud AI .[169]OpenAI estimated the hardware computation used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017) and found a 300,000-fold increase in the amount of computation required, with a doubling-time trendline of 3.4 months.[170][171]
Atomically thinsemiconductors are considered promising for energy-efficient deep learning hardware where the same basic device structure is used for both logic operations and data storage.In 2020, Marega et al. published experiments with a large-area active channel material for developing logic-in-memory devices and circuits based onfloating-gatefield-effect transistors (FGFETs).[176]
In 2021, J. Feldmann et al. proposed an integratedphotonichardware accelerator for parallel convolutional processing.[177] The authors identify two key advantages of integrated photonics over its electronic counterparts: (1) massively parallel data transfer throughwavelength divisionmultiplexing in conjunction withfrequency combs, and (2) extremely high data modulation speeds.[177] Their system can execute trillions of multiply-accumulate operations per second, indicating the potential ofintegratedphotonics in data-heavy AI applications.[177]
Large-scale automatic speech recognition is the first and most convincing successful case of deep learning. LSTM RNNs can learn "Very Deep Learning" tasks[9] that involve multi-second intervals containing speech events separated by thousands of discrete time steps, where one time step corresponds to about 10 ms. LSTM with forget gates[156] is competitive with traditional speech recognizers on certain tasks.[93]
The initial success in speech recognition was based on small-scale recognition tasks based on TIMIT. The data set contains 630 speakers from eight majordialects ofAmerican English, where each speaker reads 10 sentences.[178] Its small size lets many configurations be tried. More importantly, the TIMIT task concernsphone-sequence recognition, which, unlike word-sequence recognition, allows weak phonebigram language models. This lets the strength of the acoustic modeling aspects of speech recognition be more easily analyzed. The error rates listed below, including these early results and measured as percent phone error rates (PER), have been summarized since 1991.
Hierarchical Convolutional Deep Maxout Network[182]
16.5
The debut of DNNs for speaker recognition in the late 1990s and speech recognition around 2009-2011 and of LSTM around 2003–2007, accelerated progress in eight major areas:[23][108][106]
Scale-up/out and accelerated DNN training and decoding
Sequence discriminative training
Feature processing by deep models with solid understanding of the underlying mechanisms
A common evaluation set for image classification is theMNIST database data set. MNIST is composed of handwritten digits and includes 60,000 training examples and 10,000 test examples. As with TIMIT, its small size lets users test multiple configurations. A comprehensive list of results on this set is available.[185]
Deep learning-based image recognition has become "superhuman", producing more accurate results than human contestants. This first occurred in 2011 in recognition of traffic signs, and in 2014, with recognition of human faces.[186][187]
Deep learning-trained vehicles now interpret 360° camera views.[188] Another example is Facial Dysmorphology Novel Analysis (FDNA) used to analyze cases of human malformation connected to a large database of genetic syndromes.
Visual art processing of Jimmy Wales in France, with the style of Munch's "The Scream" applied using neural style transfer
Closely related to the progress that has been made in image recognition is the increasing application of deep learning techniques to various visual art tasks. DNNs have proven themselves capable, for example, of
identifying the style period of a given painting[189][190]
Neural Style Transfer – capturing the style of a given artwork and applying it in a visually pleasing manner to an arbitrary photograph or video[189][190]
generating striking imagery based on random visual input fields.[189][190]
Neural networks have been used for implementing language models since the early 2000s.[150] LSTM helped to improve machine translation and language modeling.[151][152][153]
Other key techniques in this field are negative sampling[191] andword embedding. Word embedding, such asword2vec, can be thought of as a representational layer in a deep learning architecture that transforms an atomic word into a positional representation of the word relative to other words in the dataset; the position is represented as a point in avector space. Using word embedding as an RNN input layer allows the network to parse sentences and phrases using an effective compositional vector grammar. A compositional vector grammar can be thought of asprobabilistic context free grammar (PCFG) implemented by an RNN.[192] Recursive auto-encoders built atop word embeddings can assess sentence similarity and detect paraphrasing.[192] Deep neural architectures provide the best results for constituency parsing,[193]sentiment analysis,[194] information retrieval,[195][196] spoken language understanding,[197] machine translation,[151][198] contextual entity linking,[198] writing style recognition,[199]named-entity recognition (token classification),[200] text classification, and others.[201]
A large percentage of candidate drugs fail to win regulatory approval. These failures are caused by insufficient efficacy (on-target effect), undesired interactions (off-target effects), or unanticipatedtoxic effects.[207][208] Research has explored use of deep learning to predict thebiomolecular targets,[209][210]off-targets, andtoxic effects of environmental chemicals in nutrients, household products and drugs.[211][212][213]
In 2017graph neural networks were used for the first time to predict various properties of molecules in a large toxicology data set.[217] In 2019, generative neural networks were used to produce molecules that were validated experimentally all the way into mice.[218][219]
Recommendation systems have used deep learning to extract meaningful features for a latent factor model for content-based music and journal recommendations.[221][222] Multi-view deep learning has been applied for learning user preferences from multiple domains.[223] The model uses a hybrid collaborative and content-based approach and enhances recommendations in multiple tasks.
In medical informatics, deep learning was used to predict sleep quality based on data from wearables[225] and predictions of health complications fromelectronic health record data.[226]
Deep neural networks have shown unparalleled performance inpredicting protein structure, according to the sequence of the amino acids that make it up. In 2020,AlphaFold, a deep-learning based system, achieved a level of accuracy significantly higher than all previous computational methods.[227][228]
Deep neural networks can be used to estimate the entropy of astochastic process and called Neural Joint Entropy Estimator (NJEE).[229] Such an estimation provides insights on the effects of inputrandom variables on an independentrandom variable. Practically, the DNN is trained as aclassifier that maps an inputvector ormatrix X to an outputprobability distribution over the possible classes of random variable Y, given input X. For example, inimage classification tasks, the NJEE maps a vector ofpixels' color values to probabilities over possible image classes. In practice, the probability distribution of Y is obtained by aSoftmax layer with number of nodes that is equal to thealphabet size of Y. NJEE uses continuously differentiableactivation functions, such that the conditions for theuniversal approximation theorem holds. It is shown that this method provides a stronglyconsistent estimator and outperforms other methods in case of large alphabet sizes.[229]
Deep learning has been shown to produce competitive results in medical application such as cancer cell classification, lesion detection, organ segmentation and image enhancement.[230][231] Modern deep learning tools demonstrate the high accuracy of detecting various diseases and the helpfulness of their use by specialists to improve the diagnosis efficiency.[232][233]
Finding the appropriate mobile audience formobile advertising is always challenging, since many data points must be considered and analyzed before a target segment can be created and used in ad serving by any ad server.[234] Deep learning has been used to interpret large, many-dimensioned advertising datasets. Many data points are collected during the request/serve/click internet advertising cycle. This information can form the basis of machine learning to improve ad selection.
In November 2023, researchers atGoogle DeepMind andLawrence Berkeley National Laboratory announced that they had developed an AI system known as GNoME. This system has contributed tomaterials science by discovering over 2 million new materials within a relatively short timeframe. GNoME employs deep learning techniques to efficiently explore potential material structures, achieving a significant increase in the identification of stable inorganiccrystal structures. The system's predictions were validated through autonomous robotic experiments, demonstrating a noteworthy success rate of 71%. The data of newly discovered materials is publicly available through theMaterials Project database, offering researchers the opportunity to identify materials with desired properties for various applications. This development has implications for the future of scientific discovery and the integration of AI in material science research, potentially expediting material innovation and reducing costs in product development. The use of AI and deep learning suggests the possibility of minimizing or eliminating manual lab experiments and allowing scientists to focus more on the design and analysis of unique compounds.[239][240][241]
Physics informed neural networks have been used to solvepartial differential equations in both forward and inverse problems in a data driven manner.[243] One example is the reconstructing fluid flow governed by theNavier-Stokes equations. Using physics informed neural networks does not require the often expensive mesh generation that conventionalCFD methods rely on.[244][245]
Deep backward stochastic differential equation method
Deep backward stochastic differential equation method is a numerical method that combines deep learning withBackward stochastic differential equation (BSDE). This method is particularly useful for solving high-dimensional problems in financial mathematics. By leveraging the powerful function approximation capabilities ofdeep neural networks, deep BSDE addresses the computational challenges faced by traditional numerical methods in high-dimensional settings. Specifically, traditional methods like finite difference methods or Monte Carlo simulations often struggle with the curse of dimensionality, where computational cost increases exponentially with the number of dimensions. Deep BSDE methods, however, employ deep neural networks to approximate solutions of high-dimensional partial differential equations (PDEs), effectively reducing the computational burden.[246]
In addition, the integration ofPhysics-informed neural networks (PINNs) into the deep BSDE framework enhances its capability by embedding the underlying physical laws directly into the neural network architecture. This ensures that the solutions not only fit the data but also adhere to the governing stochastic differential equations. PINNs leverage the power of deep learning while respecting the constraints imposed by the physical models, resulting in more accurate and reliable solutions for financial mathematics problems.
Image reconstruction is the reconstruction of the underlying images from the image-related measurements. Several works showed the better and superior performance of the deep learning methods compared to analytical methods for various applications, e.g., spectral imaging[247] and ultrasound imaging.[248]
Traditional weather prediction systems solve a very complex system of partial differential equations. GraphCast is a deep learning based model, trained on a long history of weather data to predict how weather patterns change over time. It is able to predict weather conditions for up to 10 days globally, at a very detailed level, and in under a minute, with precision similar to state of the art systems.[249][250]
An epigenetic clock is abiochemical test that can be used to measure age. Galkin et al. used deep neural networks to train an epigenetic aging clock of unprecedented accuracy using >6,000 blood samples.[251] The clock uses information from 1000CpG sites and predicts people with certain conditions older than healthy controls:IBD,frontotemporal dementia,ovarian cancer,obesity. The aging clock was planned to be released for public use in 2021 by anInsilico Medicine spinoff company Deep Longevity.
Deep learning is closely related to a class of theories ofbrain development (specifically, neocortical development) proposed bycognitive neuroscientists in the early 1990s.[252][253][254][255] These developmental theories were instantiated in computational models, making them predecessors of deep learning systems. These developmental models share the property that various proposed learning dynamics in the brain (e.g., a wave ofnerve growth factor) support theself-organization somewhat analogous to the neural networks utilized in deep learning models. Like theneocortex, neural networks employ a hierarchy of layered filters in which each layer considers information from a prior layer (or the operating environment), and then passes its output (and possibly the original input), to other layers. This process yields a self-organizing stack oftransducers, well-tuned to their operating environment. A 1995 description stated, "...the infant's brain seems to organize itself under the influence of waves of so-called trophic-factors ... different regions of the brain become connected sequentially, with one layer of tissue maturing before another and so on until the whole brain is mature".[256]
A variety of approaches have been used to investigate the plausibility of deep learning models from a neurobiological perspective. On the one hand, several variants of thebackpropagation algorithm have been proposed in order to increase its processing realism.[257][258] Other researchers have argued that unsupervised forms of deep learning, such as those based on hierarchicalgenerative models anddeep belief networks, may be closer to biological reality.[259][260] In this respect, generative neural network models have been related to neurobiological evidence about sampling-based processing in the cerebral cortex.[261]
Although a systematic comparison between the human brain organization and the neuronal encoding in deep networks has not yet been established, several analogies have been reported. For example, the computations performed by deep learning units could be similar to those of actual neurons[262] and neural populations.[263] Similarly, the representations developed by deep learning models are similar to those measured in the primate visual system[264] both at the single-unit[265] and at the population[266] levels.
Google'sDeepMind Technologies developed a system capable of learning how to playAtari video games using only pixels as data input. In 2015 they demonstrated theirAlphaGo system, which learned the game ofGo well enough to beat a professional Go player.[268][269][270]Google Translate uses a neural network to translate between more than 100 languages.
In 2017, Covariant.ai was launched, which focuses on integrating deep learning into factories.[271]
As of 2008,[272] researchers atThe University of Texas at Austin (UT) developed a machine learning framework called Training an Agent Manually via Evaluative Reinforcement, or TAMER, which proposed new methods for robots or computer programs to learn how to perform tasks by interacting with a human instructor.[242] First developed as TAMER, a new algorithm called Deep TAMER was later introduced in 2018 during a collaboration betweenU.S. Army Research Laboratory (ARL) and UT researchers. Deep TAMER used deep learning to provide a robot with the ability to learn new tasks through observation.[242] Using Deep TAMER, a robot learned a task with a human trainer, watching video streams or observing a human perform a task in-person. The robot later practiced the task with the help of some coaching from the trainer, who provided feedback such as "good job" and "bad job".[273]
A main criticism concerns the lack of theory surrounding some methods.[274] Learning in the most common deep architectures is implemented using well-understood gradient descent. However, the theory surrounding other algorithms, such as contrastive divergence is less clear.[citation needed] (e.g., Does it converge? If so, how fast? What is it approximating?) Deep learning methods are often looked at as ablack box, with most confirmations done empirically, rather than theoretically.[275]
In further reference to the idea that artistic sensitivity might be inherent in relatively low levels of the cognitive hierarchy, a published series of graphic representations of the internal states of deep (20-30 layers) neural networks attempting to discern within essentially random data the images on which they were trained[276] demonstrate a visual appeal: the original research notice received well over 1,000 comments, and was the subject of what was for a time the most frequently accessed article onThe Guardian's[277] website.
Some deep learning architectures display problematic behaviors,[278] such as confidently classifying unrecognizable images as belonging to a familiar category of ordinary images (2014)[279] and misclassifying minuscule perturbations of correctly classified images (2013).[280]Goertzel hypothesized that these behaviors are due to limitations in their internal representations and that these limitations would inhibit integration into heterogeneous multi-componentartificial general intelligence (AGI) architectures.[278] These issues may possibly be addressed by deep learning architectures that internally form states homologous to image-grammar[281] decompositions of observed entities and events.[278]Learning a grammar (visual or linguistic) from training data would be equivalent to restricting the system tocommonsense reasoning that operates on concepts in terms of grammaticalproduction rules and is a basic goal of both human language acquisition[282] andartificial intelligence (AI).[283]
As deep learning moves from the lab into the world, research and experience show that artificial neural networks are vulnerable to hacks and deception.[284] By identifying patterns that these systems use to function, attackers can modify inputs to ANNs in such a way that the ANN finds a match that human observers would not recognize. For example, an attacker can make subtle changes to an image such that the ANN finds a match even though the image looks to a human nothing like the search target. Such manipulation is termed an "adversarial attack".[285]
In 2016 researchers used one ANN to doctor images in trial and error fashion, identify another's focal points, and thereby generate images that deceived it. The modified images looked no different to human eyes. Another group showed that printouts of doctored images then photographed successfully tricked an image classification system.[286] One defense is reverse image search, in which a possible fake image is submitted to a site such asTinEye that can then find other instances of it. A refinement is to search using only parts of the image, to identify images from which that piece may have been taken.[287]
Another group showed that certainpsychedelic spectacles could fool afacial recognition system into thinking ordinary people were celebrities, potentially allowing one person to impersonate another. In 2017 researchers added stickers tostop signs and caused an ANN to misclassify them.[286]
ANNs can however be further trained to detect attempts atdeception, potentially leading attackers and defenders into an arms race similar to the kind that already defines themalware defense industry. ANNs have been trained to defeat ANN-based anti-malware software by repeatedly attacking a defense with malware that was continually altered by agenetic algorithm until it tricked the anti-malware while retaining its ability to damage the target.[286]
In 2016, another group demonstrated that certain sounds could make theGoogle Now voice command system open a particular web address, and hypothesized that this could "serve as a stepping stone for further attacks (e.g., opening a web page hosting drive-by malware)".[286]
In "data poisoning", false data is continually smuggled into a machine learning system's training set to prevent it from achieving mastery.[286]
The deep learning systems that are trained using supervised learning often rely on data that is created and/or annotated by humans.[288] It has been argued that not only low-paidclickwork (such as onAmazon Mechanical Turk) is regularly deployed for this purpose, but also implicit forms of humanmicrowork that are often not recognized as such.[289] The philosopherRainer Mühlhoff distinguishes five types of "machinic capture" of human microwork to generate training data: (1)gamification (the embedding of annotation or computation tasks in the flow of a game), (2) "trapping and tracking" (e.g.CAPTCHAs for image recognition or click-tracking on Googlesearch results pages), (3) exploitation of social motivations (e.g.tagging faces onFacebook to obtain labeled facial images), (4)information mining (e.g. by leveragingquantified-self devices such asactivity trackers) and (5)clickwork.[289]
^Bengio, Yoshua; Lamblin, Pascal; Popovici, Dan; Larochelle, Hugo (2007).Greedy layer-wise training of deep networks(PDF). Advances in neural information processing systems. pp. 153–160.Archived(PDF) from the original on 2019-10-20. Retrieved2019-10-06.
^Rina Dechter (1986). Learning while searching in constraint-satisfaction problems. University of California, Computer Science Department, Cognitive Systems Laboratory.OnlineArchived 2016-04-19 at theWayback Machine
^abFukushima, K. (1969). "Visual feature extraction by a multilayered network of analog threshold elements".IEEE Transactions on Systems Science and Cybernetics.5 (4):322–333.doi:10.1109/TSSC.1969.300225.
^Sonoda, Sho; Murata, Noboru (2017). "Neural network with unbounded activation functions is universal approximator".Applied and Computational Harmonic Analysis.43 (2):233–268.arXiv:1505.03654.doi:10.1016/j.acha.2015.12.005.S2CID12149203.
^Nakano, Kaoru (1972). "Associatron-A Model of Associative Memory".IEEE Transactions on Systems, Man, and Cybernetics. SMC-2 (3):380–388.doi:10.1109/TSMC.1972.4309133.
^Turing, Alan (1948). "Intelligent Machinery".Unpublished (Later Published in Ince DC, Editor, Collected Works of AM Turing—Mechanical Intelligence, Elsevier Science Publishers, 1992).
^Amari, Shun'ichi (1967). "A theory of adaptive pattern classifier".IEEE Transactions.EC (16):279–307.
^Ramachandran, Prajit; Barret, Zoph; Quoc, V. Le (October 16, 2017). "Searching for Activation Functions".arXiv:1710.05941 [cs.NE].
^Fukushima, K. (1979). "Neural network model for a mechanism of pattern recognition unaffected by shift in position—Neocognitron".Trans. IECE (In Japanese). J62-A (10):658–665.doi:10.1007/bf00344251.PMID7370364.S2CID206775608.
^Fukushima, K. (1980). "Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position".Biol. Cybern.36 (4):193–202.doi:10.1007/bf00344251.PMID7370364.S2CID206775608.
^Linnainmaa, Seppo (1970).The representation of the cumulative rounding error of an algorithm as a Taylor expansion of the local rounding errors (Masters) (in Finnish). University of Helsinki. p. 6–7.
^Werbos, Paul J. (1994).The Roots of Backpropagation : From Ordered Derivatives to Neural Networks and Political Forecasting. New York: John Wiley & Sons.ISBN0-471-59897-6.
^Rumelhart, David E., Geoffrey E. Hinton, and R. J. Williams. "Learning Internal Representations by Error PropagationArchived 2022-10-13 at theWayback Machine". David E. Rumelhart, James L. McClelland, and the PDP research group. (editors), Parallel distributed processing: Explorations in the microstructure of cognition, Volume 1: Foundation. MIT Press, 1986.
^Gers, Felix; Schmidhuber, Jürgen; Cummins, Fred (1999). "Learning to forget: Continual prediction with LSTM".9th International Conference on Artificial Neural Networks: ICANN '99. Vol. 1999. pp. 850–855.doi:10.1049/cp:19991218.ISBN0-85296-721-7.
^abSchmidhuber, Jürgen (1991). "A possibility for implementing curiosity and boredom in model-building neural controllers".Proc. SAB'1991. MIT Press/Bradford Books. pp. 222–227.
^Baker, J.; Deng, Li; Glass, Jim; Khudanpur, S.; Lee, C.-H.; Morgan, N.; O'Shaughnessy, D. (2009). "Research Developments and Directions in Speech Recognition and Understanding, Part 1".IEEE Signal Processing Magazine.26 (3):75–80.Bibcode:2009ISPM...26...75B.doi:10.1109/msp.2009.932166.hdl:1721.1/51891.S2CID357467.
^Deng, L.; Hassanein, K.; Elmasry, M. (1994). "Analysis of correlation structure for a neural predictive model with applications to speech recognition".Neural Networks.7 (2):331–339.doi:10.1016/0893-6080(94)90027-2.
^Doddington, G.; Przybocki, M.; Martin, A.; Reynolds, D. (2000). "The NIST speaker recognition evaluation ± Overview, methodology, systems, results, perspective".Speech Communication.31 (2):225–254.doi:10.1016/S0167-6393(99)00080-1.
^abHeck, L.; Konig, Y.; Sonmez, M.; Weintraub, M. (2000). "Robustness to Telephone Handset Distortion in Speaker Recognition by Discriminative Feature Design".Speech Communication.31 (2):181–192.doi:10.1016/s0167-6393(99)00077-1.
^L.P Heck and R. Teunen. "Secure and Convenient Transactions with Nuance Verifier". Nuance Users Conference, April 1998.
^abGraves, Alex; Eck, Douglas; Beringer, Nicole; Schmidhuber, Jürgen (2003)."Biologically Plausible Speech Recognition with LSTM Neural Nets"(PDF).1st Intl. Workshop on Biologically Inspired Approaches to Advanced Information Technology, Bio-ADIT 2004, Lausanne, Switzerland. pp. 175–184.Archived(PDF) from the original on 2021-05-09. Retrieved2016-04-09.
^Graves, Alex; Fernández, Santiago; Gomez, Faustino;Schmidhuber, Jürgen (2006). "Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks".Proceedings of the International Conference on Machine Learning, ICML 2006:369–376.CiteSeerX10.1.1.75.6306.
^Graves, Alex; and Schmidhuber, Jürgen;Offline Handwriting Recognition with Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale; Lafferty, John; Williams, Chris K. I.; and Culotta, Aron (eds.),Advances in Neural Information Processing Systems 22 (NIPS'22), December 7th–10th, 2009, Vancouver, BC, Neural Information Processing Systems (NIPS) Foundation, 2009, pp. 545–552
^Sze, Vivienne; Chen, Yu-Hsin; Yang, Tien-Ju; Emer, Joel (2017). "Efficient Processing of Deep Neural Networks: A Tutorial and Survey".arXiv:1703.09039 [cs.CV].
^Ciresan, Dan; Giusti, Alessandro; Gambardella, Luca M.; Schmidhuber, Jürgen (2012). Pereira, F.; Burges, C. J. C.; Bottou, L.; Weinberger, K. Q. (eds.).Advances in Neural Information Processing Systems 25(PDF). Curran Associates, Inc. pp. 2843–2851.Archived(PDF) from the original on 2017-08-09. Retrieved2017-06-13.
^Ciresan, D.; Giusti, A.; Gambardella, L.M.; Schmidhuber, J. (2013). "Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks".Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013. Lecture Notes in Computer Science. Vol. 7908. pp. 411–418.doi:10.1007/978-3-642-40763-5_51.ISBN978-3-642-38708-1.PMID24579167.
^Ng, Andrew; Dean, Jeff (2012). "Building High-level Features Using Large Scale Unsupervised Learning".arXiv:1112.6209 [cs.LG].
^Simonyan, Karen; Andrew, Zisserman (2014). "Very Deep Convolution Networks for Large Scale Image Recognition".arXiv:1409.1556 [cs.CV].
^Vinyals, Oriol; Toshev, Alexander; Bengio, Samy; Erhan, Dumitru (2014). "Show and Tell: A Neural Image Caption Generator".arXiv:1411.4555 [cs.CV]..
^Fang, Hao; Gupta, Saurabh; Iandola, Forrest; Srivastava, Rupesh; Deng, Li; Dollár, Piotr; Gao, Jianfeng; He, Xiaodong; Mitchell, Margaret; Platt, John C; Lawrence Zitnick, C; Zweig, Geoffrey (2014). "From Captions to Visual Concepts and Back".arXiv:1411.4952 [cs.CV]..
^Kiros, Ryan; Salakhutdinov, Ruslan; Zemel, Richard S (2014). "Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models".arXiv:1411.2539 [cs.LG]..
^Simonyan, Karen; Zisserman, Andrew (2015-04-10),Very Deep Convolutional Networks for Large-Scale Image Recognition,arXiv:1409.1556
^He, Kaiming; Zhang, Xiangyu; Ren, Shaoqing; Sun, Jian (2016). "Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification".arXiv:1502.01852 [cs.CV].
^He, Kaiming; Zhang, Xiangyu; Ren, Shaoqing; Sun, Jian (10 Dec 2015).Deep Residual Learning for Image Recognition.arXiv:1512.03385.
^Gatys, Leon A.; Ecker, Alexander S.; Bethge, Matthias (26 August 2015). "A Neural Algorithm of Artistic Style".arXiv:1508.06576 [cs.CV].
^Goodfellow, Ian; Pouget-Abadie, Jean; Mirza, Mehdi; Xu, Bing; Warde-Farley, David; Ozair, Sherjil; Courville, Aaron; Bengio, Yoshua (2014).Generative Adversarial Networks(PDF). Proceedings of the International Conference on Neural Information Processing Systems (NIPS 2014). pp. 2672–2680.Archived(PDF) from the original on 22 November 2019. Retrieved20 August 2019.
^Karras, T.; Aila, T.; Laine, S.; Lehtinen, J. (26 February 2018). "Progressive Growing of GANs for Improved Quality, Stability, and Variation".arXiv:1710.10196 [cs.NE].
^Li, Xiangang; Wu, Xihong (2014). "Constructing Long Short-Term Memory based Deep Recurrent Neural Networks for Large Vocabulary Speech Recognition".arXiv:1410.4281 [cs.CL].
^Ferrie, C., & Kaiser, S. (2019).Neural Networks for Babies. Sourcebooks.ISBN978-1492671206.{{cite book}}: CS1 maint: multiple names: authors list (link)
^Silver, David; Huang, Aja; Maddison, Chris J.; Guez, Arthur; Sifre, Laurent; Driessche, George van den; Schrittwieser, Julian; Antonoglou, Ioannis; Panneershelvam, Veda (January 2016). "Mastering the game of Go with deep neural networks and tree search".Nature.529 (7587):484–489.Bibcode:2016Natur.529..484S.doi:10.1038/nature16961.ISSN1476-4687.PMID26819042.S2CID515925.
^Szegedy, Christian; Toshev, Alexander; Erhan, Dumitru (2013)."Deep neural networks for object detection".Advances in Neural Information Processing Systems:2553–2561.Archived from the original on 2017-06-29. Retrieved2017-06-13.
^Robinson, Tony (30 September 1991). "Several Improvements to a Recurrent Error Propagation Network Phone Recognition System".Cambridge University Engineering Department Technical Report. CUED/F-INFENG/TR82.doi:10.13140/RG.2.2.15418.90567.
^abSocher, Richard; Manning, Christopher."Deep Learning for NLP"(PDF).Archived(PDF) from the original on 6 July 2014. Retrieved26 October 2014.
^Socher, Richard; Bauer, John; Manning, Christopher; Ng, Andrew (2013)."Parsing With Compositional Vector Grammars"(PDF).Proceedings of the ACL 2013 Conference.Archived(PDF) from the original on 2014-11-27. Retrieved2014-09-03.
^Brocardo, Marcelo Luiz; Traore, Issa; Woungang, Isaac; Obaidat, Mohammad S. (2017). "Authorship verification using deep belief network systems".International Journal of Communication Systems.30 (12): e3259.doi:10.1002/dac.3259.S2CID40745740.
^abBoitet, Christian; Blanchon, Hervé; Seligman, Mark; Bellynck, Valérie (2010)."MT on and for the Web"(PDF). Archived fromthe original(PDF) on 29 March 2017. Retrieved1 December 2016.
^Wallach, Izhar; Dzamba, Michael; Heifets, Abraham (9 October 2015). "AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery".arXiv:1510.02855 [cs.LG].
^Gilmer, Justin; Schoenholz, Samuel S.; Riley, Patrick F.; Vinyals, Oriol; Dahl, George E. (2017-06-12). "Neural Message Passing for Quantum Chemistry".arXiv:1704.01212 [cs.LG].
^Tkachenko, Yegor (8 April 2015). "Autonomous CRM Control via CLV Approximation with Deep Reinforcement Learning in Discrete and Continuous Action Space".arXiv:1504.01840 [cs.LG].
^van den Oord, Aaron; Dieleman, Sander; Schrauwen, Benjamin (2013). Burges, C. J. C.; Bottou, L.; Welling, M.; Ghahramani, Z.; Weinberger, K. Q. (eds.).Advances in Neural Information Processing Systems 26(PDF). Curran Associates, Inc. pp. 2643–2651.Archived(PDF) from the original on 2017-05-16. Retrieved2017-06-14.
^Litjens, Geert; Kooi, Thijs; Bejnordi, Babak Ehteshami; Setio, Arnaud Arindra Adiyoso; Ciompi, Francesco; Ghafoorian, Mohsen; van der Laak, Jeroen A.W.M.; van Ginneken, Bram; Sánchez, Clara I. (December 2017). "A survey on deep learning in medical image analysis".Medical Image Analysis.42:60–88.arXiv:1702.05747.Bibcode:2017arXiv170205747L.doi:10.1016/j.media.2017.07.005.PMID28778026.S2CID2088679.
^De, Shaunak; Maity, Abhishek; Goel, Vritti; Shitole, Sanjay; Bhattacharya, Avik (2017). "Predicting the popularity of instagram posts for a lifestyle magazine using deep learning".2017 2nd International Conference on Communication Systems, Computing and IT Applications (CSCITA). pp. 174–177.doi:10.1109/CSCITA.2017.8066548.ISBN978-1-5090-4381-1.S2CID35350962.
^Shrager, J.; Johnson, MH (1996). "Dynamic plasticity influences the emergence of function in a simple cortical array".Neural Networks.9 (7):1119–1129.doi:10.1016/0893-6080(96)00033-0.PMID12662587.
^O'Reilly, Randall C. (1 July 1996). "Biologically Plausible Error-Driven Learning Using Local Activation Differences: The Generalized Recirculation Algorithm".Neural Computation.8 (5):895–938.doi:10.1162/neco.1996.8.5.895.ISSN0899-7667.S2CID2376781.
^Testolin, Alberto; Stoianov, Ivilin; Zorzi, Marco (September 2017). "Letter perception emerges from unsupervised deep learning and recycling of natural image features".Nature Human Behaviour.1 (9):657–664.doi:10.1038/s41562-017-0186-2.ISSN2397-3374.PMID31024135.S2CID24504018.
^Nguyen, Anh; Yosinski, Jason; Clune, Jeff (2014). "Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images".arXiv:1412.1897 [cs.CV].