Generative adversarial network

 A generative adversarial network (GAN) is a class of machine learning frameworks designed by Ian Goodfellow and his colleagues in 2014. Two neural networks contest with each other in a game (in the form of a zero-sum game, where one agent's gain is another agent's loss.

Given a training set, this technique learns to generate new data with the same statistics as the training set. For example, a GAN trained on photographs can generate new photographs that look at least superficially authentic to human observers, having many realistic characteristics. Though originally proposed as a form of generative model for unsupervised learning, GANs have also proven useful for semi-supervised learning, fully supervised learning, and reinforcement learning.

The core idea of a GAN is based on the "indirect" training through the discriminator, which itself is also being updated dynamically. This basically means that the generator is not trained to minimize the distance to a specific image, but rather to fool the discriminator. This enables the model to learn in an unsupervised manner.

Method

The generative network generates candidates while the discriminative network evaluates them. The contest
operates in terms of data distributions. Typically, the generative network learns to map from a latent space to a data distribution of interest, while the discriminative network distinguishes candidates produced by the generator from the true data distribution. The generative network's training objective is to increase the error rate of the discriminative network (i.e., "fool" the discriminator network by producing novel candidates that the discriminator thinks are not synthesized (are part of the true data distribution)).

A known dataset serves as the initial training data for the discriminator. Training it involves presenting it with samples from the training dataset, until it achieves acceptable accuracy. The generator trains based on whether it succeeds in fooling the discriminator. Typically the generator is seeded with randomized input that is sampled from a predefined latent space (e.g. a multivariate normal distribution). Thereafter, candidates synthesized by the generator are evaluated by the discriminator. Independent backpropagation procedures are applied to both networks so that the generator produces better images, while the discriminator becomes more skilled at flagging synthetic images. The generator is typically a deconvolutional neural network, and the discriminator is a convolutional neural network.

GANs often suffer from a "mode collapse" where they fail to generalize properly, missing entire modes from the input data. For example, a GAN trained on the MNIST dataset containing many samples of each digit, might nevertheless timidly omit a subset of the digits from its output. Some researchers perceive the root problem to be a weak discriminative network that fails to notice the pattern of omission, while others assign blame to a bad choice of objective function. Many solutions have been proposed.

Applications

GAN applications have increased rapidly.

Fashion, art and advertising

GANs can be used to generate art; The Verge wrote in March 2019 that "The images created by GANs have become the defining look of contemporary AI art." GANs can also be used to inpaint photographs or create photos of imaginary fashion models, with no need to hire a model, photographer or makeup artist, or pay for a studio and transportation.

Science

GANs can improve astronomical images and simulate gravitational lensing for dark matter research. They

were used in 2019 to successfully model the distribution of dark matter in a particular direction in space and to predict the gravitational lensing that will occur.

GANs have been proposed as a fast and accurate way of modeling high energy jet formation and modeling showers through calorimeters of high-energy physics experiments. GANs have also been trained to accurately approximate bottlenecks in computationally expensive simulations of particle physics experiments. Applications in the context of present and proposed CERN experiments have demonstrated the potential of these methods for accelerating simulation and/or improving simulation fidelity.

Video games

In 2018, GANs reached the video game modding community, as a method of up-scaling low-resolution 2D textures in old video games by recreating them in 4k or higher resolutions via image training, and then down-sampling them to fit the game's native resolution (with results resembling the supersampling method of anti-aliasing). With proper training, GANs provide a clearer and sharper 2D texture
image magnitudes higher in quality than the original, while fully retaining the original's level of details, colors, etc. Known examples of extensive GAN usage include Final Fantasy VIII, Final Fantasy IX, Resident Evil REmake HD Remaster, and Max Payne.

Concerns about malicious applications

Concerns have been raised about the potential use of GAN-based human image synthesis for sinister purposes, e.g., to produce fake, possibly incriminating, photographs and videos. GANs can be used to generate unique, realistic profile photos of people who do not exist, in order to automate creation of fake social media profiles.

In 2019 the state of California considered and passed on October 3, 2019 the bill AB-602, which bans the use of human image synthesis technologies to make fake pornography without the consent of the people depicted, and bill AB-73, which prohibits distribution of manipulated videos of a political candidate within 60 days of an election. Both bills were authored by Assembly member Marc Berman and signed by Governor Gavin Newsom. The laws will come into effect in 2020.

DARPA's Media Forensics program studies ways to counteract fake media, including fake media produced using GANs

Miscellaneous applications

GAN can be used to detect glaucomatous images helping the early diagnosis which is essential to avoid partial or total loss of vision.

GANs that produce photorealistic images can be used to visualize interior design, industrial design, shoes, bags, and clothing items or items for computer games' scenes. Such networks were reported to be used by Facebook.

GANs can reconstruct 3D models of objects from images, and model patterns of motion in video.

GANs can be used to age face photographs to show how an individual's appearance might change with age.

GANs can also be used to transfer map styles in cartography or augment street view imagery.

Relevance feedback on GANs can be used to generate images and replace image search systems.

A variation of the GANs is used in training a network to generate optimal control inputs to nonlinear dynamical systems. Where the discriminatory network is known as a critic that checks the optimality of the solution and the generative network is known as an Adaptive network that generates the optimal control. The critic and adaptive network train each other to approximate a nonlinear optimal control.

GANs have been used to visualize the effect that climate change will have on specific houses.

A GAN model called Speech2Face can reconstruct an image of a person's face after listening to their voice.

In 2016 GANs were used to generate new molecules for a variety of protein targets implicated in cancer, inflammation, and fibrosis. In 2019 GAN-generated molecules were validated experimentally all the way into mice

History

The most direct inspiration for GANs was noise-contrastive estimation, which uses the same loss function as GANs and which Goodfellow studied during his PhD in 2010–2014.

Other people had similar ideas but did not develop them similarly. An idea involving adversarial networks was published in a 2010 blog post by Olli Niemitalo. This idea was never implemented and did not involve stochasticity in the generator and thus was not a generative model. It is now known as a conditional GAN or cGAN. An idea similar to GANs was used to model animal behavior by Li, Gauci and Gross in 2013.

Adversarial machine learning has other uses besides generative modeling and can be applied to models other than neural networks. In control theory, adversarial learning based on neural networks
was used in 2006 to train robust controllers in a game theoretic sense, by alternating the iterations between a minimizer policy, the controller, and a maximizer policy, the disturbance.

In 2017, a GAN was used for image enhancement focusing on realistic textures rather than pixel-accuracy, producing a higher image quality at high magnification. In 2017, the first faces were generated. These were exhibited in February 2018 at the Grand Palais. Faces generated by StyleGAN in 2019 drew comparisons with deepfakes.

Beginning in 2017, GAN technology began to make its presence felt in the fine arts arena with the appearance of a newly developed implementation which was said to have crossed the threshold of being able to generate unique and appealing abstract paintings, and thus dubbed a "CAN", for "creative adversarial network". A GAN system was used to create the 2018 painting Edmond de Belamy, which sold for US$432,500. An early 2019 article by members of the original CAN team discussed further progress with that system, and gave consideration as well to the overall prospects for an AI-enabled art.

In May 2019, researchers at Samsung demonstrated a GAN-based system that produces videos of a person speaking, given only a single photo of that person.

In August 2019, a large dataset consisting of 12,197 MIDI songs each with paired lyrics and melody alignment was created for neural melody generation from lyrics using conditional GAN-LSTM (refer to sources at GitHub AI Melody Generation from Lyrics).

In May 2020, Nvidia researchers taught an AI system (termed "GameGAN") to recreate the game of Pac-Man simply by watching it being played.

Classification

Bidirectional GAN

While the standard GAN model learns a mapping from a latent space to the data distribution, inverse models such as Bidirectional GAN (BiGAN)  and Adversarial Autoencoders  also
learn a mapping from data to the latent space. This inverse mapping allows real or generated data examples to be projected back into the latent space, similar to the encoder of a variational autoencoder. Applications of bidirectional models include semi-supervised learning,
 interpretable machine learning, and neural machine translation

COVID-19 Pandemic ???

The COVID-19 pandemic, also known as the coronavirus pandemic, is an
ongoing pandemic of coronavirus disease 2019 (COVID-19) caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The disease was first identified in December 2019 in Wuhan, China, became a Public Health Emergency of International Concern in January 2020, and subsequently recognised as a pandemic in March 2020. As of 2 October 2020, more than 34.3 million cases have been reported worldwide, although the true number of cases are likely to be more higher. A more reliable indicator for case spread is the more than 1.02 million deaths attributed to COVID-19. Many recoveries from confirmed infections go unreported, but at least 23,711,591 people have recovered from confirmed infections.

The disease spreads between people most often when they are physically close. It spreads very easily and sustainably through the air, primarily via small droplets or particles such as aerosols, produced after an infected person breathes, coughs, sneezes, talks or sings. It may also be transmitted via contaminated surfaces, although this has not been conclusively demonstrated. It can spread for up to two days prior to symptom onset and from people who are asymptomatic. People remain infectious for 7–12 days in moderate cases and up to two weeks in severe cases. 

Longer-term damage to organs (in particular lungs and heart) has been observed, and there is concern about a significant number of patients who have recovered from the acute phase of the disease but continue to experience a range of effects including severe fatigue, memory loss and other cognitive issues, low grade fever, muscle weakness, breathlessness and other symptoms for months afterwards

Common symptoms include fever, cough, fatigue, shortness of breath or breathing difficulties, and loss of smell.

Complications may include pneumonia and acute respiratory distress syndrome. The incubation period is typically around five days but may range from one to 14 days. There are several vaccine candidates in development, although none have completed clinical trials to prove their safety and efficacy. There is no known specific antiviral medication, so primary treatment is currently symptomatic.

Recommended preventive measures include hand washing, covering mouth when sneezing or

coughing, social distancing, wearing a face mask in public, disinfecting surfaces, ventilating and air-filtering, and monitoring and self-isolation for people who suspect they may be infected. Authorities worldwide have responded by implementing travel restrictions, lockdowns, workplace hazard controls, and facility closures to slow the spread of the disease. Many places have also worked to increase testing capacity and trace contacts of the infected.

The pandemic has caused global social and economic disruption, including the largest global recession since the Great Depression. According to estimations, up to

100 million people have fallen into extreme poverty and global famines are affecting 130 million people. It has led to the postponement or cancellation of sporting, religious, political, and cultural events, widespread supply shortages exacerbated by panic buying, and decreased emissions of pollutants and greenhouse gases. Educational institutions have been partially or fully closed, with many switching to online schooling. Later in the pandemic certain countries reopened schools, usually with heavy restrictions and higher funding. Misinformation about the virus has circulated through social media and mass media. There have been many incidents of xenophobia and racism against Chinese people and against those perceived as being Chinese or as being from areas with high infection rates. The general public often call "coronavirus" both the virus and the disease it causes. U.S. President Donald Trump referred to the virus as the "Chinese virus" in tweets, interviews, and White House press briefings, which drew some criticism that he was stigmatizing the disease with racial or nationalistic overtones.

Virology

Infection and transmission

Human-to-human transmission of SARS-CoV-2 was confirmed on 20 January 2020, during the COVID-19 pandemic. Transmission was initially assumed to occur primarily via respiratory droplets from coughs

and sneezes within a range of about 1.8 metres (6 ft). Laser light scattering experiments suggest speaking as an additional mode of transmission. Other studies have suggested that the virus may be airborne as well, with aerosols potentially being able to transmit the virus.

Indirect contact via contaminated surfaces is another possible cause of infection. Preliminary research indicates that the virus may remain viable on

plastic (polypropylene) and stainless steel (AISI 304) for up to three days, but does not survive on cardboard for more than one day or on copper for more than four hours; the virus is inactivated by soap, which destabilises its lipid bilayer. Viral RNA has also been found in stool samples and semen from infected individuals.

Prevention and treatment

There are no vaccines or antiviral drugs to prevent or treat human coronavirus infections. Treatment is only supportive. A number of antiviral targets have been identified such as viral proteases, polymerases, and entry proteins. Drugs are in development which target these proteins and the different steps of viral replication. A number of vaccines using different methods are also under development for different human coronaviruses.

There are no antiviral drugs to treat animal coronaviruses. Vaccines are available for IBV, TGEV, and Canine CoV, although their effectiveness is limited. In the case of outbreaks of highly contagious animal coronaviruses, such as PEDV, measures such as destruction of entire herds of pigs may be used to prevent transmission to other herds.

Swine influenza and respiratory disease

Swine influenza (also known as swine flu or pig flu) is a respiratory disease that occurs in pigs that is caused by the Influenza A virus. Influenza viruses that are normally found in swine are known as swine influenza viruses (SIVs). The known SIV strains include influenza C and the subtypes of influenza A known as H1N1, H1N2, H3N1, H3N2 and H2N3. Pigs can also become infected with the H4N6, H9N2 subtypes and ASF (African Swine Fever)

Swine influenza virus is common throughout pig populations worldwide. Transmission of the virus from pigs to humans is not common and does not always lead to human influenza, often resulting only in the production of antibodies in the blood. If transmission does cause human influenza, it is called zoonotic swine flu or a variant virus. People with regular exposure to pigs are at increased risk of swine flu infection. Properly cooking the meat of an infected animal removes the risk of infection.

Pigs experimentally infected with the strain of swine flu that caused the human pandemic of 2009–10 showed clinical signs of flu within four days, and the virus spread to other uninfected pigs housed with the infected ones.

With around 1 billion individuals alive at any time, the domestic pig is one of the most numerous large mammals on the planet.

Feral pigs like other introduced mammals are major drivers of extinction and ecosystem change.

They have been introduced into many parts of the world, and will damage crops and home gardens as well as potentially spreading disease. They uproot large areas of land, eliminating native vegetation and spreading weeds. This results in habitat alteration, a change in plant succession and composition and a decrease in native fauna dependent on the original habitat.

Pigs and food safety

The pandemic virus is a type of swine influenza, derived originally from a strain which lived in pigs, and this

origin gave rise to the common name of "swine flu". This term is widely used by mass media, though the Paris-based World Organisation for Animal Health as well as industry groups such as the U.S. National Pork Board, the American Meat Institute, and the Canadian Pork Council objected to widespread media use of the name "swine flu" and suggested it should be called "North American flu" instead, while the World Health Organization switched its designation from "swine influenza" to "influenza A (H1N1)" in late April 2009. The virus has been found in U.S. hogs, and Canadian as well as in hogs in Northern Ireland, Argentina, and Norway. Leading health agencies and the United States Secretary of Agriculture have stressed that eating properly cooked pork or other food products derived from pigs will not cause flu.

 

Deforestation

Kate Jones, chair of ecology and biodiversity at University College London, says zoonotic diseases are Increasingly linked to environmental change and human behaviour. The disruption of pristine

forests driven by logging, mining, road building through remote places, rapid urbanisation and population growth is bringing people into closer contact with animal species they may never have been near before. The resulting transmission of disease from wildlife to humans, she says, is now “a hidden cost of human economic development". In a guest article published by IPBES, Peter Daszak and three co-chairs of the 2019 Global Assessment Report on Biodiversity and Ecosystem Services, Josef Settele, Sandra Díaz and Eduardo Brondizio, write that "rampant deforestation, uncontrolled expansion of agriculture, intensive farming, mining and infrastructure development, as well as the exploitation of wild species have created a ‘perfect storm’ for the spillover of diseases from wildlife to people. Deforestation, wildlife farming and trade in unsanitary conditions increases the risk of new zoonotic diseases, biodiversity experts have warned.

Misinformation related to the COVID-19 pandemic

The COVID-19 pandemic has resulted in misinformation and conspiracy theories about the scale of the pandemic and the origin, prevention, diagnosis, and treatment of the disease. False information, including intentional disinformation, has been spread through social media, text messaging, and mass media, including the tabloid media, conservative media, and state media of countries such as China, Russia, Iran, and Turkmenistan. It has also been reportedly spread by covert operations backed by states such as Saudi Arabia, Russia and China to generate panic and sow distrust in other countries. In some countries, such as India, Bangladesh, and Ethiopia, journalists have been arrested for allegedly spreading fake news about the pandemic.

Misinformation has been propagated by celebrities, politicians (including heads of state in countries such as the United States, Iran, and Brazil), and other prominent public figures. Commercial scams have claimed to offer at-home tests, supposed preventives, and "miracle" cures. Several religious groups have claimed their faith will protect them from the virus. Some people have claimed the virus is a bioweapon 

accidentally or purposefully leaked from a laboratory, a population control scheme, the result of a spy operation, or the side effect of 5G upgrades to cellular networks.

The World Health Organization has declared an "infodemic" of incorrect information about the virus, which poses risks to global health.