Generative Adversarial Network (GAN)

From
Jump to: navigation, search

YouTube ... Quora ...Google search ...Google News ...Bing News


Comprised of two nets, pitting one against the other (thus the “adversarial”). GANs’ potential is huge, because they can learn to mimic any distribution of data. That is, GANs can be taught to create worlds eerily similar to our own in any domain: images, music, speech, prose. Discriminative algorithms map features to labels. They are concerned solely with that correlation. One way to think about generative algorithms is that they do the opposite. Instead of predicting a label given certain features, they attempt to predict features given a certain label. Generative adversarial networks (GAN) are from a different breed of networks, they are twins: two networks working together. GANs consist of any two networks (although often a combination of FFs and CNNs), with one tasked to generate content and the other has to judge content. The discriminating network receives either training data or generated content from the generative network. How well the discriminating network was able to correctly predict the data source is then used as part of the error for the generating network. This creates a form of competition where the discriminator is getting better at distinguishing real data from generated data and the generator is learning to become less predictable to the discriminator. This works well in part because even quite complex noise-like patterns are eventually predictable but generated content similar in features to the input data is harder to learn to distinguish. GANs can be quite difficult to train, as you don’t just have to train two networks (either of which can pose it’s own problems) but their dynamics need to be balanced as well. If prediction or generation becomes to good compared to the other, a GAN won’t converge as there is intrinsic divergence. Goodfellow, Ian, et al. “Generative adversarial nets.” Advances in Neural Information Processing Systems. 2014.

An example of self-supervised learning is generative adversarial networks, or GANs. These are generative models that are most commonly used for creating synthetic photographs using only a collection of unlabeled examples from the target domain. GAN models are trained indirectly via a separate discriminator model that classifies examples of photos from the domain as real or fake (generated), the result of which is fed back to update the GAN model and encourage it to generate more realistic photos on the next iteration. 14 Different Types of Learning in Machine Learning | Jason Brownlee - Machine Learning Mastery

Generator and Discriminator

Generator: The generator is a neural network that takes random noise as input and generates synthetic data samples, such as images, text, or audio. Its goal is to produce samples that are indistinguishable from real data samples. The generator is initialized with random weights that define an initial probability distribution over the output space (e.g., images). This initial distribution is essentially random noise, as the generator has not yet learned to map the input noise vectors to meaningful outputs.

Discriminator: The discriminator is another neural network that acts as a binary classifier. It takes both real data samples and synthetic samples generated by the generator as input, and its task is to distinguish between the two. The discriminator is trained to output a high probability for real data samples and a low probability for fake (generated) samples. The discriminator is initialized with random weights that define an initial probability distribution over its binary classification task (real vs. fake). At the start, the discriminator's outputs are essentially random guesses, as it has not yet learned to distinguish real data from the generator's outputs.

The generator and discriminator are trained in an adversarial manner:

  • The discriminator is trained to accurately classify real and fake samples.
  • The generator is trained to produce samples that can fool the discriminator into classifying them as real.

This adversarial training process continues iteratively, with the generator and discriminator improving against each other, until the generator produces samples that are indistinguishable from real data to the discriminator.

Game Theory for Training Language Models

Game theory provides a powerful framework for training language models to be more accurate and consistent. The consensus game pits the generative and discriminative modes of a large language model against each other in a game-theoretic setup. The generator proposes open-ended answers, while the discriminator evaluates and chooses between options. By incentivizing agreement between these two modes, the model is driven to find responses that satisfy both its generative and discriminative knowledge.

Crucially, the discriminator is initialized with a different prior probability distribution over outputs, reflecting its pre-trained knowledge from data. As the game progresses, the discriminator gets rewarded not just for reaching consensus with the generator, but also for not deviating too far from its initial beliefs grounded in real-world data. This encourages the model to incorporate factual knowledge into the agreed-upon responses, improving overall accuracy.

Without such a mechanism, models like large language models could theoretically converge on completely incorrect answers that satisfy their internal consistency, but contradict reality. The consensus game acts as a regulator, pushing the model away from such degenerate solutions. By leveraging the different inductive biases of the generative and discriminative modes, and using game-theoretic rewards, the consensus game allows language models to become more correct and efficient at mapping inputs to truthful outputs.

GAN Lab

Play with Generative Adversarial Networks (GAN) in your browser

figure-animated-ring.gif

Computer Fraud and Abuse Act (CFAA)

Is the Adversarial ML Researcher violating the CFAA when attacking an ML system? Depending on the nature of the adversarial ML attack, and which US State the lawsuit is brought, the answer varies. Legal Risks of Adversarial Machine Learning Research | Ram Shankar Siva Kumar - Medium

Adversarial Machine Learning is booming with ML researchers increasingly targeting commercial ML systems such as those used in Facebook, Tesla, Microsoft, IBM, Google to demonstrate vulnerabilities. In this paper, we ask, "What are the potential legal risks to adversarial ML researchers when they attack ML systems?" Studying or testing the security of any operational system potentially runs afoul the Computer Fraud and Abuse Act (CFAA), the primary United States federal statute that creates liability for hacking. We claim that Adversarial ML research is likely no different. Our analysis show that because there is a split in how CFAA is interpreted, aspects of adversarial ML attacks, such as model inversion, membership inference, model stealing, reprogramming the ML system and poisoning attacks, may be sanctioned in some jurisdictions and not penalized in others. We conclude with an analysis predicting how the US Supreme Court may resolve some present inconsistencies in the CFAA's application in Van Buren v. United States, an appeal expected to be decided in 2021. We argue that the court is likely to adopt a narrow construction of the CFAA, and that this will actually lead to better adversarial ML security outcomes in the long term. Legal Risks of Adversarial Machine Learning Research | R. Shankar, S. Kumar, J. Penney, B. Schneier, and K. Albert - arXiv.org


Retrieved from "https://primo.ai/index.php?title=Generative_Adversarial_Network_(GAN)&oldid=37245"