-
Effective Dynamics of Generative Adversarial Networks
Authors:
Steven Durr,
Youssef Mroueh,
Yuhai Tu,
Shenshen Wang
Abstract:
Generative adversarial networks (GANs) are a class of machine-learning models that use adversarial training to generate new samples with the same (potentially very complex) statistics as the training samples. One major form of training failure, known as mode collapse, involves the generator failing to reproduce the full diversity of modes in the target probability distribution. Here, we present an…
▽ More
Generative adversarial networks (GANs) are a class of machine-learning models that use adversarial training to generate new samples with the same (potentially very complex) statistics as the training samples. One major form of training failure, known as mode collapse, involves the generator failing to reproduce the full diversity of modes in the target probability distribution. Here, we present an effective model of GAN training, which captures the learning dynamics by replacing the generator neural network with a collection of particles in the output space; particles are coupled by a universal kernel valid for certain wide neural networks and high-dimensional inputs. The generality of our simplified model allows us to study the conditions under which mode collapse occurs. Indeed, experiments which vary the effective kernel of the generator reveal a mode collapse transition, the shape of which can be related to the type of discriminator through the frequency principle. Further, we find that gradient regularizers of intermediate strengths can optimally yield convergence through critical dam** of the generator dynamics. Our effective GAN model thus provides an interpretable physical framework for understanding and improving adversarial training.
△ Less
Submitted 8 December, 2022;
originally announced December 2022.
-
What Makes a Message Persuasive? Identifying Adaptations Towards Persuasiveness in Nine Exploratory Case Studies
Authors:
Sebastian Duerr,
Krystian Teodor Lange,
Peter A. Gloor
Abstract:
The ability to persuade others is critical to professional and personal success. However, crafting persuasive messages is demanding and poses various challenges. We conducted nine exploratory case studies to identify adaptations that professional and non-professional writers make in written scenarios to increase their subjective persuasiveness. Furthermore, we identified challenges that those writ…
▽ More
The ability to persuade others is critical to professional and personal success. However, crafting persuasive messages is demanding and poses various challenges. We conducted nine exploratory case studies to identify adaptations that professional and non-professional writers make in written scenarios to increase their subjective persuasiveness. Furthermore, we identified challenges that those writers faced and identified strategies to resolve them with persuasive natural language generation, i.e., artificial intelligence. Our findings show that humans can achieve high degrees of persuasiveness (more so for professional-level writers), and artificial intelligence can complement them to achieve increased celerity and alignment in the process.
△ Less
Submitted 26 April, 2021;
originally announced April 2021.
-
Persuasive Natural Language Generation -- A Literature Review
Authors:
Sebastian Duerr,
Peter A. Gloor
Abstract:
This literature review focuses on the use of Natural Language Generation (NLG) to automatically detect and generate persuasive texts. Extending previous research on automatic identification of persuasion in text, we concentrate on generative aspects through conceptualizing determinants of persuasion in five business-focused categories: benevolence, linguistic appropriacy, logical argumentation, tr…
▽ More
This literature review focuses on the use of Natural Language Generation (NLG) to automatically detect and generate persuasive texts. Extending previous research on automatic identification of persuasion in text, we concentrate on generative aspects through conceptualizing determinants of persuasion in five business-focused categories: benevolence, linguistic appropriacy, logical argumentation, trustworthiness, tools and datasets. These allow NLG to increase an existing message's persuasiveness. Previous research illustrates key aspects in each of the above mentioned five categories. A research agenda to further study persuasive NLG is developed. The review includes analysis of seventy-seven articles, outlining the existing body of knowledge and showing the steady progress in this research field.
△ Less
Submitted 14 January, 2021;
originally announced January 2021.
-
Eurythmic Dancing with Plants -- Measuring Plant Response to Human Body Movement in an Anthroposophic Environment
Authors:
Sebastian Duerr,
Josephine van Delden,
Buenyamin Oezkaya,
Peter A. Gloor
Abstract:
This paper describes three experiments measuring interaction of humans with garden plants. In particular, body movement of a human conducting eurythmic dances near the plants (beetroots, tomatoes, lettuce) is correlated with the action potential measured by a plant SpikerBox, a device measuring the electrical activity of plants, and the leaf movement of the plant, tracked with a camera. The first…
▽ More
This paper describes three experiments measuring interaction of humans with garden plants. In particular, body movement of a human conducting eurythmic dances near the plants (beetroots, tomatoes, lettuce) is correlated with the action potential measured by a plant SpikerBox, a device measuring the electrical activity of plants, and the leaf movement of the plant, tracked with a camera. The first experiment shows that our measurement system captures external stimuli identically for different plants, validating the measurement system. The second experiment illustrates that the plants' response is correlated to the movements of the dancer. The third experiment indicates that plants that have been exposed for multiple weeks to eurythmic dancing might respond differently to plants which are exposed for the first time to eurythmic dancing.
△ Less
Submitted 23 December, 2020;
originally announced December 2020.
