The Experts below are selected from a list of 318 Experts worldwide ranked by ideXlab platform
Aaron Courville - One of the best experts on this subject based on the ideXlab platform.
-
improved training of wasserstein gans
Neural Information Processing Systems, 2017Co-Authors: Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron CourvilleAbstract:Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only poor samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to Undesired Behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models with continuous generators. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
-
improved training of wasserstein gans
arXiv: Learning, 2017Co-Authors: Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron CourvilleAbstract:Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to Undesired Behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
Mary Magee Quinn - One of the best experts on this subject based on the ideXlab platform.
-
Warning Signs of Problems in Schools: Ecological Perspectives and Effective Practices for Combating School Aggression and Violence.
Journal of School Violence, 2004Co-Authors: David Osher, Richard Vanacker, Gale M. Morrison, Robert A. Gable, Kevin P. Dwyer, Mary Magee QuinnAbstract:SUMMARY One need not look hard to find evidence of concern related to the nature of student Behavior in our schools. School violence, aggression, bullying, and harassment (e.g., racial or sexual) are often cited as challenging Behaviors confronting educators and community leaders. Unfortunately, most schools address these concerns with aversive consequences delivered to individual perpetrators in a hope of reducing the future probability of Undesired Behavior. A growing body of literature identifies the need to explore the social context of Behavior. The community, school, classroom, family, and peer group interact with student characteristics to help prevent, support the development of, and even exacerbate the display of both desired and Undesired Behavior. This article applies the logic of warning signs and functional Behavioral assessment to schools as it explores the social context of the school and the classroom. The school-wide and classroom-based factors that have been associated with or found to s...
Ishaan Gulrajani - One of the best experts on this subject based on the ideXlab platform.
-
improved training of wasserstein gans
Neural Information Processing Systems, 2017Co-Authors: Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron CourvilleAbstract:Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only poor samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to Undesired Behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models with continuous generators. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
-
improved training of wasserstein gans
arXiv: Learning, 2017Co-Authors: Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron CourvilleAbstract:Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to Undesired Behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
Liqiang Zhang - One of the best experts on this subject based on the ideXlab platform.
-
IoT-Praetor: Undesired Behaviors Detection for IoT Devices
IEEE Internet of Things Journal, 2021Co-Authors: Juan Wang, Shirong Hao, Ru Wen, Boxian Zhang, Liqiang ZhangAbstract:Due to insecure design and configuration, the Internet-of-Things (IoT) devices are vulnerable to various security issues. In most attacks against IoT, e.g., Mirai, attackers control devices to perform malicious Behaviors that are not expected by owners and administrators. Therefore, how to effectively detect malicious Behaviors is crucial to protect the security of IoT devices. Different from powerful PCs and servers, resource-constrained IoT devices are generally used to execute the specific function and their Behaviors are limited. Based on this observation, we propose IoT-Praetor, an Undesired Behavior security detection system for IoT devices. In IoT-Praetor, a new device usage description (DUD) model is proposed to construct an IoT device Behavior specification, including communication and interaction Behaviors. Furthermore, automatic Behavior extraction approaches are presented. We also design a Behavior rule engine to detect device Behaviors in real time. To evaluate the effectiveness of IoT-Praetor, we implemented our methods on Samsung SmartThings and performed a security test. The evaluation results show that the successful detection rate of malicious interaction Behavior is 94.5% on average, and the detection rate of malicious communication Behavior is above 98%, and system running time delay is only in millisecond level.
David Osher - One of the best experts on this subject based on the ideXlab platform.
-
Warning Signs of Problems in Schools: Ecological Perspectives and Effective Practices for Combating School Aggression and Violence.
Journal of School Violence, 2004Co-Authors: David Osher, Richard Vanacker, Gale M. Morrison, Robert A. Gable, Kevin P. Dwyer, Mary Magee QuinnAbstract:SUMMARY One need not look hard to find evidence of concern related to the nature of student Behavior in our schools. School violence, aggression, bullying, and harassment (e.g., racial or sexual) are often cited as challenging Behaviors confronting educators and community leaders. Unfortunately, most schools address these concerns with aversive consequences delivered to individual perpetrators in a hope of reducing the future probability of Undesired Behavior. A growing body of literature identifies the need to explore the social context of Behavior. The community, school, classroom, family, and peer group interact with student characteristics to help prevent, support the development of, and even exacerbate the display of both desired and Undesired Behavior. This article applies the logic of warning signs and functional Behavioral assessment to schools as it explores the social context of the school and the classroom. The school-wide and classroom-based factors that have been associated with or found to s...
