Learning Informative and Private Representations via Generative Adversarial Networks

Yang, Tsung-Yen; Brinton, Christopher G.; Mittal, Prateek; Chiang, Mung; Lan, Andrew S.

doi:10.1109/bigdata.2018.8622089

Cited by 20 publications

(24 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, the privacy of medical images [58] and streetview images [129,141] have captured research interest in recent years. Furthermore, a number of GAN-based schemes have been developed for image steganography [87,115], image anonymization [62,68,122], and image encoding [27,94,98,147], which indeed can be exploited on any type of image besides face/medical/street-view images. Currently, the study of face images and medical images focuses on a single object, such as one face and one human organ, whereas the study of street-view images deals with multiple objects, including pedestrians, vehicles, buildings, and so on.…”

Section: Image Data Privacymentioning

confidence: 99%

“…In this adversarial framework, the generator is an encoding function that outputs limited information of private attributes while preserving non-private attributes, and the discriminators are neural network-based estimators for privacy protection. Similarly, other works [27,94,147] attempted to learn representations or dimension-reduced features from raw data based on DCGAN, in which the desired variables are maintained for utility representations, and the sensitive variables are hidden for privacy protection. Especially, these representations encoded from users' data can exhibit the predictive ability and protect privacy.…”

Section: Image Encodingmentioning

confidence: 99%

“…There is an increasing concern in computer vision devices invading users' privacy by recording unwanted videos [87,107,134,141]. On the one hand, some previous works that focus on image privacy can be applied to hide sensitive content in videos by simply considering a video as a sequence of image frames [62,147]. On the other hand, videos can also be used to recognize important events and assist humans' daily lives by advanced deep learning models, which differ from image recognition applications.…”

Section: Video Data Privacymentioning

confidence: 99%

See 2 more Smart Citations

Generative Adversarial Networks

et al. 2021

View full text Add to dashboard Cite

Generative Adversarial Networks (GANs) have promoted a variety of applications in computer vision and natural language processing, among others, due to its generative model’s compelling ability to generate realistic examples plausibly drawn from an existing distribution of samples. GAN not only provides impressive performance on data generation-based tasks but also stimulates fertilization for privacy and security oriented research because of its game theoretic optimization strategy. Unfortunately, there are no comprehensive surveys on GAN in privacy and security, which motivates this survey to summarize systematically. The existing works are classified into proper categories based on privacy and security functions, and this survey conducts a comprehensive analysis of their advantages and drawbacks. Considering that GAN in privacy and security is still at a very initial stage and has imposed unique challenges that are yet to be well addressed, this article also sheds light on some potential privacy and security applications with GAN and elaborates on some future research directions.

show abstract

Section: Image Data Privacymentioning

confidence: 99%

Section: Image Encodingmentioning

confidence: 99%

Section: Video Data Privacymentioning

confidence: 99%

See 1 more Smart Citation

Generative Adversarial Networks

et al. 2021

View full text Add to dashboard Cite

show abstract

“…These features are then used to train either standard machine learning models [13], [24], [25] or deep learning models [26], [27] to predict engagement or performance. However, e process of hand-crafting my discard valuable information within the clickstreams, such as their sequential patterns [20], [28]. We are therefore motivated to investigate modeling students' clicking behavior from their raw clickstreams, i.e., without any feature engineering, to maintain the primitive information.…”

Section: Introductionmentioning

confidence: 99%

“…In another study, [8] proposed a maximum likelihood estimation approach for CFA prediction based on visits to specific positions and transitions in lecture videos. Finally, [20] employed Generative Adversarial Networks (GAN) to predict CFA from hand-crafted features while simultaneously safeguarding sensitive attributes within the data. Compared to these approaches, we develop techniques that remove the hand-crafted feature pre-processing step, which we will show enhances CFA prediction performance.…”

Section: Introductionmentioning

confidence: 99%

Click-Based Student Performance Prediction: A Clustering Guided Meta-Learning Approach

Chu¹,

Tenorio²,

Cruz³

et al. 2021

Preprint

View full text Add to dashboard Cite

We study the problem of predicting student knowledge acquisition in online courses from clickstream behavior. Motivated by the proliferation of eLearning lecture delivery, we specifically focus on student in-video activity in lectures videos, which consist of content and in-video quizzes. Our methodology for predicting in-video quiz performance is based on three key ideas we develop. First, we model students' clicking behavior via time-series learning architectures operating on raw event data, rather than defining hand-crafted features as in existing approaches that may lose important information embedded within the click sequences. Second, we develop a self-supervised clickstream pre-training to learn informative representations of clickstream events that can initialize the prediction model effectively. Third, we propose a clustering guided meta-learning-based training that optimizes the prediction model to exploit clusters of frequent patterns in student clickstream sequences. Through experiments on three real-world datasets, we demonstrate that our method obtains substantial improvements over two baseline models in predicting students' in-video quiz performance. Further, we validate the importance of the pre-training and meta-learning components of our framework through ablation studies. Finally, we show how our methodology reveals insights on video-watching behavior associated with knowledge acquisition for useful learning analytics.

show abstract