Theoretical Insights into Memorization in GANs

By Vaishnavh Nagarajan. 2019

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning (2022, background): “With respect to the generative model, this training process is much more computationally and memory intensive compared to a classification model and additionally may violate data legality concerns because using a generative model increases the chance of memorizing potentially sensitive data [45].”
Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning (2021, background): “Additionally, it is not clear whether generating images from the data distribution will violate data legality concerns because using a generative model increases the chance of memorizing potentially sensitive data [42].”
Hausdorff GAN: Improving GAN Generation Quality With Hausdorff Metric (2021, background): “That being said, these solutions may occur only from certain hyperparameters’ combinations [4] or from memorization of training samples [35].”
A Closer Look at Rehearsal-Free Continual Learning * (2022, background): “With respect to the generative model, this training process is much more computationally and memory intensive compared to a classification model and additionally may violate data legality concerns because using a generative model increases the chance of memorizing potentially sensitive data [41].”
R-DFCIL: Relation-Guided Representation Learning for Data-Free Class Incremental Learning (2022, background): “These approaches usually perform poorly [27] due to the domain gap between generated and real data, and they also cause data privacy concerns because the generator may remember sensitive information in the real data [16].”
A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications (2020, background): “[315] argue that making the generator “learn to memorize” the training data is a more difficult task than making it “learn to output realistic but unseen data”.”
Extracting Training Data from Diffusion Models (2023, background): “Moreover, other work provides insights into when and why GANs may replicate training examples [50, 26], as well as how to mitigate such effects [50].”