New error bounds for GANs with nonlinear objective functions derived.
problem Statistical consistency of GANs with nonlinear objective functions.
method Derivation of statistical error bounds for (f,Γ)-GANs using Rademacher complexity. result Proves the statistical consistency of (f,Γ)-GANs. CapsuleGAN uses capsule networks in GANs for better image data modeling.
problem Improving image data modeling in GANs.
method CapsuleGAN uses capsule networks as discriminators in GANs, with a new objective function incorporating margin loss.
result CapsuleGAN outperforms standard GANs on image data modeling and semi-supervised classification.
The paper explores how Lipschitz-continuity improves GAN training stability and quality.
problem Failure and instability in GAN training due to unreliable gradient from optimal discriminative function.
method Investigates the property of optimal discriminative function and proves Lipschitz-continuity is a solution.
result Lipschitz-continuity condition ensures convergence and leads to more stable and higher quality generated samples.
Dual-objective GANs reduce training instabilities with tunable α-loss parameters.
problem Training instabilities in Generative Adversarial Networks (GANs).
method Introduce (αD,αG)-GANs with dual objectives modeled using α-loss. result Upper bounds on estimation error show improved performance under certain conditions.
E-GAN improves GANs by evolving a population of generators.
problem Training instability and mode collapse in GANs.
method E-GAN uses mutation operations as different adversarial training objectives to evolve a population of generators.
result E-GAN achieves better generative performance and reduces training problems.
A new GAN training method using primal-dual subgradient methods.
problem Training GANs to avoid mode collapse and generate diverse samples.
method Relating GANs to convex optimization via Lagrangian perspective and primal-dual subgradient methods.
result The proposed method resolves mode collapse and generates diverse samples.
Paper analyzes GANs training difficulties and proposes a control framework.
problem Difficulties in training GANs, especially for financial time series.
method Stochastic control framework for hyper-parameters tuning.
result Explicit forms for optimal adaptive learning rate and batch size derived.
GANs can generate realistic data without minimizing a divergence, contrary to current theory.
problem Current theory suggests GANs minimize a divergence to generate realistic data.
method Discussed various loss functions for G, showing they are not divergences and do not have the same equilibrium.
result GANs can use a wide range of loss functions, not just divergences, to generate realistic data.
Stackelberg GAN improves GAN stability by reducing minimax gap.
problem Stability issues in GAN training procedure.
method New multi-generator architecture and application of Shapley-Folkman lemma.
result Minimax gap shrinks to ε with rate O(1/ε) as the number of generators increases.
GANs may not have Nash equilibria, but proximal training can find solutions.
problem Existence of Nash equilibria in GANs optimization.
method Proximal training approach to find solutions.
result Proximal training finds solutions to GAN problems.
Paper bridges f-GANs and WGANs for better image generation.
problem Learning high-dimensional distributions using GANs.
method List constraints, minimize Lagrangian relaxation, propose KL-Wasserstein GAN.
result Empirical success on synthetic and real-world image generation benchmarks.
Improved GAN training stability through tunable classification losses.
problem Training instabilities in GANs.
method Reformulated GAN value function using class probability estimation (CPE) losses, defined (αD,αG)-GANs. result Tuning (αD,αG) can alleviate training instabilities. Survey of GANs challenges and solutions for better model design and optimization.
problem Challenges in training GANs, including mode collapse, non-convergence, and instability.
method Comprehensive survey of GANs design and optimization solutions, proposing a new taxonomy.
result Presentation of promising research directions in GANs.
Paper introduces a new method to improve GANs by leveraging the discriminator's energy.
problem Improving the quality of generated samples in GANs.
method Discriminator Contrastive Divergence (DCD) method.
result Significant improvement in generation quality on synthetic and real-world data.
Generative Adversarial Networks are adapted for point cloud generation.
problem Applying GAN to point clouds is challenging due to data structure differences.
method Proposed PC-GAN framework with hierarchical sampling, posterior inference, and sandwiching objective.
result PC-GAN trained with sandwiching objective outperforms existing methods on point cloud generation.
Generalized dual discriminator GANs improve upon traditional GANs by using two discriminators and a flexible loss function.
problem Mode collapse in GANs.
method Introducing dual discriminator α-GANs and extending the approach to arbitrary functions. result The approach reduces the optimization problem to a linear combination of an f-divergence and a reverse f-divergence. GANs struggle with discontinuous distributions and object counting in images.
problem GANs' limitations in learning from complex distributions and counting objects.
method Evaluated GANs on synthetic datasets including discontinuous and noisy points, and images with varying polygons.
result GANs fail to accurately recreate discontinuous distributions and count objects in images.
Paper analyzes game dynamics with negative momentum for improved stability and convergence.
problem Complexity and instability in game dynamics, especially in adversarial settings.
method Analyzed gradient-based methods with negative momentum on simple games and adversarial problems.
result Alternating gradient updates with negative momentum achieve convergence in difficult adversarial problems.
New GAN loss functions improve image generation quality and stability.
problem Improving the performance of GANs in generating high-quality images.
method Introducing least kth-order GAN (LkGAN) and Rényi-centric GAN loss functions. result The proposed loss functions lead to better image quality and stability.
LSGANs improve GANs by using least squares loss, leading to better image quality and stability.
problem Vanishing gradients in GANs during training.
method Introducing LSGANs with least squares loss for both discriminator and generator.
result LSGANs generate higher quality images and are more stable during training.
We present a framework to understand GAN training as alternating density ratio estimation and approximate divergence minimization. This provides an interpretation for the mismatched GAN generator and discriminator objectives often used in practice, and explains the problem of poor sample diversity. We also derive a fam…
This paper compares two GAN architectures to reduce mode collapse.
problem Reducing mode collapse in Generative Adversarial Networks (GANs).
method Explains and compares PacGAN and VEEGAN models.
result PacGAN performs slightly better than vanilla GAN in terms of mode collapse.
A new GAN variant avoids mode collapse and generates high-quality images.
problem Mode collapse in GANs.
method Mixture Density GAN uses Gaussian mixtures to encourage the discriminator to form clusters, leading the generator to discover multiple data modes.
result Mixture Density GAN generates images with high quality and avoids mode collapse, as shown by FID metrics.
New research shows encoder-decoder GANs can still fail even on real data.
problem Theoretical limitations of Encoder-Decoder GAN architectures.
method Rigorous analysis of Encoder-Decoder GAN training objectives.
result The training objectives cannot prevent mode collapse or learning meaningless codes.
The paper proposes a method to learn 3D object pose manifolds using GANs and elasticae.
problem Learning image manifolds of 3D objects with limited data.
method Geom-SGAN and elasticae for geometry-preserving image interpolation.
result The method outperforms state-of-the-art GANs and VAEs in learning rotation paths.
Modified GAN-CLS generates more plausible images from text descriptions.
problem Inference issues with GAN-CLS algorithm.
method Modified GAN-CLS algorithm with a corrected objective function.
result Generated images are more plausible and match input texts better.
Analyzes GAN units for better understanding and improvement.
problem Lack of understanding GAN internal representations and artifacts.
method Interpretable units identification, causal effect quantification, contextual relationship examination.
result Visualizes and understands GANs at various levels, enabling new insights and improvements.
Generative adversarial networks (GANs) have received a tremendous amount of attention in the past few years, and have inspired applications addressing a wide range of problems. Despite its great potential, GANs are difficult to train. Recently, a series of papers (Arjovsky & Bottou, 2017a; Arjovsky et al. 2017b; and Gu…
Generative adversarial networks (GANs) are successful deep generative models. GANs are based on a two-player minimax game. However, the objective function derived in the original motivation is changed to obtain stronger gradients when learning the generator. We propose a novel algorithm that repeats the density ratio e…
Introduces PGAN, a GAN variant with a probabilistic model.
problem GAN instability and lack of quality measure.
method Integrates a Gaussian Mixture Model into GAN framework with a likelihood-based loss function.
result PGAN generates realistic images and correlates likelihood with image quality.
Wasserstein GANs use different p-metrics to improve model performance.
problem Improving stability and performance of Wasserstein GANs.
method Introduce (q,p)-Wasserstein GANs using various p-metrics. result Different p-metrics can notably improve GAN performance. KM-GAN generates images without labels using K-Means and GANs.
problem Reducing label dependence in GANs for efficient image generation.
method KM-GAN integrates K-Means clustering into GANs' framework for feature extraction and updates.
result KM-GAN generates high-quality images comparable to conditional GANs.
A new hybrid VAE-GAN framework improves mode coverage and quality.
problem Mode collapse and poor sample quality in GANs and VAEs.
method Integrates a 'Best-of-Many-Samples' reconstruction cost and a stable synthetic likelihood estimate.
result Significant improvement in mode coverage and quality compared to hybrid VAE-GANs and plain GANs.
Generative models create indistinguishable adversarial objects for object detection.
problem Creating unrestricted adversarial examples for object detection.
method Search over latent space of GAN for adversarial objects.
result Generated adversarial objects are indistinguishable from non-adversarial objects and transferable.
Generative Adversarial Networks (GAN) have limitations when the goal is to generate sequences of discrete elements. The reason for this is that samples from a distribution on discrete objects such as the multinomial are not differentiable with respect to the distribution parameters. This problem can be avoided by using…
ProAGAN stabilizes GANs for learning SOMs from noisy medical imaging data.
problem Learning stochastic object models from noisy and indirect medical imaging measurements.
method Developed Progressive Growing of AmbientGANs (ProAGAN) to stabilize GANs training.
result Signal detection performance improved using ProAGAN-generated images.
This paper analyzes theoretical properties of GANs.
problem Understanding the mathematical and statistical properties of GANs.
method Analyzing the adversarial principle, Jensen-Shannon divergence, and large sample properties.
result Proves a central limit theorem for estimated distribution.
Generative Adversarial Networks (GANs) excel at creating realistic images with complex models for which maximum likelihood is infeasible. However, the convergence of GAN training has still not been proved. We propose a two time-scale update rule (TTUR) for training GANs with stochastic gradient descent on arbitrary GAN…
A new GAN model improves object generation and reconstruction quality.
problem Generating realistic objects and their reconstructions.
method Joint training of a generator and encoder in an adversarial manner, introducing an augmented adversarial reconstruction loss.
result The model generates samples and reconstructions of quality competitive with state-of-the-art.
GAN-based semi-supervised learning improves classifier generalization.
problem Improving classifier performance with limited labeled data.
method Theoretical analysis of GAN-SSL, proving equivalence of discriminator optimization and supervised learning, and exploring conditions for perfect discriminator.
result GAN-SSL theoretically outputs a perfect discriminator on both labeled and unlabeled data.
Generative Adversarial Networks improve credit card fraud detection.
problem Detecting fraudulent credit card transactions accurately.
method Using GANs to generate synthetic data for oversampling.
result Wasserstein-GAN produces more realistic fraudulent transactions.
This paper analyzes GANs using Fourier modes to stabilize training.
problem Stability and convergence issues in GAN training.
method Decompose GAN objective function into Fourier series and study dynamics.
result Convergent orbits in GANs are small perturbations of periodic orbits, justifying slow training.
Proposes new loss functions for GANs to improve estimation accuracy and robustness.
problem Improving the training of GANs to achieve more accurate and robust models.
method Introduces Hellinger-type loss functions and analyzes their statistical properties.
result Demonstrates improved estimation accuracy and robustness of the proposed loss functions.
Generative adversarial networks (GANs) provide an algorithmic framework for constructing generative models with several appealing properties: they do not require a likelihood function to be specified, only a generating procedure; they provide samples that are sharp and compelling; and they allow us to harness our knowl…
This research shows unsupervised GANs can perform object segmentation without labels.
problem Performing object segmentation without pixel or image-level labels.
method Used large-scale unsupervised GAN models to differentiate foreground from background.
result Demonstrated high-quality saliency masks and new state-of-the-art performance.
CP-GAN generates images selectively conditioned on class specificity, capturing between-class relationships.
problem Generating images selectively conditioned on class specificity in class-overlapping data.
method Proposed Classifier's Posterior GAN (CP-GAN) that redesigns generator input and objective function for class-overlapping data.
result Demonstrated effectiveness of CP-GAN using both controlled and real-world class-overlapping data.
ACtuAL uses actor-critic for GANs to train on discrete data.
problem Training GANs with discrete data is difficult due to non-differentiability.
method Reframe GAN as actor-critic with TD objective for sequence modeling.
result Improvements in language modeling tasks over Teacher-Forcing.
Generative model learns to compose images of objects from different distributions.
problem Capturing complex interactions between objects in scenes.
method Composition-by-Decomposition (CoDe) network.
result Model generates realistic composite images capturing interactions between input objects.