Rob-GAN combines generator, discriminator, and adversarial attack for improved robustness and quality.
problem Improving robustness and quality of GAN-generated images under adversarial attacks.
method Rob-GAN framework that jointly optimizes generator and discriminator in the presence of adversarial attacks.
result Rob-GAN improves convergence speed, image quality, and robustness of discriminators under strong adversarial attacks.
Discrimination-aware classification is receiving an increasing attention in data science fields. The pre-process methods for constructing a discrimination-free classifier first remove discrimination from the training data, and then learn the classifier from the cleaned data. However, they lack a theoretical guarantee f…
Pairwise discriminators stabilize adversarial training by ensuring the generator's alignment is preserved.
problem Stability issues in adversarial training when using sub-optimal discriminators.
method Introducing a family of objectives using pairwise discriminators, ensuring the generator's alignment is preserved.
result Only the generator needs to converge, and the alignment is preserved with any discriminator.
Neural nets trained with linear discriminant initialization converge faster and more accurately.
problem Training feed-forward neural networks efficiently and accurately.
method Initialize first layer weights with linear discriminants.
result Asymptotic higher accuracy and faster convergence.
This paper proposes a curriculum learning method for GANs using multiple discriminators.
problem Training GANs with sufficient convergence conditions and avoiding mode collapse.
method A framework for training the generator against an ensemble of discriminator networks, formalized in the full-information adversarial bandit framework.
result Our approach improves samples quality and diversity over existing baselines by effectively learning a curriculum.
Self-supervised GAN prevents forgetting in sequential tasks.
problem Discriminator forgetting in GANs leads to training instability.
method Add self-supervision to the discriminator to maintain useful representations.
result Self-supervised GAN outperforms regular GANs in learning better representations.
Discriminator optimizes to approximate optimal transport for better image generation.
problem Improving the quality of generated images using GANs.
method Trains discriminator to optimize a lower bound of Wasserstein distance, approximating optimal transport.
result Trained discriminator improves inception score and FID metrics.
DNLL loss improves deep LDA accuracy and consistency.
problem Pathological solutions in unconstrained Deep LDA.
method Introducing Discriminative Negative Log-Likelihood (DNLL) loss.
result Deep LDA trained with DNLL produces clean latent spaces and better calibrated probabilities.
Generative adversarial nets (GANs) are a promising technique for modeling a distribution from samples. It is however well known that GAN training suffers from instability due to the nature of its maximin formulation. In this paper, we explore ways to tackle the instability problem by dualizing the discriminator. We sta…
New method improves GANs by training a mixed batch discriminator.
problem GANs struggle with mode collapse due to focusing on individual samples.
method Train a discriminator on mixed batches of true and fake samples.
result Significantly reduces mode collapse in GANs on various datasets.
We analyze GANs using neural tangent kernels, revealing flaws and advancing understanding.
problem Flaws in previous GAN analysis models.
method Neural Tangent Kernel framework for infinite-width discriminator.
result New insights into GAN convergence and generated distribution.
New method uses random discriminators to train GANs more efficiently.
problem Difficulty in training Generative Adversarial Networks (GANs).
method Proposes a new generative network using random discriminators.
result The method leads to a more stable and efficient optimization problem.
Develops a novel GAN method for better image generation.
problem Training stability and generalization in GANs.
method LD-GAN trains discriminator to maximize separability and generator based on LDA.
result Improved training stability and generalization in class conditional generation.
We present local discriminative Gaussian (LDG) dimensionality reduction, a supervised dimensionality reduction technique for classification. The LDG objective function is an approximation to the leave-one-out training error of a local quadratic discriminant analysis classifier, and thus acts locally to each training po…
AdvAs improves GAN training by penalizing the generator based on discriminator gradients.
problem Mismatch between theoretical analysis and practice in GAN training.
method AdvAs is a penalty imposed on the generator based on the discriminator's gradient norm.
result AdvAs reduces the mismatch between theory and practice and improves GAN training.
Metrics specifying distances between data points can be learned in a discriminative manner or from generative models. In this paper, we show how to unify generative and discriminative learning of metrics via a kernel learning framework. Specifically, we learn local metrics optimized from parametric generative models. T…
Paper analyzes the tradeoff between a discriminator's ability to identify true distribution and its ability to generalize.
problem Discrimination-generalization tradeoff in GANs.
method Analyzes the linear span of discriminator sets and their relationship to generalization and discriminative ability.
result Discriminator sets with dense linear span are both discriminative and generalizable.
Proposes a sparse classifier for discriminative Gaussian Mixture Models.
problem Softmax-based discriminative models assume unimodality, leading to parameter redundancy.
method Sparse Bayesian learning for GMM-based discriminative model, reducing parameters and complexity.
result The SDGM outperforms existing softmax-based discriminative models.
The paper examines GANs' forgetting and mode collapse, showing how they relate and impact training.
problem Catastrophic forgetting and mode collapse in GANs during continual learning.
method Investigates the continual learning nature of GANs, analyzing discriminator's output landscapes and convergence.
result Catastrophic forgetting and mode collapse are interrelated and prevent GANs from converging.
Linking GANs to binary classification through divergences.
problem Training GANs and understanding their divergence properties.
method Revisiting the discriminator's role in computing f-divergences.
result Alternative training perspective for f-GANs by designing discriminator loss.
The paper proposes using a discriminator for both domain adaptation and pseudo labeling confidence.
problem Improving generalization of classifiers trained on labeled source data to unlabeled target data.
method Multi-purposing the discriminator to learn domain-invariant feature representations and generate pseudo labels based on confidence.
result The approach enhances classifier performance by providing confidence measures for pseudo labels.
FairGAN generates fair data to prevent discrimination in GANs.
problem Preventing discrimination in generated data.
method FairGAN uses GANs to learn fair data generation.
result FairGAN generates fair data that also preserves data utility.
Multi-output Gaussian processes (MOGP) are probability distributions over vector-valued functions, and have been previously used for multi-output regression and for multi-class classification. A less explored facet of the multi-output Gaussian process is that it can be used as a generative model for vector-valued rando…
We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce con…
Solves GAN mode collapse by assigning minibatches to multiple discriminators.
problem Mode collapse in GANs where models generate similar samples.
method Multiple discriminators, microbatching, and changing tasks.
result Promotes sample diversity in generated sets.
This work improves GAN training by optimizing multiple discriminator losses.
problem Training GANs with multiple discriminators using single-objective methods.
method Formulates multi-objective optimization of multiple discriminator losses.
result Hypervolume maximization outperforms previous methods in sample quality and computational cost.
FairUDT uses uplift decision trees to detect and mitigate discrimination in training data.
problem Bias in machine learning classifiers due to historical discrimination or underrepresentation of minority groups.
method Integrates uplift modeling with decision trees and introduces a modified leaf relabeling approach for fairness.
result Achieves an acceptable accuracy-discrimination tradeoff while maintaining interpretability.
Study local convergence of GDA for training GANs with kernel-based discriminators.
problem Analyzing the local dynamics of GDA for GANs with kernel-based discriminators.
method Linearization of a non-linear dynamical system, under an isolated points model assumption.
result Showed phase transitions indicating convergence, oscillation, or divergence of GDA.
Feature learning forms the cornerstone for tackling challenging learning problems in domains such as speech, computer vision and natural language processing. In this paper, we consider a novel class of matrix and tensor-valued features, which can be pre-trained using unlabeled samples. We present efficient algorithms f…
New CGAN technique for efficient constrained topology design.
problem Constrained topology optimization with high computational cost.
method Conditional GAN (crCGAN) with deep CNNs and relaxed formulations.
result Improved efficiency and accuracy in topology design.
Improves GAN training by guiding the discriminator to have more diverse binary activation patterns.
problem Stability and convergence issues in GAN training.
method Binarized Representation Entropy (BRE) regularization to guide the discriminator's model capacity allocation.
result Improves GAN training stability and convergence speed, higher sample quality, and higher classification accuracy.
Feature learning forms the cornerstone for tackling challenging learning problems in domains such as speech, computer vision and natural language processing. In this paper, we consider a novel class of matrix and tensor-valued features, which can be pre-trained using unlabeled samples. We present efficient algorithms f…
Improves GANs by enforcing diverse feature learning.
problem GANs can collapse to a single configuration and be unstable.
method Enforces diverse feature learning by penalizing correlated features.
result Enforces diverse features, stabilizes training, and improves image synthesis.
DoPaNet uses multiple discriminators to prevent mode collapse in GANs.
problem Mode collapse in adversarial training.
method Employing multiple discriminators and a classifier to guide the generator.
result DoPaNet effectively covers the target distribution and outperforms competing methods.
Paper proposes faster incremental subclass discriminant analysis.
problem Efficiently classify subclasses in incremental data.
method Exact and approximate linear and kernelized solutions.
result Superior training time and accuracy compared to existing methods.
We introduce a method to stabilize Generative Adversarial Networks (GANs) by defining the generator objective with respect to an unrolled optimization of the discriminator. This allows training to be adjusted between using the optimal discriminator in the generator's objective, which is ideal but infeasible in practice…
This work compares lattice-free and lattice-based training criteria for LVCSR.
problem Improving acoustic model performance in speech recognition.
method Direct comparison of lattice-free and lattice-based sequence discriminative training criteria using GPU.
result Lattice-free MMI performance is comparable to lattice-based criteria, while lattice-based sMBR remains superior.
Develops DDC to improve clustering with deep neural networks.
problem Low-level indiscriminative representations and lack of pattern relationships in traditional clustering methods.
method Introduces global and local constraints to a deep neural network for adaptive relationship estimation and high-level representation learning.
result DDC outperforms current methods on multiple datasets.
Improved GAN performance with incomplete data using factorised discriminators.
problem Limited availability of labelled data for GAN training.
method Factorising data distribution into sub-distributions and training sub-discriminators.
result Improved performance in image generation, segmentation, and audio separation tasks.
Develops Triangle GAN for semi-supervised cross-domain learning.
problem Semi-supervised cross-domain joint distribution matching with limited labeled data.
method Triangle GAN architecture with two generators and two discriminators trained adversarially.
result Generators learn conditional distributions between domains, discriminators define ternary function.
New method boosts skill learning by encouraging optimistic exploration.
problem Intrinsic reward for exploration is inherently pessimistic.
method Derive an information gain auxiliary objective involving an ensemble of discriminators and rewarding policy disagreement.
result Improves skill learning in grid worlds and Atari games.
Improves GAN training stability and quality through a tempered learning process.
problem Training instability and low quality samples in GANs.
method Integrates a 'tempering' module that controls the real data distribution, balancing generator and discriminator.
result Improves quality, stability, and convergence speed across various GAN architectures.
Curriculum GANs improve GAN training by making tasks harder over time.
problem Improving training of Generative Adversarial Networks (GANs).
method A curriculum learning strategy that increases discriminator strength progressively.
result State-of-the-art image generation results achieved.
Generative model improves latent space convexity through adversarial training on interpolations.
problem Improving latent space convexity in generative models.
method Adversarial training on latent space interpolations within an AE-GAN architecture.
result Convex latent distribution of generated images, preserving realistic resemblances.
Paper analyzes and improves GANs' generalization and stability.
problem Poor generalization of GANs' discriminators in practical settings.
method Proposes a zero-centered gradient penalty to improve discriminator's generalization.
result Improves GANs' generalization and convergence through the proposed penalty.
This paper proposes an online knowledge distillation method that transfers feature map information in addition to class probabilities.
problem Previous online knowledge distillation methods only utilized class probabilities, missing feature map information.
method Adversarial training framework to transfer feature map information; multiple networks trained simultaneously with discriminators.
result Our method performs better than direct alignment methods and is more suitable for online distillation.
Paper introduces new loss functions for Siamese networks using FDA.
problem Training Siamese networks with improved loss functions.
method Proposes Fisher Discriminant Triplet (FDT) and Fisher Discriminant Contrastive (FDC) loss functions based on FDA.
result Shows effectiveness of FDT and FDC on MNIST and histopathology datasets.
VDB constrains discriminator to improve imitation learning and GANs.
problem Stability and performance issues in adversarial learning methods.
method Information bottleneck technique to modulate discriminator's accuracy.
result Significant improvements across imitation learning, inverse RL, and GANs.