arXiv research

New control theory approach stabilizes GANs training.

Generative adversarial networks (GANs) are the state of the art in generative modeling. Unfortunately, most GAN methods are susceptible to mode collapse, meaning that they tend to capture only a subset of the modes of the true distribution. A possible way of dealing with this problem is to use an ensemble of GANs, wher…

Survey of GANs and autoencoders, addressing mode collapse and likelihood issues.

A new GAN model $α$-GAN with tunable loss function addresses gradient vanishing and mode collapse issues.

We define two categories of Dirac manifolds, i.e. manifolds with complex Dirac structures. The first notion of maps I call \emph{Dirac maps}, and the category of Dirac manifolds is seen to contain the categories of Poisson and complex manifolds as full subcategories. The second notion, \emph{dual-Dirac maps}, defines a…

Unbalanced GANs stabilize GAN training by pre-training the generator with VAE.

GANs can approximate SDEs for large time steps.

Alternative discrete Dirac mechanics using Dirac structures.

Introduces weak $(p,k)$-Dirac structures in geometric settings.

Paper bridges $f$-GANs and WGANs for better image generation.

Defines Dirac pairs on Jacobi algebroids, generalizing Lie algebroids.

In this paper we introduce the Dirac and spin-Dirac operators associated to a connection on Riemann-Cartan space(time) and standard Dirac and spin-Dirac operators associated with a Levi-Civita connection on a Riemannian (Lorentzian) space(time) and calculate the square of these operators, which play an important role i…

Stokes-Dirac structures are infinite-dimensional Dirac structures defined in terms of differential forms on a smooth manifold with boundary. These Dirac structures lay down a geometric framework for the formulation of Hamiltonian systems with a nonzero boundary energy flow. Simplicial triangulation of the underlaying m…

Introduces compatibility between Dirac structures and Nijenhuis tensors.

Given a Dirac subbundle and an isotropic subbundle of a Courant algebroid, we provide a canonical method to obtain a new Dirac subbundle. When the original Dirac subbundle is involutive (i.e., a Dirac structure) this construction has interesting applications, for instance to Dirac's theory of constraints and to the Mar…

We show that a suitable notion of Dirac-Jacobi structure on a generic line bundle $L$, is provided by Dirac structures in the omni-Lie algebroid of $L$. Dirac-Jacobi structures on line bundles generalize Wade's $\mathcal E^1 (M)$-Dirac structures and unify generic (i.e.~non-necessarily coorientable) precontact distribu…

Prb-GAN uses dropout and variational inference to improve GAN performance.

We introduce Dirac processes, using Dirac delta functions, for short-rate-type pricing of financial derivatives. Dirac processes add spikes to the existing building blocks of diffusions and jumps. Dirac processes are Generalized Processes, which have not been used directly before because the dollar value of non-Real nu…

xAI-GAN improves GANs by providing richer feedback, enhancing image quality.

New spectral torsion defined for rescaled Dirac operators.

This paper reviews GANs algorithms, theory, and applications.

Generative Adversarial Networks (GANs) are powerful models for learning complex distributions. Stable training of GANs has been addressed in many recent works which explore different metrics between distributions. In this paper we introduce Fisher GAN which fits within the Integral Probability Metrics (IPM) framework f…

A classical theorem of Drinfel'd states that the category of simply connected Poisson Lie groups H is isomorphic to the category of Manin triples (d, g, h), where h is the Lie algebra of H. In this paper, we consider Dirac Lie groups, that is, Lie groups H endowed with a multiplicative Courant algebroid A and a Dirac s…

A recent technical breakthrough in the domain of machine learning is the discovery and the multiple applications of Generative Adversarial Networks (GANs). Those generative models are computationally demanding, as a GAN is composed of two deep neural networks, and because it trains on large datasets. A GAN is generally…

We investigate the training and performance of generative adversarial networks using the Maximum Mean Discrepancy (MMD) as critic, termed MMD GANs. As our main theoretical contribution, we clarify the situation with bias in GAN loss functions raised by recent work: we show that gradient estimators used in the optimizat…

GANs may not have Nash equilibria, but proximal training can find solutions.

Empirical study shows GANs overfit and drop modes when training is deterministic.

GANs generate images of emotions from datasets.

We considered an extension of the standard functional for the Einstein-Dirac equation where the Dirac operator is replaced by the square of the Dirac operator and a real parameter controlling the length of spinors is introduced. For one distinguished value of the parameter, the resulting Euler-Lagrange equations provid…

Generative Adversarial Networks (GANs) have been used in several machine learning tasks such as domain transfer, super resolution, and synthetic data generation. State-of-the-art GANs often use tens of millions of parameters, making them expensive to deploy for applications in low SWAP (size, weight, and power) hardwar…

GANs analyzed for performance and training issues.

Characterizes obstacles to variational formulation of Dirac dynamics.

Despite excellent progress in recent years, mode collapse remains a major unsolved problem in generative adversarial networks (GANs).In this paper, we present spectral regularization for GANs (SR-GANs), a new and robust method for combating the mode collapse problem in GANs. Theoretical analysis shows that the optimal …

Vanilla GANs are connected to Wasserstein distance for better understanding.

Several dihedral angles prediction methods were developed for protein structure prediction and their other applications. However, distribution of predicted angles would not be similar to that of real angles. To address this we employed generative adversarial networks (GAN). Generative adversarial networks are composed …

Generative adversarial networks (GAN) have been effective for learning generative models for real-world data. However, existing GANs (GAN and its variants) tend to suffer from training problems such as instability and mode collapse. In this paper, we propose a novel GAN framework called evolutionary generative adversar…

Unified framework compresses GANs up to 47x with minimal quality loss.

Study on symplectic Dirac operators on foliations, estimating eigenvalues.

NR-GANs learn clean images from noisy data.

New method improves GANs by estimating density ratios in feature space with SP loss.

This work improves GAN stability with theoretical conditions.

This paper uses SDEs to analyze GANs training and long-run behavior.

Connected space of Dirac-minimal metrics in 2 and 4 dimensions.

The paper introduces discrete Dirac structures for mechanics, simplifying dynamics.

Study on magnetic Dirac operators and their spectrum.

The paper defines Dirac structures on connection spaces and their properties.

Generative Adversarial Networks (GANs) are a powerful class of generative models. Despite their successes, the most appropriate choice of a GAN network architecture is still not well understood. GAN models for image synthesis have adopted a deep convolutional network architecture, which eliminates or minimizes the use …

Study on the existence of Dirac complements for Dirac structures.