Continuous Methods : Hamiltonian Domain Translation
This paper proposes a novel approach to domain translation. Leveraging established parallels between generative models and dynamical systems, we propose a reformulation of the Cycle-GAN architecture. By embedding our model with a Hamiltonian structure, we obtain a continuous, expressive and most importantly invertible generative model for domain translation.
Code (0)
등록된 구현이 없습니다.
Tasks
TranslationSimilar Papers 제목 키워드 기반
Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
The translation equivariance of convolutional layers enables convolutional neural networks to generalize well on image problems. While translation equivariance provides a powerful inductive bias for images, we often addi…
Inductive BiasTranslationNumerical Generalized Randomized Hamiltonian Monte Carlo for piecewise smooth target densities
Traditional gradient-based sampling methods, like standard Hamiltonian Monte Carlo, require that the desired target distribution is continuous and differentiable. This limits the types of models one can define, although …
Hamiltonian Generative Networks
The Hamiltonian formalism plays a central role in classical and quantum physics. Hamiltonians are the main tool for modelling the continuous time evolution of systems with conserved quantities, and they come equipped wit…
Reinforcement LearningStructure preserving discontinuous Galerkin approximation of one-dimensional port-Hamiltonian systems
In this article, we present the structure-preserving discretization of linear one-dimensional port-Hamiltonian (PH) systems of two conservation laws using discontinuous Galerkin (DG) methods. We recall the DG discretizat…
Continuous and Diverse Image-to-Image Translation via Signed Attribute Vectors
Recent image-to-image (I2I) translation algorithms focus on learning the mapping from a source to a target domain. However, the continuous translation problem that synthesizes intermediate results between two domains has…
AttributeImage-to-Image TranslationTranslation