All Projects → wangleiphy → MongeAmpereFlow

wangleiphy / MongeAmpereFlow

Licence: other
Continuous-time gradient flow for generative modeling and variational inference

Programming Languages

python
139335 projects - #7 most used programming language
Mathematica
289 projects

Projects that are alternatives of or similar to MongeAmpereFlow

egtplot
egtplot: A python package for 3-Strategy Evolutionary Games
Stars: ✭ 33 (+13.79%)
Mutual labels:  dynamical-systems
benchmark VAE
Unifying Variational Autoencoder (VAE) implementations in Pytorch (NeurIPS 2022)
Stars: ✭ 1,211 (+4075.86%)
Mutual labels:  normalizing-flows
radCAD
A framework for generalised dynamical systems modelling & simulation (inspired by and compatible with cadCAD.org)
Stars: ✭ 59 (+103.45%)
Mutual labels:  dynamical-systems
awesome-nn-optimization
Awesome list for Neural Network Optimization methods.
Stars: ✭ 39 (+34.48%)
Mutual labels:  dynamical-systems
flowtorch-old
Separating Normalizing Flows code from Pyro and improving API
Stars: ✭ 36 (+24.14%)
Mutual labels:  normalizing-flows
NanoFlow
PyTorch implementation of the paper "NanoFlow: Scalable Normalizing Flows with Sublinear Parameter Complexity." (NeurIPS 2020)
Stars: ✭ 63 (+117.24%)
Mutual labels:  normalizing-flows
rom-operator-inference-Python3
Operator Inference for data-driven, non-intrusive model reduction of dynamical systems.
Stars: ✭ 31 (+6.9%)
Mutual labels:  dynamical-systems
deeprob-kit
A Python Library for Deep Probabilistic Modeling
Stars: ✭ 32 (+10.34%)
Mutual labels:  normalizing-flows
Pontryagin-Differentiable-Programming
A unified end-to-end learning and control framework that is able to learn a (neural) control objective function, dynamics equation, control policy, or/and optimal trajectory in a control system.
Stars: ✭ 111 (+282.76%)
Mutual labels:  dynamical-systems
progressive-growing-of-gans.pytorch
Unofficial PyTorch implementation of "Progressive Growing of GANs for Improved Quality, Stability, and Variation".
Stars: ✭ 51 (+75.86%)
Mutual labels:  optimal-transport
mbsim
A multi-body simulation software
Stars: ✭ 22 (-24.14%)
Mutual labels:  dynamical-systems
Differentialequations.jl
Multi-language suite for high-performance solvers of differential equations and scientific machine learning (SciML) components
Stars: ✭ 2,023 (+6875.86%)
Mutual labels:  dynamical-systems
pressio
Model reduction for linear and nonlinear dynamical systems: core C++ library
Stars: ✭ 35 (+20.69%)
Mutual labels:  dynamical-systems
sssMOR
sssMOR - Sparse State-Space and Model Order Reduction Toolbox
Stars: ✭ 20 (-31.03%)
Mutual labels:  dynamical-systems
SDETools
Matlab Toolbox for the Numerical Solution of Stochastic Differential Equations
Stars: ✭ 80 (+175.86%)
Mutual labels:  dynamical-systems
dynamical-systems-with-applications-using-python
Source code for 'Dynamical Systems with Applications Using Python' by Stephen Lynch
Stars: ✭ 100 (+244.83%)
Mutual labels:  dynamical-systems
NonlinearDynamicsTextbook
Nonlinear Dynamics: A concise introduction interlaced with code
Stars: ✭ 145 (+400%)
Mutual labels:  dynamical-systems
eom
Configurable ODE/PDE solver
Stars: ✭ 50 (+72.41%)
Mutual labels:  dynamical-systems
BrainPy
Brain Dynamics Programming in Python
Stars: ✭ 242 (+734.48%)
Mutual labels:  dynamical-systems
sinkhorn-label-allocation
Sinkhorn Label Allocation is a label assignment method for semi-supervised self-training algorithms. The SLA algorithm is described in full in this ICML 2021 paper: https://arxiv.org/abs/2102.08622.
Stars: ✭ 49 (+68.97%)
Mutual labels:  optimal-transport

PyTorch implementation of “Monge-Ampère Flow for Generative Modeling” arXiv:1809.10188

How to run the code

Density estimation of MNIST

python density_estimation.py -dataset MNIST -hdim 1024 -Nsteps 100 -train -cuda 7

Variational free energy of Ising

python variational_free_energy.py -L 16 -fe_exact -2.3159198563359373 -train -cuda 7 -hdim 512 -Nsteps 50 -Batchsize 64 -symmetrize

Plots in the paper

  • MNIST NLL
python paper/plot_nll.py -outname nll.pdf 
  • Gaussianization MNIST
python density_estimation.py -hdim 1024 -Nsteps 100 -epsilon 0.1 -checkpoint data/learn_mnist/Simple_MLP_hdim1024_Batchsize100_lr0.001_Nsteps100_epsilon0.1/epoch-1.chkp -show 
  • Direct sample Ising
python variational_free_energy.py -hdim 512 -Nsteps 50 -checkpoint data/learn_ot/ising_L16_d2_T2.269185314213022_symmetrize_Simple_MLP_hdim512_Batchsize64_lr0.001_delta0.0_Nsteps50_epsilon0.1/epoch-1.chkp -show  -L 16  -symmetrize 

Reference: Exact Ising free energy density at critical temperature on $L\times L$ lattices (For details see Appendix B of the paper)

$L$ periodic open
4 -2.33604476445 -1.9470001244979966
8 -2.3227349295609376 -2.1909718508291
16 -2.3159198563359373 -2.272901214087426
32 -2.3140498159960936 -2.2993352217736573
64 -2.3135805785878905 -2.3080749864821253
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].