ODE${}^2$VAE: Deep generative second order ODEs with Bayesian neural networks

Abstract

We present Ordinary Differential Equation Variational Auto-Encoder (ODE${}^2$VAE), a latent second order ODE model for high-dimensional sequential data. Leveraging the advances in deep generative models, ODE${}^2$VAE can simultaneously learn the embedding of high dimensional trajectories and infer arbitrarily complex continuous-time latent dynamics. Our model explicitly decomposes the latent space into momentum and position components and solves a second order ODE system, which is in contrast to recurrent neural network (RNN) based time series models and recently proposed black-box ODE techniques. In order to account for uncertainty, we propose probabilistic latent ODE dynamics parameterized by deep Bayesian neural networks. We demonstrate our approach on motion capture, image rotation and bouncing balls datasets. We achieve state-of-the-art performance in long term motion prediction and imputation tasks.

References in corpus (14)

• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• A Recurrent Latent Variable Model for Sequential Data
• Deep Predictive Coding Networks for Video Prediction and Unsupervised Learning
• Recent Advances in Autoencoder-Based Representation Learning
• InfoVAE: Information Maximizing Variational Autoencoders
• Wasserstein Auto-Encoders
• FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models
• Unsupervised Learning of Visual Structure using Predictive Generative Networks
• Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data
• Deep Temporal Sigmoid Belief Networks for Sequence Modeling
• Variational Gaussian Process Dynamical Systems
• Fixing a Broken ELBO
• Learning unknown ODE models with Gaussian processes
• Continuous-Time Flows for Efficient Inference and Density Estimation

Cited by in corpus (31)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• Scalable Gradients for Stochastic Differential Equations
• How to train your neural ODE: the world of Jacobian and kinetic regularization
• Dissecting Neural ODEs
• Graph Neural Ordinary Differential Equations
• Parameterized Neural Ordinary Differential Equations: Applications to Computational Physics Problems
• Recipes for when Physics Fails: Recovering Robust Learning of Physics Informed Neural Networks
• Evolutionary Preference Learning via Graph Nested GRU ODE for Session-based Recommendation
• Reducing the complexity of chemical networks via interpretable autoencoders
• Hypersolvers: Toward Fast Continuous-Depth Models
• STEER: Simple Temporal Regularization For Neural ODEs
• LagNetViP: A Lagrangian Neural Network for Video Prediction
• Bayesian Neural Ordinary Differential Equations
• Infinitely Deep Bayesian Neural Networks with Stochastic Differential Equations
• Continuous PDE Dynamics Forecasting with Implicit Neural Representations
• PDE-Driven Spatiotemporal Disentanglement
• Stable Neural Flows
• Physics-Integrated Variational Autoencoders for Robust and Interpretable Generative Modeling
• Stochasticity in Neural ODEs: An Empirical Study
• A Shooting Formulation of Deep Learning
• Learning Continuous-Time Dynamics by Stochastic Differential Networks
• Continuous-Time Model-Based Reinforcement Learning
• Continuous-Depth Neural Models for Dynamic Graph Prediction
• Latent Neural Differential Equations for Video Generation
• Efficient and Accurate Gradients for Neural SDEs
• Neural Lyapunov Model Predictive Control: Learning Safe Global Controllers from Sub-optimal Examples
• Ensemble Kalman Variational Objectives: Nonlinear Latent Trajectory Inference with A Hybrid of Variational Inference and Ensemble Kalman Filter
• Capturing Actionable Dynamics with Structured Latent Ordinary Differential Equations
• Continual Learning of Multi-modal Dynamics with External Memory
• VidStyleODE: Disentangled Video Editing via StyleGAN and NeuralODEs
• SurvODE: Extrapolating Gene Expression Distribution for Early Cancer Identification