In this section, we give an overview of related work in stable neural ODE networks. We also give an overview of common adversarial attacks and recent works that defend against adversarial examples. Stable Neural Network Gradient vanishing and gradient exploding are two well-known phenomena in deep learning [1]. The gradient of the objective function, which strongly relies on the training method as well as the neural network architecture, indicates how sensitive the output is with respect to (w.r.t.) input perturbation. Exploding gradient implies instability of the output w.r.t. the input and thus resulting in a non-robust learning architecture.

Traffic forecasting is an important issue in intelligent traffic systems (ITS). Graph neural networks (GNNs) are effective deep learning models to capture the complex spatio-temporal dependency of traffic data, achieving ideal prediction performance. In this paper, we propose attention-based graph neural ODE (ASTGODE) that explicitly learns the dynamics of the traffic system, which makes the prediction of our machine learning model more explainable. Our model aggregates traffic patterns of different periods and has satisfactory performance on two real-world traffic data sets. The results show that our model achieves the highest accuracy of the root mean square error metric among all the existing GNN models in our experiments.

Deep residual networks (ResNets) have shown state-of-the-art performance in various real-world applications. Recently, the ResNets model was reparameterized and interpreted as solutions to a continuous ordinary differential equation or Neural-ODE model. In this study, we propose a neural generalized ordinary differential equation (Neural-GODE) model with layer-varying parameters to further extend the Neural-ODE to approximate the discrete ResNets. Specifically, we use nonparametric B-spline functions to parameterize the Neural-GODE so that the trade-off between the model complexity and computational efficiency can be easily balanced. It is demonstrated that ResNets and Neural-ODE models are special cases of the proposed Neural-GODE model. Based on two benchmark datasets, MNIST and CIFAR-10, we show that the layer-varying Neural-GODE is more flexible and general than the standard Neural-ODE. Furthermore, the Neural-GODE enjoys the computational and memory benefits while performing comparably to ResNets in prediction accuracy.

Embedding nonlinear dynamical systems into artificial neural networks is a powerful new formalism for machine learning. By parameterizing ordinary differential equations (ODEs) as neural network layers, these Neural ODEs are memory-efficient to train, process time-series naturally and incorporate knowledge of physical systems into deep learning models. However, the practical applications of Neural ODEs are limited due to long inference times, because the outputs of the embedded ODE layers are computed numerically with differential equation solvers that can be computationally demanding. Here we show that mathematical model order reduction methods can be used for compressing and accelerating Neural ODEs by accurately simulating the continuous nonlinear dynamics in low-dimensional subspaces. We implement our novel compression method by developing Neural ODEs that integrate the necessary subspace-projection and interpolation operations as layers of the neural network. We validate our approach by comparing it to neuron pruning and SVD-based weight truncation methods from the literature in image and time-series classification tasks. The methods are evaluated by acceleration versus accuracy when adjusting the level of compression. On this spectrum, we achieve a favourable balance over existing methods by using model order reduction when compressing a convolutional Neural ODE. In compressing a recurrent Neural ODE, SVD-based weight truncation yields good performance. Based on our results, our integration of model order reduction with Neural ODEs can facilitate efficient, dynamical system-driven deep learning in resource-constrained applications.

In this paper we introduce a provably stable architecture for Neural Ordinary Differential Equations (ODEs) which achieves non-trivial adversarial robustness under white-box adversarial attacks even when the network is trained naturally. For most existing defense methods withstanding strong white-box attacks, to improve robustness of neural networks, they need to be trained adversarially, hence have to strike a trade-off between natural accuracy and adversarial robustness. Inspired by dynamical system theory, we design a stabilized neural ODE network named SONet whose ODE blocks are skew-symmetric and proved to be input-output stable. With natural training, SONet can achieve comparable robustness with the state-of-the-art adversarial defense methods, without sacrificing natural accuracy. Even replacing only the first layer of a ResNet by such a ODE block can exhibit further improvement in robustness, e.g., under PGD-20 ($\ell_\infty=0.031$) attack on CIFAR-10 dataset, it achieves 91.57\% and natural accuracy and 62.35\% robust accuracy, while a counterpart architecture of ResNet trained with TRADES achieves natural and robust accuracy 76.29\% and 45.24\%, respectively. To understand possible reasons behind this surprisingly good result, we further explore the possible mechanism underlying such an adversarial robustness. We show that the adaptive stepsize numerical ODE solver, DOPRI5, has a gradient masking effect that fails the PGD attacks which are sensitive to gradient information of training loss; on the other hand, it cannot fool the CW attack of robust gradients and the SPSA attack that is gradient-free. This provides a new explanation that the adversarial robustness of ODE-based networks mainly comes from the obfuscated gradients in numerical ODE solvers.

Normalization is an important and vastly investigated technique in deep learning. However, its role for Ordinary Differential Equation based networks (neural ODEs) is still poorly understood. This paper investigates how different normalization techniques affect the performance of neural ODEs. Particularly, we show that it is possible to achieve 93% accuracy in the CIFAR-10 classification task, and to the best of our knowledge, this is the highest reported accuracy among neural ODEs tested on this problem.

In this paper, we propose a method, which allows us to alleviate or completely avoid the notorious problem of numerical instability and stiffness of the adjoint method for training neural ODE. On the backward pass, we propose to use the machinery of smooth function interpolation to restore the trajectory obtained during the forward integration. We show the viability of our approach, both in theory and practice.