Wasserstein Gradient Flows for Moreau Envelopes of f-Divergences in Reproducing Kernel Hilbert Spaces

Most commonly used $f$-divergences of measures, e.g., the Kullback-Leibler divergence, are subject to limitations regarding the support of the involved measures. A remedy consists of regularizing the $f$-divergence by a squared maximum mean discrepancy (MMD) associated with a characteristic kernel $K$. In this paper, we use the so-called kernel mean embedding to show that the corresponding regularization can be rewritten as the Moreau envelope of some function in the reproducing kernel Hilbert space associated with $K$. Then, we exploit well-known results on Moreau envelopes in Hilbert spaces to prove properties of the MMD-regularized $f$-divergences and, in particular, their gradients. Subsequently, we use our findings to analyze Wasserstein gradient flows of MMD-regularized $f$-divergences. Finally, we consider Wasserstein gradient flows starting from empirical measures and provide proof-of-the-concept numerical examples with Tsallis-$\alpha$ divergences.

Learning Weakly Convex Regularizers for Convergent Image-Reconstruction Algorithms

We propose to learn non-convex regularizers with a prescribed upper bound on their weak-convexity modulus. Such regularizers give rise to variational denoisers that minimize a convex energy. They rely on few parameters (less than 15,000) and offer a signal-processing interpretation as they mimic handcrafted sparsity-promoting regularizers. Through numerical experiments, we show that such denoisers outperform convex-regularization methods as well as the popular BM3D denoiser. Additionally, the learned regularizer can be deployed to solve inverse problems with iterative schemes that provably converge. For both CT and MRI reconstruction, the regularizer generalizes well and offers an excellent tradeoff between performance, number of parameters, guarantees, and interpretability when compared to other data-driven approaches.

On the Effect of Initialization: The Scaling Path of 2-Layer Neural Networks

In supervised learning, the regularization path is sometimes used as a convenient theoretical proxy for the optimization path of gradient descent initialized with zero. In this paper, we study a modification of the regularization path for infinite-width 2-layer ReLU neural networks with non-zero initial distribution of the weights at different scales. By exploiting a link with unbalanced optimal transport theory, we show that, despite the non-convexity of the 2-layer network training, this problem admits an infinite dimensional convex counterpart. We formulate the corresponding functional optimization problem and investigate its main properties. In particular, we show that as the scale of the initialization ranges between $0$ and $+\infty$, the associated path interpolates continuously between the so-called kernel and rich regimes. The numerical experiments confirm that, in our setting, the scaling path and the final states of the optimization path behave similarly even beyond these extreme points.

A Neural-Network-Based Convex Regularizer for Image Reconstruction

The emergence of deep-learning-based methods for solving inverse problems has enabled a significant increase in reconstruction quality. Unfortunately, these new methods often lack reliability and explainability, and there is a growing interest to address these shortcomings while retaining the performance. In this work, this problem is tackled by revisiting regularizers that are the sum of convex-ridge functions. The gradient of such regularizers is parametrized by a neural network that has a single hidden layer with increasing and learnable activation functions. This neural network is trained within a few minutes as a multi-step Gaussian denoiser. The numerical experiments for denoising, CT, and MRI reconstruction show improvements over methods that offer similar reliability guarantees.

Improving Lipschitz-Constrained Neural Networks by Learning Activation Functions

Lipschitz-constrained neural networks have several advantages compared to unconstrained ones and can be applied to various different problems. Consequently, they have recently attracted considerable attention in the deep learning community. Unfortunately, it has been shown both theoretically and empirically that networks with ReLU activation functions perform poorly under such constraints. On the contrary, neural networks with learnable 1-Lipschitz linear splines are known to be more expressive in theory. In this paper, we show that such networks are solutions of a functional optimization problem with second-order total-variation regularization. Further, we propose an efficient method to train such 1-Lipschitz deep spline neural networks. Our numerical experiments for a variety of tasks show that our trained networks match or outperform networks with activation functions specifically tailored towards Lipschitz-constrained architectures.

Stability of image reconstruction algorithms

Robustness and stability of image reconstruction algorithms have recently come under scrutiny. Their importance to medical imaging cannot be overstated. We review the known results for the topical variational regularization strategies ($\ell_2$ and $\ell_1$ regularization), and present new stability results for $\ell_p$ regularized linear inverse problems for $p\in(1,\infty)$. Our results generalize well to the respective $L_p(\Omega)$ function spaces.

Approximation of Lipschitz Functions using Deep Spline Neural Networks

Lipschitz-constrained neural networks have many applications in machine learning. Since designing and training expressive Lipschitz-constrained networks is very challenging, there is a need for improved methods and a better theoretical understanding. Unfortunately, it turns out that ReLU networks have provable disadvantages in this setting. Hence, we propose to use learnable spline activation functions with at least 3 linear regions instead. We prove that this choice is optimal among all component-wise $1$-Lipschitz activation functions in the sense that no other weight constrained architecture can approximate a larger class of functions. Additionally, this choice is at least as expressive as the recently introduced non component-wise Groupsort activation function for spectral-norm-constrained weights. Previously published numerical results support our theoretical findings.

Convolutional Proximal Neural Networks and Plug-and-Play Algorithms

In this paper, we introduce convolutional proximal neural networks (cPNNs), which are by construction averaged operators. For filters of full length, we propose a stochastic gradient descent algorithm on a submanifold of the Stiefel manifold to train cPNNs. In case of filters with limited length, we design algorithms for minimizing functionals that approximate the orthogonality constraints imposed on the operators by penalizing the least squares distance to the identity operator. Then, we investigate how scaled cPNNs with a prescribed Lipschitz constant can be used for denoising signals and images, where the achieved quality depends on the Lipschitz constant. Finally, we apply cPNN based denoisers within a Plug-and-Play (PnP) framework and provide convergence results for the corresponding PnP forward-backward splitting algorithm based on an oracle construction.

Stabilizing Invertible Neural Networks Using Mixture Models

In this paper, we analyze the properties of invertible neural networks, which provide a way of solving inverse problems. Our main focus lies on investigating and controlling the Lipschitz constants of the corresponding inverse networks. Without such an control, numerical simulations are prone to errors and not much is gained against traditional approaches. Fortunately, our analysis indicates that changing the latent distribution from a standard normal one to a Gaussian mixture model resolves the issue of exploding Lipschitz constants. Indeed, numerical simulations confirm that this modification leads to significantly improved sampling quality in multimodal applications.