We consider the decentralized optimization problem, where a network of $n$ agents aims to collaboratively minimize the average of their individual smooth and convex objective functions through peer-to-peer communication in a directed graph. To tackle this problem, we propose two accelerated gradient tracking methods, namely APD and APD-SC, for non-strongly convex and strongly convex objective functions, respectively. We show that APD and APD-SC converge at the rates $O\left(\frac{1}{k^2}\right)$ and $O\left(\left(1 - C\sqrt{\frac{\mu}{L}}\right)^k\right)$, respectively, up to constant factors depending only on the mixing matrix. APD and APD-SC are the first decentralized methods over unbalanced directed graphs that achieve the same provable acceleration as centralized methods. Numerical experiments demonstrate the effectiveness of both methods.

Gaussian noise was then added to produce a noisy simulated data. Given the data, the posterior of a model (a pixelated image of the undistorted background source) could be calculated by adding the likelihood and the prior terms. Furthermore since the model is perfectly linear (and known) and the noise and the prior are Gaussian, the posterior is a high-dimensional Gaussian posterior that could be calculated analytically, allowing us to compare the samples drawn with GGNS with the analytic solution. Figure 2 shows a comparison between the true image, and its noise, and the one recovered by GGNS. We see that we can recover both the correct image, and the noise distribution. We emphasize that this is a uni-modal problem and that the experiment's goal is to demonstrate the capability of GGNS to sample in high dimensions (in this case, 256), such as images, and to test the agreement between the samples and a baseline analytic solution.

The Exponential Mechanism (ExpM), a differentially private optimization method, promises many advantages over Differentially Private Stochastic Gradient Descent (DPSGD), the state-of-the-art (SOTA) and de facto method for differentially private machine learning (ML). Yet, ExpM has been historically stymied from differentially private training of modern ML algorithms by two obstructions: ExpM requires a sensitivity bound for the given loss function; ExpM requires sampling from a historically intractable density. We prove a sensitivity bound for $\ell(2)$ loss, and investigate using Normalizing Flows (NFs), deep networks furnishing approximate sampling from the otherwise intractable ExpM distribution. We prove that as the NF output converges to ExpM distribution, the privacy ($\varepsilon$) of an NF sample converges to that of the ExpM distribution. Under the assumption that the NF output distribution is the ExpM distribution, we empirically test ExpM+NF against DPSGD using the SOTA implementation (Opacus \cite{opacus} with PRV accounting) in multiple classification tasks on the Adult Dataset (census data) and MIMIC-III Dataset (healthcare records) using Logistic Regression and GRU-D, a deep learning recurrent neural network with \smallsim 20K-100K parameters. In all experiments we find ExpM+NF achieves greater than 94\% of the non-private training accuracy (AUC) with $\varepsilon$-DP for $\varepsilon$ a low as $1\mathrm{e}{-3}$ -- three orders of magnitude stronger privacy with similar accuracy. Further, performance results show ExpM+NF training time is comparable to (slightly less) than DPSGD. Limitations and future directions are provided; notably, research on NF approximation accuracy and its effect on privacy are a promising avenue to substantially advancing the field. Code for these experiments \hl{will be provided after review}.

We propose a solution for linear inverse problems based on higher-order Langevin diffusion. More precisely, we propose pre-conditioned second-order and third-order Langevin dynamics that provably sample from the posterior distribution of our unknown variables of interest while being computationally more efficient than their first-order counterpart and the non-conditioned versions of both dynamics. Moreover, we prove that both pre-conditioned dynamics are well-defined and have the same unique invariant distributions as the non-conditioned cases. We also incorporate an annealing procedure that has the double benefit of further accelerating the convergence of the algorithm and allowing us to accommodate the case where the unknown variables are discrete. Numerical experiments in two different tasks in communications (MIMO symbol detection and channel estimation) and in three tasks for images showcase the generality of our method and illustrate the high performance achieved relative to competing approaches (including learning-based ones) while having comparable or lower computational complexity.

Shapley values are extensively used in explainable artificial intelligence (XAI) as a framework to explain predictions made by complex machine learning (ML) models. In this work, we focus on conditional Shapley values for predictive models fitted to tabular data and explain the prediction $f(\boldsymbol{x}^{*})$ for a single observation $\boldsymbol{x}^{*}$ at the time. Numerous Shapley value estimation methods have been proposed and empirically compared on an average basis in the XAI literature. However, less focus has been devoted to analyzing the precision of the Shapley value explanations on an individual basis. We extend our work in Olsen et al. (2023) by demonstrating and discussing that the explanations are systematically less precise for observations on the outer region of the training data distribution for all used estimation methods. This is expected from a statistical point of view, but to the best of our knowledge, it has not been systematically addressed in the Shapley value literature. This is crucial knowledge for Shapley values practitioners, who should be more careful in applying these observations' corresponding Shapley value explanations.

Diabetes encompasses a complex landscape of glycemic control that varies widely among individuals. However, current methods do not faithfully capture this variability at the meal level. On the one hand, expert-crafted features lack the flexibility of data-driven methods; on the other hand, learned representations tend to be uninterpretable which hampers clinical adoption. In this paper, we propose a hybrid variational autoencoder to learn interpretable representations of CGM and meal data. Our method grounds the latent space to the inputs of a mechanistic differential equation, producing embeddings that reflect physiological quantities, such as insulin sensitivity, glucose effectiveness, and basal glucose levels. Moreover, we introduce a novel method to infer the glucose appearance rate, making the mechanistic model robust to unreliable meal logs. On a dataset of CGM and self-reported meals from individuals with type-2 diabetes and pre-diabetes, our unsupervised representation discovers a separation between individuals proportional to their disease severity. Our embeddings produce clusters that are up to 4x better than naive, expert, black-box, and pure mechanistic features. Our method provides a nuanced, yet interpretable, embedding space to compare glycemic control within and across individuals, directly learnable from in-the-wild data.

Self-training and contrastive learning have emerged as leading techniques for incorporating unlabeled data, both under distribution shift (unsupervised domain adaptation) and when it is absent (semi-supervised learning). However, despite the popularity and compatibility of these techniques, their efficacy in combination remains unexplored. In this paper, we undertake a systematic empirical investigation of this combination, finding that (i) in domain adaptation settings, self-training and contrastive learning offer significant complementary gains; and (ii) in semi-supervised learning settings, surprisingly, the benefits are not synergistic. Across eight distribution shift datasets (e.g., BREEDs, WILDS), we demonstrate that the combined method obtains 3--8% higher accuracy than either approach independently. We then theoretically analyze these techniques in a simplified model of distribution shift, demonstrating scenarios under which the features produced by contrastive learning can yield a good initialization for self-training to further amplify gains and achieve optimal performance, even when either method alone would fail.

Kernel methods are a popular class of nonlinear predictive models in machine learning. Scalable algorithms for learning kernel models need to be iterative in nature, but convergence can be slow due to poor conditioning. Spectral preconditioning is an important tool to speed-up the convergence of such iterative algorithms for training kernel models. However computing and storing a spectral preconditioner can be expensive which can lead to large computational and storage overheads, precluding the application of kernel methods to problems with large datasets. A Nystrom approximation of the spectral preconditioner is often cheaper to compute and store, and has demonstrated success in practical applications. In this paper we analyze the trade-offs of using such an approximated preconditioner. Specifically, we show that a sample of logarithmic size (as a function of the size of the dataset) enables the Nystrom-based approximated preconditioner to accelerate gradient descent nearly as well as the exact preconditioner, while also reducing the computational and storage overheads.

In the process of training a generative model, it becomes essential to measure the discrepancy between two high-dimensional probability distributions: the generative distribution and the ground-truth distribution of the observed dataset. Recently, there has been growing interest in an approach that involves slicing high-dimensional distributions, with the Cramer-Wold distance emerging as a promising method. However, we have identified that the Cramer-Wold distance primarily focuses on joint distributional learning, whereas understanding marginal distributional patterns is crucial for effective synthetic data generation. In this paper, we introduce a novel measure of dissimilarity, the mixture Cramer-Wold distance. This measure enables us to capture both marginal and joint distributional information simultaneously, as it incorporates a mixture measure with point masses on standard basis vectors. Building upon the mixture Cramer-Wold distance, we propose a new generative model called CWDAE (Cramer-Wold Distributional AutoEncoder), which shows remarkable performance in generating synthetic data when applied to real tabular datasets. Furthermore, our model offers the flexibility to adjust the level of data privacy with ease.

We propose to solve inverse problems involving the temporal evolution of physics systems by leveraging recent advances from diffusion models. Our method moves the system's current state backward in time step by step by combining an approximate inverse physics simulator and a learned correction function. A central insight of our work is that training the learned correction with a single-step loss is equivalent to a score matching objective, while recursively predicting longer parts of the trajectory during training relates to maximum likelihood training of a corresponding probability flow. We highlight the advantages of our algorithm compared to standard denoising score matching and implicit score matching, as well as fully learned baselines for a wide range of inverse physics problems. The resulting inverse solver has excellent accuracy and temporal stability and, in contrast to other learned inverse solvers, allows for sampling the posterior of the solutions. Code and experiments are available at https://github.com/tum-pbs/SMDP.