We consider an energy-harvesting communication system where a transmitter powered by exogenous energy arrival process and equipped with finite battery of size B <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">max</sub> communicates over discrete-time AWGN channel. first concentrate on simple Bernoulli at each time step, either packet E is harvested probability p, or no all, independent the other steps. provide near optimal control policy...

Federated learning is a distributed paradigm that aims at training models using samples across multiple users in network while keeping the on users' devices with aim of efficiency and protecting privacy. In such settings, data often statistically heterogeneous manifests various distribution shifts users, which degrades performance learnt model. The primary goal this paper to develop robust federated algorithm achieves satisfactory against samples. To achieve goal, we first consider...

Diffusion-based Neural Combinatorial Optimization (NCO) has demonstrated effectiveness in solving NP-complete (NPC) problems by learning discrete diffusion models for solution generation, eliminating hand-crafted domain knowledge. Despite their success, existing NCO methods face significant challenges both cross-scale and cross-problem generalization, high training costs compared to traditional solvers. While recent studies have introduced training-free guidance approaches that leverage...

Deep neural networks (DNNs) have set benchmarks on a wide array of supervised learning tasks. Trained DNNs, however, often lack robustness to minor adversarial perturbations the input, which undermines their true practicality. Recent works increased DNNs by fitting using adversarially-perturbed training samples, but improved performance can still be far below seen in non-adversarial settings. A significant portion this gap attributed decrease generalization due training. In work, we extend...

Given a task of predicting $Y$ from $X$, loss function $L$, and set probability distributions $Γ$ on $(X,Y)$, what is the optimal decision rule minimizing worst-case expected over $Γ$? In this paper, we address question by introducing generalization principle maximum entropy. Applying to sets with marginal $X$ constrained be empirical data, develop general minimax approach for supervised learning problems. While some functions such as squared-error log loss, rederives well-knwon regression...

.Generative adversarial networks (GANs) learn the distribution of observed samples through a zero-sum game between two machine players, generator and discriminator. While GANs achieve great success in learning complex image, sound, text data, they perform suboptimally multimodal distribution-learning benchmarks such as Gaussian mixture models (GMMs). In this paper, we propose Generative Adversarial Training for Mixture Models (GAT-GMM), minimax GAN framework GMMs. Motivated by optimal...

Generative Adversarial Networks (GANs) have become a popular method to learn probability model from data. In this paper, we provide an understanding of basic issues surrounding GANs including their formulation, generalization and stability on simple LQG benchmark where the generator is Linear, discriminator Quadratic data has high-dimensional Gaussian distribution. Even in benchmark, GAN problem not been well-understood as observe that existing state-of-the-art architectures may fail proper...

Generative adversarial networks (GANs) represent a zero-sum game between two machine players, generator and discriminator, designed to learn the distribution of data. While GANs have achieved state-of-the-art performance in several benchmark learning tasks, GAN minimax optimization still poses great theoretical empirical challenges. trained using first-order methods commonly fail converge stable solution where players cannot improve their objective, i.e., Nash equilibrium underlying game....

Federated learning is a distributed machine paradigm, which aims to train model using the local data of many clients. A key challenge in federated that samples across clients may not be identically distributed. To address this challenge, personalized with goal tailoring learned distribution every individual client has been proposed. In paper, we focus on problem and propose novel Learning scheme based Optimal Transport (FedOT) as algorithm learns optimal transport maps for transferring...

Generative Adversarial Networks (GANs) have become a popular method to learn probability model from data. In this paper, we aim provide an understanding of some the basic issues surrounding GANs including their formulation, generalization and stability on simple benchmark where data has high-dimensional Gaussian distribution. Even in benchmark, GAN problem not been well-understood as observe that existing state-of-the-art architectures may fail proper generative distribution owing (1) (i.e.,...

Generative adversarial network (GAN) is a minimax game between generator mimicking the true model and discriminator distinguishing samples produced by from real training samples. Given an unconstrained able to approximate any function, this reduces finding generative minimizing divergence measure, e.g. Jensen-Shannon (JS) divergence, data distribution. However, in practice constrained be smaller class $\mathcal{F}$ such as neural nets. Then, natural question how minimization interpretation...

Interpreting neural network classifiers using gradient-based saliency maps has been extensively studied in the deep learning literature. While existing algorithms manage to achieve satisfactory performance application standard image recognition datasets, recent works demonstrate vulnerability of widely-used interpretation schemes norm-bounded perturbations adversarially designed for every individual input sample. However, such adversarial are commonly knowledge an sample, and hence perform...

Deep generative models dominate the existing literature in layout pattern generation. However, leaving guarantee of legality to an inexplicable neural network could be problematic several applications. In this paper, we propose DiffPattern generate reliable patterns. introduces a novel diverse topology generation method via discrete diffusion model with compute-efficiently lossless representation. Then white-box assessment is utilized legal patterns given desired design rules. Our...

Given low order moment information over the random variables X = (X <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> , xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> …, xmlns:xlink="http://www.w3.org/1999/xlink">p</inf> ) and Y, what distribution minimizes Hirschfeld-Gebelein-Rényi (HGR) maximal correlation coefficient between while remains faithful to given moments? The answer this question is important especially in fit models (X,...

We consider an energy-harvesting communication system where a transmitter powered by exogenous energy arrival process and equipped with finite battery of size $B_{max}$ communicates over discrete-time AWGN channel. first concentrate on simple Bernoulli at each time step, either packet $E$ is harvested probability $p$, or no all, independent the other steps. provide near optimal control policy approximation to information-theoretic capacity this Our approximations for both problems are...

The success of deep neural networks stems from their ability to generalize well on real data; however, et al. have observed that can easily overfit randomly-generated labels. This observation highlights the following question: why do gradient methods succeed in finding generalizable solutions for while there exist with poor generalization behavior? In this work, we use a Fourier-based approach study properties gradient-based over 2-layer band-limited activation functions. Our results...