Research in graph signal processing (GSP) aims to develop tools for data defined on irregular domains. In this paper, we first provide an overview of core ideas GSP and their connection conventional digital processing, along with a brief historical perspective highlight how concepts recently developed build top prior research other areas. We then summarize recent advances developing basic tools, including methods sampling, filtering, or learning. Next, review progress several application...

In social settings, individuals interact through webs of relationships. Each individual is a node in complex network (or graph) interdependencies and generates data, lots data. We label the data by its source, or formally stated, we index nodes graph. The resulting signals (data indexed nodes) are far removed from time image well ordered samples pixels. DSP, discrete signal processing, provides comprehensive, elegant, efficient methodology to describe, represent, transform, analyze, process,...

Fast changing, increasingly complex, and diverse computing platforms pose central problems in scientific computing: How to achieve, with reasonable effort, portable optimal performance? We present SPIRAL, which considers this problem for the performance-critical domain of linear digital signal processing (DSP) transforms. For a specified transform, SPIRAL automatically generates high-performance code that is tuned given platform. formulates tuning as an optimization exploits domain-specific...

Signals and datasets that arise in physical engineering applications, as well social, genetics, biomolecular, many other domains, are becoming increasingly larger more complex. In contrast to traditional time image signals, data these domains supported by arbitrary graphs. Signal processing on graphs extends concepts techniques from signal indexed generic This paper studies the of low high frequencies graphs, low-, high- band-pass graph signals filters. processing, easily defined because a...

Analysis and processing of very large data sets, or big data, poses a significant challenge. Massive sets are collected studied in numerous domains, from engineering sciences to social networks, biomolecular research, commerce, security. Extracting valuable information requires innovative approaches that efficiently process amounts as well handle and, moreover, utilize their structure. This article discusses paradigm for large-scale analysis based on the discrete signal (DSP) graphs (DSPG)....

Gossip algorithms are attractive for in-network processing in sensor networks because they do not require any specialized routing, there is no bottleneck or single point of failure, and robust to unreliable wireless network conditions. Recently, has been a surge activity the computer science, control, signal processing, information theory communities, developing faster more gossip deriving theoretical performance guarantees. This paper presents an overview recent work area. We describe...

The paper studies average consensus with random topologies (intermittent links) and noisy channels. Consensus noise in the network links leads to bias-variance dilemma-running for long reduces bias of final estimate but increases its variance. We present two different compromises this tradeoff: <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A</i> - xmlns:xlink="http://www.w3.org/1999/xlink">ND</i> algorithm modifies conventional by forcing...

We study distributed optimization problems when N nodes minimize the sum of their individual costs subject to a common vector variable. The are convex, have Lipschitz continuous gradient (with constant L), and bounded gradient. propose two fast algorithms based on centralized Nesterov algorithm establish convergence rates in terms per-node communications K evaluations k. Our first method, Distributed Gradient, achieves O( logK/K) O(logk/k). second with Consensus iterations, assumes at all...

Explainable machine learning offers the potential to provide stakeholders with insights into model behavior by using various methods such as feature importance scores, counterfactual explanations, or influential training data. Yet there is little understanding of how organizations use these in practice. This study explores view and explainability for stakeholder consumption. We find that, currently, majority deployments are not end users affected but rather engineers, who debug itself. There...

The paper studies distributed static parameter (vector) estimation in sensor networks with nonlinear observation models and noisy intersensor communication. It introduces separably estimable that generalize the observability condition linear centralized to estimation. two algorithms models, <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">NU</i> (with its counterpart xmlns:xlink="http://www.w3.org/1999/xlink">LU</i> )...

This paper presents a distributed Kalman filter to estimate the state of sparsely connected, large-scale, n -dimensional, dynamical system monitored by network N sensors. Local filters are implemented on <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</sub> -dimensional subsystems, Lt n, obtained spatially decomposing large-scale system. The is optimal under an Lth order Gauss-Markov approximation centralized filter. We quantify information...

The paper studies the problem of distributed average consensus in sensor networks with quantized data and random link failures. To achieve consensus, dither (small noise) is added to states before quantization. When quantizer range unbounded (countable number levels), stochastic approximation shows that asymptotically achieved probability one mean square a finite variable. We show mean-squared error (mse) can be made arbitrarily small by tuning weight sequence, at cost convergence rate...

We introduce the task of Visual Dialog, which requires an AI agent to hold a meaningful dialog with humans in natural, conversational language about visual content. Specifically, given image, history, and question has ground infer context from answer accurately. Dialog is disentangled enough specific downstream so as serve general test machine intelligence, while being grounded vision allow objective evaluation individual responses benchmark progress. develop novel two-person chat...

The paper introduces DILOC, a distributed, iterative algorithm to locate M sensors (with unknown locations) in R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</sup> , m ges 1, with respect minimal number of + 1 anchors known locations. and anchors, nodes the network, exchange data their neighbors only; no centralized processing or communication occurs, nor is there fusion center compute sensors' DILOC uses barycentric coordinates node its...

We introduce the first goal-driven training for visual question answering and dialog agents. Specifically, we pose a cooperative `image guessing' game between two agents - Q-BOT A-BOT- who communicate in natural language so that can select an unseen image from lineup of images. use deep reinforcement learning (RL) to learn policies these end-to-end pixels multi-agent multi-round reward.,,We demonstrate experimental results.,,First, as `sanity check' demonstration pure RL (from scratch), show...

This paper proposes modeling the rapidly evolving energy systems as cyber-based physical systems. It introduces a novel dynamical model whose mathematical description depends on cyber technologies supporting system. discusses how such can be used to ensure full observability through cooperative information exchange among its components; this is achieved without requiring local of system components. also shows cyber-physical develop interactive protocols between controllers embedded within...

We consider the problem of signal recovery on graphs as model data with complex structure signals a graph. Graph implies one or multiple smooth graph from noisy, corrupted, incomplete measurements. propose and formulate corresponding optimization problem. provide general solution by using alternating direction methods multipliers. next show how inpainting, matrix completion, robust principal component analysis, anomaly detection all relate to recovery, specific solutions theoretical...

This paper considers gossip distributed estimation of a (static) random field (a.k.a., large-scale unknown parameter vector) observed by sparsely interconnected sensors, each which only observes small fraction the field. We consider linear estimators whose structure combines information flow among sensors (the consensus term resulting from local gossiping exchange when they are able to communicate) and gathering measured sensing or innovations term). leads mixed time scale algorithms-one...

This paper reviews signal processing research for applications in the future electric power grid, commonly referred to as smart grid. Generally, it is expected that grid of would differ from current system by increased integration distributed generation, storage, demand response, electronics, and communications sensing technologies. The consequence physical structure becomes significantly more distributed. existing centralized control not suitable any operate such a highly system. Hence,...

Monitoring electricity consumption in the home is an important way to help reduce energy usage and Non-Intrusive Load (NILM) techniques are a promising approach obtain estimates of electrical power individual appliances from aggregate measurements voltage and/or current distribution system. In this paper, we discuss event detection algorithms used NILM literature propose new metrics for evaluating them. particular, introduce that incorporate information contained signal instead strict rates....

Spectral graph convolutional neural networks (CNNs) require approximation to the convolution alleviate computational complexity, resulting in performance loss. This paper proposes topology adaptive network (TAGCN), a novel defined vertex domain. We provide systematic way design set of fixed-size learnable filters perform convolutions on graphs. The topologies these are when they scan convolution. TAGCN not only inherits properties CNN for grid-structured data, but it is also consistent with...

In this paper, we develop deep spatio-temporal neural networks to sequentially count vehicles from low quality videos captured by city cameras (citycams). Citycam have resolution, frame rate, high occlusion and large perspective, making most existing methods lose their efficacy. To overcome limitations of incorporate the temporal information traffic video, design a novel FCN-rLSTM network jointly estimate vehicle density connecting fully convolutional (FCN) with long short term memory (LSTM)...