The pressing need to modernize the aging power grid has culminated into the smart grid vision, which entails the widespread use of state-of-the-art sensing, control, and communication technologies. The deployment of these smart technologies calls for novel grid monitoring and optimization techniques. This tutorial focuses on how current research challenges in power grid monitoring and optimization can be addressed through signal processing, communications, and networking toolboxes. After an overview of fundamental power engineering concepts, a wide range of modern research topics will be presented, including power system state estimation, phasor measurement units, line outage identification, price and load forecasting, economic operation of power systems, demand response, electric vehicles, and renewable energy management.

Microphone array algorithms emerged in the early 1990s as viable solutions to speech enhancement problems. Although more than 20 years have elapsed, the adaptation of beamforming methods to speech enhancement problems remains an open issue. These difficulties may be attributed to the wide-band and nonstationary characteristics of a speech signal and to the very long, typically time-varying, room impulse responses (RIRs) relating moving speakers and microphones in acoustic enclosures.

This tutorial will begin with an introduction to speech enhancement and speaker separation algorithms for microphone arrays. We will explore some of the leading criteria for spatial array design in the context of speech enhancement using the spatial properties of the acoustic environment. We will discuss the multi-channel Wiener filter (MWF) and its variant the speech distortion weighted MWF (SDW-MWF); the minimum variance distortionless response (MVDR) beamformer; and the linearly constrained minimum variance (LCMV) beamformer. The relations between the various criteria will be explored. We will then discuss an adaptive implementation of the MVDR and LCMV criteria, namely the generalized sidelobe canceller (GSC), and its importance in speech processing tasks.

We will then turn to the introduction of an emerging field of research, i.e., distributed algorithm design for ad hoc microphone arrays and wireless acoustic sensor networks (WASNs). A WASN is comprised of multiple (often battery powered) microphone nodes, each of which is equipped with one or more microphones, a signal processing unit and a wireless communication module. The large spatial distribution of such microphone constellations yields a large amount of spatial information, and it increases the probability that a subset of the microphones is close to a relevant sound source. However, new challenges arise in these new distributed architectures, in particular when data centralization is not possible (either due to the lack of a dedicated central processing device or due to overly demanding transmission/processing requirements). One then has to rely on distributed (in-network) processing, where nodes only share compressed/fused microphone signals with each other. In this part of the tutorial we will introduce some novel algorithms for solving the speech enhancement problem in a distributed fashion, while limiting the amount of data that is shared between the nodes. We will also explore theoretical performance bounds of such distributed microphone arrays and address some practical issues, e.g. synchronization and efficient adaptation to changing environmental conditions.

In particular we will discuss the DANSE-family of algorithms (Distributed Adaptive Node-specific Signal Estimation), which allow for a distributed implementation of several of the speech enhancement algorithms that were introduced in the first part of this tutorial (including SDW-MWF and LCMV beamforming). We first describe their implementation in fully-connected networks in which a signal broadcast by any node can be received by all other nodes in the network. We will then extend these results towards more challenging partially-connected topologies, where a node can only communicate with the nodes in its close neighborhood. Finally, we will also discuss the distributed GSC algorithm and its variants.

The tutorial will be accompanied, when applicable, by sound file demonstrations.

As media content is prominent in our everyday life, efficient content-based management of this massive source of information through its full life cycle (from creation to consumption by the user) defines the general scope of this tutorial. Multimedia content features (also called descriptors) play a central role in many computer vision and image processing applications. Features are various types of information extracted from the content and represent some of its characteristics or signatures. However, especially the (low-level) features, which can be extracted automatically usually lack the discrimination power needed for accurate processing especially in the case of a large and varied media content data reserves. Therefore, we shall present a recent evolutionary technique, the Multi-Dimensional Particle Swarm Optimization (MD PSO) to synthesize highly discriminative features using a novel evolutionary feature synthesis framework. We shall further show how MD PSO can effectively be used in many data mining and pattern recognition applications where the optimal solution space dimension is unknown as long as the potential solution is encoded in the swarm particles accordingly and a proper fitness function is used. Furthermore, for classifying and indexing a large media content data reserve, the key questions are: 1) how to select certain features so as to achieve the highest discrimination over certain classes, 2) how to combine them in the most effective way, 3) which distance metric to apply, 4) how to find the optimal classifier configuration for the classification problem at hand, and 5) how to scale/adapt the classifier if a large number of classes/features are present and finally, 6) how to train the classifier efficiently to maximize the classification accuracy. In this tutorial novel classifier topologies based on recent evolutionary techniques will be introduced to address these questions.

The following topics will particularly be focused on during the tutorial along with the recent techniques and applications.

1) Evolutionary Algorithms and Multi-Dimensional Particle Swarm Optimization

2) Data Clustering by MD-PSO

3) Multi-dimensional Evolutionary Feature Synthesis

4) Collective Network of Evolutionary Binary Classifiers

Natural sounds in real-world environments are typically composed of multiple source signals, such as speech and noise, or multiple instruments in a music signal. Moreover, each of these source signals may be composed of parts, for example multiple notes played by a musical instrument. Compositional models, including those based on non-negative matrix factorization (NMF), explicitly consider the fact that sound components largely combine constructively in the composition of more complex sounds. The use of compositional models has yielded state-of-the-art results in many audio processing tasks, such as sound source separation, music content analysis and noise-robust automatic speech recognition. These methods are also closely related to the sparse representations popularized in Compressed Sensing.

In this tutorial we will discuss both basics concepts, such as feature representations, dictionary learning and algorithms, as well as more advanced topics such as regularisation with sparsity, probabilistic formulations such as latent variable models, convolutive models and tensor factorization models. From the start, every topic will illustrated with representative application examples, ranging from automatic music transcription to noise-robust automatic speech recognition. The examples will be done using MATLAB and the codes used will be made available to the participants. The tutorial is targeted at students and researchers who have basic knowledge of signal processing, but not necessarily any previous experience with compositional models.

Compressed sensing has been a very active area of research in the past few years. Recently, focus has shifted to Bayesian based approaches that are able to perform sparse signal recovery at much lower complexity while invoking constraint and/or a priori information about the data. While Bayesian approaches have their advantages these must have access to a priori statistics. Usually, these statistics are unknown and are often difficult or even impossible to predict. An effective workaround is to assume some kind of distribution which is typically considered to be Gaussian. Seemingly attractive, this assumption necessitates the estimation of the associated parameters; which could be hard if not impossible. The tutorial will focus on this aspect of Bayesian recovery and present a framework to address the challenges mentioned above. The tutorial will start by giving an overview of Bayesian approaches as well as graph-based approaches for sparse signal recovery. It will then move to show that the proposed framework allows construction of algorithms for sparse signal recovery which are Bayesian at their core but at the same time are agnostic to the distribution of the unknown sparse signal components. The tutorial will then move to show how different types of problem structures could be utilized to assist in sparse signal recovery. It will be shown especially that how the intrinsic features of the proposed framework could be exploited to build simple and efficient structure-aware sparse signal recovery algorithms. The discussion will illuminate through several examples from signal processing, communication, seismic deconvolution, and health monitoring.

Due to their wide applicability, sparse and low-rank modelling tools have quickly become some of the most important techniques for today's researchers in signal, image, and video processing as well as machine learning. Application areas in which sparse and low-rank modelling tools have been applied span a wide range of topics in these fields including: compressive sensing, image inpainting, super-resolution, denoising, foreground/background separation, image and video compression, object recognition and classification, l-1 regularized regression, least absolute deviations, Robust Principle Component Analysis, as well as Collaborative Filtering.

However while sparse and low-rank models themselves are typically fairly straightforward to grasp in applications, often times the optimization machinery required to make use of these models can be unfamiliar to students and researchers with a more traditional electrical, biomedical, statistics, or computer engineering/science background. Therefore a major distinctive feature of this tutorial course will be a practical focus on connecting fundamental concepts in convex optimization with their natural (and cutting edge) extensions for solving sparse and low-rank problems. No previous exposure to nonlinear programming is required for this tutorial. We will review the fundamental concepts as needed throughout the course.

In this tutorial, based on a book manuscript currently under development, we will spend the first third of the class introducing sparse and low-rank models in the context of various applications. After briefly touching on recovery guarantees we will then spend the remainder of the class accessibly explaining the cutting edge methods used to solve sparse and low-rank recovery problems. This coverage of optimization methods will include: a discussion of greedy methods for sparse recovery, an overview of essential concepts from nonlinear programming as well as classic algorithms for smooth constrained optimization, smooth reformulation techniques for sparse and low-rank problems, accelerated proximal gradient techniques, and sequential primal-dual algorithms.

The tutorial aims at providing an overview of data fusion techniques necessary for representing, modeling and automatically interpreting complex interaction situations occurring in interactive cognitive environments starting from observations provided by a distributed network of embedded systems interacting with the observed environment. A Bayesian oriented viewpoint is assumed in this tutorial.

The main objective is to describe a common framework based on distributed data fusion, for multi-level joint tracking and classification of objects and interactions. The distributed data fusion theory is applied to the problem of recognizing multiple interacting objects from multi-camera video sequences. In particular, the attention will be focused on those actions involving more than one subject, or a subject and elements in the environment, i.e. interactions. This approach is quite novel in the domain of human activity recognition since most methods rely just on the analysis of body motion of single subjects but they don't take into account the context where these situations take place. To this goal, video processing algorithms aiming at improving tracking performances will also be described, both for detection of subjects and body parts. Collaborative trackers, being designed to exchange information between each other in order to produce more reliable models, are a promising approach to tackle typical issues due to occlusions happening during interaction among multiple subjects.

The tutorial will demonstrate how the same distributed data fusion theories can be successfully applied at a higher level for classifying interactions between subjects on a larger scale. Object classification is a fundamental step in automatic video tracking which allows improved tracking and a more accurate description of events. The hypothesis for object classification is based on feature selection method which gives a good subset of features while the machine learns the classification task and use these selected features for object classification.

The tutorial is divided into the following parts:

• 1) Introduction on Interactive and Cognitive Environments (C.Regazzoni)
• 2) Data fusion architectures and models (L.Marcenaro)
• 3) Probabilistic Graphical Models and Dynamic Bayesian Networks for interactions modeling (C.Regazzoni)
• 4) Examples and case studies (L.Marcenaro)

The issue of energy efficiency, spectral efficiency, and resource optimization has attracted a huge interest by the information and telecommunication technology (ICT) community in the last decade, as witnessed by the vast literature available in this topic. In the field of wireless communications, efficiency can be achieved by operating at all different layers of the network, spanning from system architectures and protocols, to transmission techniques, and to opportunistic spectrum sharing, just to mention a few notable examples. Design and optimization methods of such networks are benefiting from the adoption of sophisticated optimization techniques at large. Game theory, traditionally studied and applied in the areas of economics, political science, biology and sociology, has recently emerged as an effective framework for the design of a wireless network, since it provides analytical tools to predict the outcome of interactions among rational entities with conflicting interests, like communication nodes. Interactions of the users in a wireless network for communications or sensing can be modeled as a game in which the user terminals are the players in the game competing for network resources (i.e., bandwidth and/or energy), which are typically scarce. This tutorial provides an overview of the relevant applications of game theory in wireless networks, focusing on state-of-the-art techniques for resource allocation. The very basics concepts of game theory are introduced by means of many simple examples, and special emphasis is put on how to translate a real-world problem into an analytical game model.