Predictive Analytics and Decision Support for Heart Failure patients

Thesis (Master's)--University of Washington, 2016-08

Dynamic set pieces for CAMBADA team

Robotics is an emergent branch of engineering that involves the conception, manufacture, and control of robots. It is a multidisciplinary field that combines electronics, design, computer science, artificial intelligence, mechanics and nanotechnology. Its evolution results in machines that are able to perform tasks with some level of complexity. Multi-agent systems is a researching topic within robotics, thus they allow the solving of higher complexity problems, through the execution of simple routines. Robotic soccer allows the study and development of robotics and multiagent systems, as the agents have to work together as a team, having in consideration most problems found in our quotidian, as for example adaptation to a highly dynamic environment as it is the one of a soccer game. CAMBADA is the robotic soccer team belonging to the group of research IRIS from IEETA, composed by teachers, researchers and students of the University of Aveiro, which annually has as main objective the participation in the RoboCup, in the Middle Size League. The purpose of this work is to improve the coordination in set pieces situations. This thesis introduces a new behavior and the adaptation of the already existing ones in the offensive situation, as well as the proposal of a new positioning method in defensive situations. The developed work was incorporated within the competition software of the robots. Which allows the presentation, in this dissertation, of the experimental results obtained, through simulation software as well as through the physical robots on the laboratory.

RICH AND EFFICIENT VISUAL DATA REPRESENTATION

Increasing the size of training data in many computer vision tasks has shown to be very effective. Using large scale image datasets (e.g. ImageNet) with simple learning techniques (e.g. linear classifiers) one can achieve state-of-the-art performance in object recognition compared to sophisticated learning techniques on smaller image sets. Semantic search on visual data has become very popular. There are billions of images on the internet and the number is increasing every day. Dealing with large scale image sets is intense per se. They take a significant amount of memory that makes it impossible to process the images with complex algorithms on single CPU machines. Finding an efficient image representation can be a key to attack this problem. A representation being efficient is not enough for image understanding. It should be comprehensive and rich in carrying semantic information. In this proposal we develop an approach to computing binary codes that provide a rich and efficient image representation. We demonstrate several tasks in which binary features can be very effective. We show how binary features can speed up large scale image classification. We present learning techniques to learn the binary features from supervised image set (With different types of semantic supervision; class labels, textual descriptions). We propose several problems that are very important in finding and using efficient image representation.

Self-Organizing Map Neural Architectures Based on Limit Cycle Attractors

Recent efforts to develop large-scale neural architectures have paid relatively little attention to the use of self-organizing maps (SOMs). Part of the reason is that most conventional SOMs use a static encoding representation: Each input is typically represented by the fixed activation of a single node in the map layer. This not only carries information in an inefficient and unreliable way that impedes building robust multi-SOM neural architectures, but it is also inconsistent with rhythmic oscillations in biological neural networks. Here I develop and study an alternative encoding scheme that instead uses limit cycle attractors of multi-focal activity patterns to represent input patterns/sequences. Such a fundamental change in representation raises several questions: Can this be done effectively and reliably? If so, will map formation still occur? What properties would limit cycle SOMs exhibit? Could multiple such SOMs interact effectively? Could robust architectures based on such SOMs be built for practical applications? The principal results of examining these questions are as follows. First, conditions are established for limit cycle attractors to emerge in a SOM through self-organization when encoding both static and temporal sequence inputs. It is found that under appropriate conditions a set of learned limit cycles are stable, unique, and preserve input relationships. In spite of the continually changing activity in a limit cycle SOM, map formation continues to occur reliably. Next, associations between limit cycles in different SOMs are learned. It is shown that limit cycles in one SOM can be successfully retrieved by another SOM’s limit cycle activity. Control timings can be set quite arbitrarily during both training and activation. Importantly, the learned associations generalize to new inputs that have never been seen during training. Finally, a complete neural architecture based on multiple limit cycle SOMs is presented for robotic arm control. This architecture combines open-loop and closed-loop methods to achieve high accuracy and fast movements through smooth trajectories. The architecture is robust in that disrupting or damaging the system in a variety of ways does not completely destroy the system. I conclude that limit cycle SOMs have great potentials for use in constructing robust neural architectures.

Spiking neural network-based target tracking control for autonomous mobile robots

In this paper, a target tracking controller based on spiking neural network is proposed for autonomous robots. This controller encodes the preprocessed environmental and target information provided by CCD cameras, encoders and ultrasonic sensors into spike trains, which are integrated by a three-layer spiking neural network (SNN). The outputs of SNN are generated based on the competition between the forward/backward neuron pair corresponding to each motor, with the weights evolved by the Hebbian learning. The application to target tracking of a mobile robot in unknown environment verifies the validity of the proposed controller.

Acoustic and device feature fusion for load recognition

Appliance-specific Load Monitoring (LM) provides a possible solution to the problem of energy conservation which is becoming increasingly challenging, due to growing energy demands within offices and residential spaces. It is essential to perform automatic appliance recognition and monitoring for optimal resource utilization. In this paper, we study the use of non-intrusive LM methods that rely on steady-state appliance signatures for classifying most commonly used office appliances, while demonstrating their limitation in terms of accurately discerning the low-power devices due to overlapping load signatures. We propose a multi-layer decision architecture that makes use of audio features derived from device sounds and fuse it with load signatures acquired from energy meter. For the recognition of device sounds, we perform feature set selection by evaluating the combination of time-domain and FFT-based audio features on the state of the art machine learning algorithms. Further, we demonstrate that our proposed feature set which is a concatenation of device audio feature and load signature significantly improves the device recognition accuracy in comparison to the use of steady-state load signatures only.

New inequality-based approach to passivity analysis of neural networks with interval time-varying delay

This paper is concerned with the problem of passivity analysis of neural networks with an interval time-varying delay. Unlike existing results in the literature, the time-delay considered in this paper is subjected to interval time-varying without any restriction on the rate of change. Based on novel refined Jensen inequalities and by constructing an improved Lyapunov-Krasovskii functional (LKF), which fully utilizes information of the neuron activation functions, new delay-dependent conditions that ensure the passivity of the network are derived in terms of tractable linear matrix inequalities (LMIs) which can be effectively solved by various computational tools. The effectiveness and improvement over existing results of the proposed method in this paper are illustrated through numerical examples.

Improving the quality of load forecasts using smart meter data

For the operator of a power system, having an accurate forecast of the day-ahead load is imperative in order to guaranty the reliability of supply and also to minimize generation costs and pollution. Furthermore, in a restructured power system, other parties, like utility companies, large consumers and in some cases even ordinary consumers, can benefit from a higher quality demand forecast. In this paper, the application of smart meter data for producing more accurate load forecasts has been discussed. First an ordinary neural network model is used to generate a forecast for the total load of a number of consumers. The results of this step are used as a benchmark for comparison with the forecast results of a more sophisticated method. In this new method, using wavelet decomposition and a clustering technique called interactive k-means, the consumers are divided into a number of clusters. Then for each cluster an individual neural network is trained. Consequently, by adding the outputs of all of the neural networks, a forecast for the total load is generated. A comparison between the forecast using a single model and the forecast generated by the proposed method, proves that smart meter data can be used to significantly improve the quality of load forecast.

Introduction : special issue of selected papers from ACML 2014

We are delighted to present this special issue of Machine Learning Journal with selected papers from the Sixth Asian Conference on Machine Learning (ACML 2014) held in Nha Trang City, Vietnam from 26 to 28 November 2014. ACML aims at providing a leading international forum for researchers in machine learning and related fields to share their new ideas and achievements. While located in Asia, the conference has a wide visibility to the international community. ACML was the first machine learning conference with two cycles of submissions with a strict double-blind review process, and this tradition continues. ACML 2014 received 80 submissions from 20 countries across Asia, Australasia, Europe and North America. Each paper was assigned with two meta-reviewers and at least four reviewers. In the end, 25 papers were accepted into the main program, accounting for an acceptance rate of 31.25 % (Phung and Li 2014).

Identifying social influence in complex networks: a novel conductance eigenvector centrality model

Identifying influential peers is an important issue for business to promote commercial strategies in social networks. This paper proposes a conductance eigenvector centrality (CEC) model to measure peer influence in the complex social network. The CEC model considers the social network as a conductance network and constructs methods to calculate the conductance matrix of the network. By a novel random walk mechanism, the CEC model obtains stable CEC values which measure the peer influence in the network. The experiments show that the CEC model can achieve robust performance in identifying peer influence. It outperforms the benchmark algorithms and obtains excellent outcomes when the network has high clustering coefficient.

An embedding scheme for detecting anomalous block structured graphs

Graph-based anomaly detection plays a vital role in various application domains such as network intrusion detection, social network analysis and road traffic monitoring. Although these evolving networks impose a curse of dimensionality on the learning models, they usually contain structural properties that anomaly detection schemes can exploit. The major challenge is finding a feature extraction technique that preserves graph structure while balancing the accuracy of the model against its scalability. We propose the use of a scalable technique known as random projection as a method for structure aware embedding, which extracts relational properties of the network, and present an analytical proof of this claim. We also analyze the effect of embedding on the accuracy of one-class support vector machines for anomaly detection on real and synthetic datasets. We demonstrate that the embedding can be effective in terms of scalability without detrimental influence on the accuracy of the learned model.