1731 research outputs found
Sort by
Personal Healthcare Agents for Monitoring and Predicting Stress and Hypertension from Biosignals
No full textWe live in exciting times. The fast paced growth in mobile computers has put powerful computational devices in the palm of our hands. Blazing fast connectivity has made human-human, human-machine, and machine-machine communication effortless. Wearable devices and the internet of things have made monitoring every aspect of our lives easier.
This has given rise to the domain of quantified self where we can continuous record and quantify the various signals generated in everyday life. Sensors on smartphones can continuously record our location and motion profile.
Sensors on wearable devices can track changes in our bodies’ physiological responses. This monitoring also has the capability to revolutionise the health care domain by creating more informed and involved patients. This has the potential to shift care-management from a physician-centric approach to a patient-centric approach allowing individuals to create more empowered patients and individuals who are in better control of their health. However, the data deluge from all these sources can sometimes be overwhelming. There is a need for intelligent technology that can help us navigate the data and take informed decisions.
The goal of this work is to develop a mobile, personal intelligent agent platform that can become a digital companion to live with the user. It can monitor the covert and overt signal streams of the user, identify activity and stress levels to help the users’ make healthy choices regarding their lives. This thesis particularly targets patients suffering from or at-risk of essential hypertension since its a difficult condition to detect and manage.
This thesis delivers the following contributions: 1) An intelligent personal agent platform for on-the-go continuous monitoring of covert and overt signals. 2) A machine learning algorithm for accurate recognition of
activities using smartphone signals recorded from in-the-wild scenarios.
3) A machine learning pipeline to combine various physiological signal streams, motion profiles, and user annotations for on-the-go stress recognition. 4) We design and train a complete signal processing and classification system for hypertension prediction. 5) Through a small pilot study we demonstrate that this system can distinguish between hypertensive and normotensive subjects with high accuracy
Application Interference in Multi-Core Architectures: Analysis and Effects
No full textClouds are an irreplaceable part of many business applications. They provide tremendous flexibility and gave birth for many related technologies – Software as a Service (SaaS) and the like. One of the biggest powers of clouds is load redistribution for scaling up and down on demand. This
helps dealing with varying loads, increasing resource utilization and cutting down electricity bills while maintaining reasonable performance isolation.
The last one is of our particular interest. Most cloud systems are accounted and billed not by useful throughput,
but by resource usage. For example, a cloud provider may charge according to cumulative CPU time and/or average memory footprint. But this does not guarantee that the application realized its full performance potential
because CPU and memory are shared resources. As a result, if there are many other applications it could experience frequent execution stalls due to contention on memory bus or cache pressure. The problem is more and more pronounced because modern hardware rapidly increases in density
leading to more applications are co-located. The performance degradation caused by co-location of applications is called application interference. In this work we study in-depth reasons of interference as well as ways to mitigate it. The first part of the work is devoted to interference analysis
and introduces a simple yet powerful empirical model of CPU performance that takes interference into account. The model is based on empirical observations and build up from extrapolation of a two-task (trivial) case. In the following part we present a method of ranking of virtual machines
according to their average interference. The method is based on analysis of performance counters. We first launch a set of very diverse benchmark programs (to be representative for wide range of programs) one-by-one together with all sorts of performance counters. This gives us their “ideal”
(isolated) performances. Then we run them in pairs to see the level of interference they create to each other. Once this is done, for each benchmark we calculate average interference. Finally we calculate the correlation
between the average interference and performance counters. The counters with the biggest correlation are to be used as interference estimators. The final part deals with measuring interference in production environment with affordable overhead. The technique is based on short (in the
order of milliseconds) freezes of virtual machines to see how they affect other VMs (hence the name of method – Freeze’nSense). By comparing the performance of the VM when other VMs active and when they frozen it is possible to conclude how much it looses in speed because of sharing
hardware with other applications
Fusion processes in low-energy collisions of weakly bound nuclei
No full textThe present thesis deals with the study of nuclear reactions of weakly-bound few-body nuclei with stable targets at energies around the Coulomb barrier. A weakly-bound nucleus is characterized by having low binding energy and thus large probability to undergo a breakup during the interaction with another nucleus. The effect of breakup on other reactions such as fusion is still not well understood and represents an interesting issue to study in detail. Among weakly-bound nuclei, there are some which are particularly interesting to study because of their peculiar structure and the key role they play in astrophysics: the halo nuclei
Time-optimal control problems in the space of measures
No full textThe thesis deals with the study of a natural extension of classical finite-dimensional time-optimal control problem to the space of positive Borel measures. This approach has two main motivations: to model real-life situations in which the knowledge of the initial state is only probabilistic, and to model the statistical distribution of a huge number of agents for applications in multi-agent systems.
We deal with a deterministic dynamics and treat the problem first in a mass-preserving setting: we give a definition of generalized target, its properties, admissible trajectories and generalized minimum time function, we prove a Dynamic Programming Principle, attainability results, regularity results and an Hamilton-Jacobi-Bellman equation solved in a suitable viscosity sense by the generalized minimum time function, and finally we study the definition of an object intended to reflect the classical Lie bracket but in a measure-theoretic setting.
We also treat a case with mass loss thought for modelling the situation in which we are interested in the study of an averaged cost functional and a strongly invariant target set. Also more general cost functionals are analysed which takes into account microscopical and macroscopical effects, and we prove sufficient conditions ensuring their lower semicontinuity and a dynamic programming principle in a general formulation
From predictive to interactive multimodal language learning
No full textThe way humans learn the meaning of words is a fundamental question in many different disciplines and, from a computational perspective, an answer to this question
could lead to important advances in artificial intelligence. While the details of the learning process are still an open question, what we do know is that humans make use of the very rich perceptual input present in the communicative setups in which learning takes place. In this work, we will present three models of human learning from naturalistic
multi-modal input. We will start by introducing a model that assumes a purely predictive learner existing in a non-communicative setup and show that such a computational learner when tested displays comparable learning behaviour to human learners on a novel word learning setup. We will then relax some of the learning assumptions and present a model that, instead of exposing the computational learner to a passive environment, such as the text corpora traditionally used in semantic learning experiments, it exposes the learner to communicative episodes, simulated in our experiments by corpora capturing multi-modal interactions between children and their caregivers, allowing the learner to make use of information beyond words and passive percept during learning. Finally, we will present on-going work towards interactive learning between two agents
Formal failure analyses for effective fault management: an aerospace perspective
No full textThe possibility of failures is a reality that all modern complex engineering systems need to deal with. In this dissertation we consider two techniques to analyze the nature and impact of faults on system dynamics, which is fundamental to reliably manage them. Timed failure propagation analysis studies how and how fast faults propagate through physical and logical parts of a system. We develop formal techniques to validate and automatically generate representations of such behavior from a more detailed model of the system under analysis.
Diagnosability analysis studies the impact of faults on observable parameters and tries to understand whether the presence of faults can be inferred from the observations within a useful time frame. We extend a recently developed framework for specifying diagnosis requirements, develop efficient algorithms to assess diagnosability under a fixed set of observables, and propose an automated technique to select optimal subsets of observables. The techniques have been implemented and evaluated on realistic models and case studies developed in collaboration with engineers from the European Space Agency, demonstrating the practicality of the contributions
Exploiting spatial and spectral information for audio source separation and speaker diarization
No full textThe goal of multichannel audio source separation is to produce high quality separated audio signals, observing mixtures of these signals. The difficulty of tackling the problem comes from not only the source propagation through noisy and echoing environments, but also overlapped source signals. Among the different research directions pursued around this problem, the adoption of probabilistic and advanced modeling aims at exploiting the diversity of multichannel propagation, and the redundancy of source signals. Moreover, prior information about the environments or the signals is helpful to improve the quality and to accelerate the separation. In this thesis, we propose methods to increase the effectiveness of model-based audio source separation methods by exploiting prior information applying spectral and sparse modeling theories. The work is divided into two main parts. In the first part, spectral modeling based on Nonnegative Matrix Factorization is adopted to represent the source signals. The parameters of Gaussian model-based source separation are estimated in sense of Maximum-Likelihood using a Generalized Expectation-Maximization algorithm by applying supervised Nonnegative Matrix and Tensor Factorization, given spectral descriptions of the source signals. Three modalities of making the descriptions available are addressed, i.e. the descriptions are on-line trained during the separation, pre-trained and made directly available, or pre-trained and made indirectly available. In the latter, a detection method is proposed in order to identify the descriptions best representing the signals in the mixtures. In the second part, sparse modeling is adopted to represent the propagation environments. Spatial descriptions of the environments, either deterministic or probabilistic, are pre-trained and made indirectly available. A detection method is proposed in order to identify the deterministic descriptions best representing the environments. The detected descriptions are then used to perform source separation by minimizing a non-convex -norm function. For speaker diarization where the task is to determine ``who spoke when" in real meetings, a Watson mixture model is optimized using an Expectation-Maximization algorithm in order to detect the probabilistic descriptions, best representing the environments, and to estimate the temporal activity of each source. The performance of the proposed methods is experimentally evaluated using different datasets, between simulated and live-recorded. The elaborated results show the superiority of the proposed methods over recently developed methods used as baselines
Observation of the Kibble-Zurek mechanism in a bosonic gas
No full textWhen a second-order phase transition is crossed at finite speed, domains with independent order parameters can appear in the system, with the consequent formation of defects at the domain boundaries. The Kibble-Zurek theory provides a description for this universal phenomenon, which applies to many different systems in nature, and it predicts a power-law dependence of the defect density on the quench rate. This thesis reports on the results of the experimental study of the Kibble-Zurek mechanism in elongated Bose-Einstein condensates of atomic sodium gases, following the observations on the spontaneous formation of defects after temperature quenches across the BEC transition. The power-law scaling of the defect number with the quench speed was observed and characterized for the first time in ultracold gases. The characterization of the density and phase profiles of the defects allowed their identification as solitonic vortices, representing the first direct experimental evidence for this kind of long living excitation, which sets a link between solitons and vortices. The measurements reported in this thesis provide a novel approach to the study of the critical phenomena happening at phase transitions, and introduce to the possibility of exploring the turbulent dynamics of quenched systems through the spontaneous production of solitonic vortices
Planning and Scheduling in Temporally Uncertain Domains
No full textAny form of model-based reasoning is limited by the adherence of the model to the actual reality. Scheduling is the problem of finding a suitable timing to execute a given set of activities accommodating complex temporal constraints. Planning is the problem of finding a strategy for an agent to achieve a desired goal given a formal model of the system and the environment it is immersed in. When time and temporal constraints are considered, the problem takes the name of temporal planning. A common assumption in existing techniques for planning and scheduling is controllability of activities: the agent is assumed to be able to control the timing of starting and ending of each activity. In several practical applications, however, the actual timing of actions is not under direct control of the plan executor. In this thesis, we focus on this temporal uncertainty issue in scheduling and in temporal planning: we propose to natively express temporal uncertainty in the model used for reasoning. We first analyze the state-of-the-art on the subject, presenting a rationalization of existing works. Second, we show how Satisfiability Modulo Theory (SMT) solvers can be exploited to quickly solve different kinds of query in the realm of scheduling under uncertainty. Finally, we address the problem of temporal planning in domains featuring real-time constraints and actions having duration that is not under the control of the planning agent
Energy Neutral Design of Embedded Systems for Resource Constrained Monitoring Applications
No full textAutomatic monitoring of environments, resouces and human processes are crucial and foundamental tasks to improve people's quality of life and to safeguard the natural environment. Today, new technologies give us the possibility to shape a greener and safer future. The more specialized is the kind of monitoring we want to achieve, more tight are the constraints in terms of reliability, low energy and maintenance-free autonomy. The challenge in case of tight energy constraints is to find new techniques to save as much power as possible or to retrieve it from the very same environment where the system operates, towards the realization of energy neutral embedded monitoring systems. Energy efficiency and battery autonomy of such devices are still the major problem impacting reliability and penetration of such systems in risk-related activities of our daily life. Energy management must not be optimized to the detriment of the quality of monitoring and sensors can not be operated without supply. In this thesis, I present different embedded system designs to bridge this gap, both from the hardware and software sides, considering specific resource constrained scenarios as case studies that have been used to develop solutions with much broader validity. Results achieved demonstrate that energy neutrality in monitoring under resource constrained conditions can be obtained without compromising efficiency and reliability of the outcomes