1,721,030 research outputs found
A CUSUM-Based Approach for the Dynamic Step-Size Selection Problem in Primal-Based Distributed Optimization Algorithms
No full textWe present a novel approach to dynamically select step-sizes in primal-based distributed optimization algorithms. The proposed method is based on the local evaluation of local parameters, and thus does not come with any additional communication burden (particularly important in communication environments with limited data rates, such as underwater systems), is computationally lightweight (particularly important when agents have limited computation or power capabilities, such as in internet of underwater things scenarios), and require no prior information about the network topology or objective function (particularly important in situations where the agents may change location and solve different tasks in different times). The analysed scheme ladders on classical hypothesis testing methods to estimate where the local dynamics of the local variables are evolving towards. In this approach, we utilized cumulative sum (CUSUM) to detect the average growth in the evolution of local parameters. This enables establishing mechanisms for changing dynamically the local step-sizes so to simultaneously avoid divergence but also increase convergence speed when necessary. We thus evaluate the performance of the novel method using a dedicated Monte Carlo analysis that compares it against oracle-based selectors, i.e., against situations where the network employs the best possible fixed step-size. The results show that the proposed method can sometimes lead to faster convergence than such optimal fixed step-sizes, often have comparable performance against them, and in any case guarantee convergence
Average consensus via max consensus
No full textSince intuition states that it is simple and fast to compute maxima over networks, we aim at understanding the limits of computing averages over networks through computing maxima. We thus build on top of max-consensus based networks' cardinality estimation protocols a novel estimation strategy that infers averages through computing maxima of opportunely and locally generated random initial conditions. We motivate the max-consensus strategy explaining why it satisfies practical requirements, we characterize completely its statistical properties, and we analyze when and under which conditions it performs favorably against classical linear consensus strategies in static Cayley graphs
Automatic control: The natural approach for a quantitative-based personalized education
Full text linkThis paper proposes an engineering-oriented framework that casts the problem of learning as an automatic control problem, and that can ultimately be used to design education activities that autonomously adapt to individual students' abilities, prerequisites, learning goals and other restrictions. The framework leverages on quantitative descriptions of knowledge flows within university programs in terms of Knowledge Components Matrices (KCMs) and Knowledge Flow Graphs (KFGs), that serve as the basis for developing the aforementioned automated approach to personalized education. Essentially, the manuscript proposes to: 1) combine these descriptions with results from exams and assessments to statistically estimate the learning status of a student; 2) combine these descriptions with data-driven approaches to derive models of how knowledge ladders logically and in time; 3) use these two ingredients to automatically design suitable and personalized study activities for a student, given his/her current knowledge status and desired learning outcome. We describe all steps (modelling of the knowledge flows, estimating the current learning status, and derivation of suitable learning activities to close the loop) with formal and control-oriented notation. The paper serves thus the purpose of showing how methods from the field of system theory and control engineering are naturally useful for the implementation of quantitative-based personalized education
Networks cardinality estimation using order statistics
No full textWe consider a network of collaborative peers that aim at distributedly estimating the network cardinality. We assume nodes to be endowed with unique identification numbers (IDs), and we study the performance of size estimators that are based on exchanging these IDs. Motivated by practical scenarios where the time-to-estimate is critical, we specifically address the case where the convergence time of the algorithm, i.e., the number of communications required to achieve the final estimate, is minimal. We thus construct estimators of the network size by exploiting statistical inference concepts on top of the distributed computation of order statistics of the IDs, i.e., of the M biggest IDs available in the network. We then characterize the statistical performance of these estimators from theoretical perspectives and show their effectiveness in practical estimation situations by means of numerical examples
Hyperparameters Tuning in Regularized System Identification with Nonzero Prior Means
No full textWe consider the problem of identifying linear time invariant systems using regularization schemes, and address the fact that generally the mean value of the corresponding parameter prior is set to zero. We thus consider the scenario where it is beneficial to use a prior with nonzero-mean, where this mean moreover depends on some hyperparameters. We show how to construct such priors and do hyperparameter tuning by marginal likelihood, and since a parameter dependent mean may slow down optimization, we also derive an efficient and stable way of treating them, leading to an overall scheme whose leading order numerical complexity is the same as in the case where the prior mean is zero. The proposed method thus allows including new types of external information in the prior, and we exemplify how this extension can improve the existing regularization techniques
Distributed estimation of diameter, radius and eccentricities in anonymous networks
No full textWe consider how a set of collaborating agents can distributedly infer some of the properties of the communication network that they form. We specifically focus on estimating quantities that can characterize the performance of other distributed algorithms, namely the eccentricities of the nodes, and the radius and diameter of the network. We propose a strategy that can be implemented in any network, even under anonymity constraints, and has the desirable properties of being fully distributed, parallel and scalable. We analytically characterize the statistics of the estimation error, and highlight how the performance of the algorithm depends on a parameter tuning the communication complexity. © 2012 IFAC
Distributed model-invariant detection of unknown inputs in networked systems
No full textThis work considers hypothesis testing in networked systems under severe lack of prior knowledge. In previous work we derived a centralized Uniformly Most Powerful Invariant (UMPI) approach to testing unknown inputs in unknown Linear Time Invariant (LTI) networked dynamics subject to unknown Gaussian noise. The detector was also shown to have Constant False Alarm Rate (CFAR) properties. Nonetheless, in large-scale systems, centralized testing may be infeasible or undesirable. Thus, we develop a distributed testing version of our previous work that utilizes a statistic that is maximally invariant to the unknown parameters and the nonlocal/neighboring measurements. Similar to the centralized approach, the distributed test is shown to have CFAR properties and to have performance that asymptotically approaches that of the centralized test. Simulation results illustrate that the performance of the distributed approach suffers marginal performance degradation in comparison to the centralized approach. Insight to this phenomena is provided through a discussion. © 2013 ACM
An outline of the story of girls in control and its success in motivating girls internationally
Full text linkGender stereotypes often deter women from pursuing STEM-related studies at secondary and tertiary levels of education. Control engineering in particular is an example of a discipline where women are underrepresented in all stages of academia; from undergraduate students through to faculty and technical board members. The Girls in Control workshop targets 10-to-15-year-old girls and aims to educate them about control engineering at a level that is understandable and engaging to stimulate an interest in STEM. Moreover, the worldwide COVID-19 pandemic resulted in new and innovative ways of collaborating and designing outreach programs. The Girls in Control workshop uses online platforms to provide accessibility to girls worldwide by removing language barriers. The workshop runs in almost 20 different languages and the materials are openly available online. The Girls in Control workshop ran successfully at the 21st World Congress of the International Federation of Automatic Control 2020, the IEEE Conference on Decision and Control 2020, the 2021 American control conference, and the 29th Mediterranean Conference on Control and Automation 2021; all of which ran on virtual platforms. Overall, over 500 girls have participated in the Girls in Control workshops in 19 languages with a large amount of positive feedback. A more advanced, follow-up workshop is being tested where girls are challenged further by tackling problems with disturbances and requiring complex control solutions
Distributed detection of topological changes in communication networks
No full textChanges in the topology of communication networks, such as sudden appearance or disappearance of links or nodes, may signal malicious attacks or malfunctions. A topology change detector may thus be useful to trigger alarms or self-reconfiguration procedures. Here we present a novel approach that enjoys several desirable qualities such as fast convergence, intrinsically distributed computations, and scalability w.r.t. communication and computational requirements. We characterize the performance of this technique from analytical and practical points of view, providing theoretical results on its performance. We thus show how it is possible to tune and trade-off the accuracy of the change detection results with the communication requirements of the procedure
Calibrating distance sensors for terrestrial applications without groundtruth information
No full textThis paper describes a new calibration procedure for distance sensors that does not require independent sources of groundtruth information, i.e., that is not based on comparing the measurements from the uncalibrated sensor against measurements from a precise device assumed as the groundtruth. Alternatively, the procedure assumes that the uncalibrated distance sensor moves in space on a straight line in an environment with fixed targets, so that the intrinsic parameters of the statistical model of the sensor readings are calibrated without requiring tests in controlled environments, but rather in environments where the sensor follows linear movement and objects do not move. The proposed calibration procedure exploits an approximated expectation maximization scheme on top of two ingredients: an heteroscedastic statistical model describing the measurement process, and a simplified dynamical model describing the linear sensor movement. The procedure is designed to be capable of not just estimating the parameters of one generic distance sensor, but rather integrating the most common sensors in robotic applications, such as Lidars, odometers, and sonar rangers and learn the intrinsic parameters of all these sensors simultaneously. Tests in a controlled environment led to a reduction of the mean squared error of the measurements returned by a commercial triangulation Lidar by a factor between 3 and 6, comparable to the efficiency of other state-of-the art groundtruth-based calibration procedures. Adding odometric and ultrasonic information further improved the performance index of the overall distance estimation strategy by a factor of up to 1.2. Tests also show high robustness against violating the linear movements assumption
- …
