Logo: to the web site of Uppsala University

We consider the joint problem of online experiment design and parameter estimation for identifying nonlinear system models, while adhering to system constraints. We utilize a receding horizon approach and propose a new adaptive input design criterion, which is tailored to continuously updated parameter estimates, along with a new sequential estimator. We demonstrate the ability of the method to design informative experiments online, while steering the system within operational constraints.

Inventory control is subject to service-level requirements, in which sufficient stock levels must be maintained despite an unknown demand. We propose a data-driven order policy that certifies any prescribed service level under minimal assumptions on the unknown demand process. The policy achieves this using any online learning method along with integral action. We further propose an inference method that is valid in finite samples. The properties and theoretical guarantees of the method are illustrated using both synthetic and real-world data. 

Time-delays are known to have a detrimentaleffect on feedback systems. In the context of networkedcyber-physical systems, delays can be injected by maliciousadversaries. Detecting them early is an important challenge.This paper proposes a novel variation of Interacting MultipleModel filtering to detect delay injection attacks in feedbackcontrol systems, when hidden in an open loop setting. Thedetection scheme is formalised by treating delay as alternativemodes of the system, and theoretical analysis of the stationarydistribution informs a reduction to a three parameter model aswell as the choices of hyper parameter values. The method isevaluated on a cruise control application, and shows detectionwithin a few seconds and a low false alarm probability.

The current alarming loss of biodiversity requires improved techniques for mitigation. The ability to predict the expected effect of interventions affecting biodiversity ahead in time, with confidence limits, will therefore be central for efficient policy making in the future. This work develops new techniques for modeling of the biodiversity response that follow an abrupt intervention, like clear-cutting of forest. The modeling quantifies the dynamics of biodiversity metrics like species richness and geometric mean abundance with differential equations, where the unknown parameters are estimated with prediction error system identification techniques. Since the data is scarce in time per clear-cutting location, a minimum amount of only four parameters is estimated for all locations in a geographical region. The parameters are estimated by aggregation over multiple clear-cutting locations to obtain an improved statistical significance based on an assumption of independence between locations. Since the associated loss function is highly asymmetric around the estimate, the bootstrap technique is applied to the data to obtain an asymmetric confidence region that is statistically correct. The method is applied to data from the Swedish forest inventory, and shown to perform as desired. From a biodiversity perspective the results are in line with previous knowledge. The methodology also leads to the formulation of a number of new research questions. A specific advantage is that the data compression of the algorithm into a low order model allows statistically confident estimates to be obtained also from very small sets of data, here only a few tens of samples per region. This allows assessment of data sets that are far smaller and below the operating range of machine learning algorithms.

Multistatic ranging with mobile receivers enables good tracking performance due to the combined adaptive sensor geometry. However, environments that contain signal obstructing obstacles require receiver trajectory planning to maintain line-of-sight (LOS) conditions with transmitters and the target of interest. In this letter, we develop a non-myopic receding-horizon framework for multistatic tracking. It uses convex collision-avoidance constraints and a control objective that focuses on maintaining good LOS signal conditions, taking into account future obstructions. We demonstrate the efficiency and tracking accuracy of the method via a numerical experiment.

Background: Only half of patients appear adherent to antihypertensive treatment. If so, the effectiveness in prevention against adverse cardiovascular outcomes may depend more on a drugs' side effects than its blood pressure effects. We hypothesized that initiating treatment with a particular antihypertensive drug class will determine later persistence.

Methods: We applied a new-user trial emulation approach to multiple mandatory health-care registers in Sweden. All individuals starting antihypertensive treatment with an angiotensin receptor blocker (ARB), angiotensin converting enzyme inhibitor (ACEi), calcium channel blocker (CCB), thiazide/thiazide-like diuretic (TD), or single-pill combination of two drugs (SPC) in 2011-2018 were assessed for inclusion. Patients with prior cardiovascular disease or diabetes were excluded. Persistence was defined using dispensed prescription refills with at least 80% adherence. We applied multistate Poisson regression models adjusting for age, sex, obesity, birth country, education, income, marital status and year of initiation. The study period was January 1st 2011-December 31st 2019.

Findings: A total of 341,182 patients were included. Initiating antihypertensive treatment with an ARB was associated with a larger proportion continuing their original drug class, or indeed any antihypertensive drug class, at any time during the five years of follow-up. At three years, 44.7% (CI 43.7-45.7) of the ARB initiators were continuously persistent to their original drug class. The corresponding number if intermittent discontinuations were allowed was 81.8% (CI 80.9-82.8). Overall, if intermittent discontinuations and changes between classes were allowed, about 80% of all new users of antihypertensive drugs (regardless of class at initiation) were persistent to at least one antihypertensive drug class over 5 years.

Interpretation: Initiating treatment with an ARB is associated with the best long-term class persistence and least treatment changes. The "rule of halves" should be replaced with the insight that, after the first year, at least 4 out of 5 patients take their antihypertensive drug at least 4 out of 5 days.

We consider the problem of tracking moving targets using mobile wireless sensors (of possibly different types). This is a joint estimation and control problem in which a tracking system must take into account both target and sensor dynamics. We make minimal assumptions about the target dynamics, namely only that their accelerations are bounded. We develop a control law that determines the sensor motion control signals so as to maximize target resolvability as the target dynamics evolve. The method is given a tractable formulation that is amenable to an efficient search method and is evaluated in a series of experiments involving both round-trip time based ranging and Doppler frequency shift measurements.

Adversarial training can be used to learn models that are robust against perturbations. For linear models, it can be formulated as a convex optimization problem. Compared to methods proposed in the context of deep learning, leveraging the optimization structure allows significantly faster convergence rates. Still, the use of generic convex solvers can be inefficient for large-scale problems. Here, we propose tailored optimization algorithms for the adversarial training of linear models, which render large-scale regression and classification problems more tractable. For regression problems, we propose a family of solvers based on iterative ridge regression and, for classification, a family of solvers based on projected gradient descent. The methods are based on extended variable reformulations of the original problem. We illustrate their efficiency in numerical examples.