Mazen Alamir

Mazen Alamir

In this note, a new formulation of Model Predictive Control (MPC) framework with no stability-related terminal constraint is proposed and its stability is proved under mild standard assumptions. The novelty in the formulation lies in the use of time-varying monotonically increasing stage cost penalty. The main result is that the $0$-reachability prediction horizon can always be made stabilizing provided that the increasing rate of the penalty is made sufficiently high.

Mazen Alamir

This paper proposes a general framework for probabilistic certification of cancer therapies. The certification is defined in terms of two key issues which are the tumor contraction and the lower admissible bound on the circulating lymphocytes which is viewed as indicator of the patient health. The certification is viewed as the ability to guarantee with a predefined high probability the success of the therapy over a finite horizon despite of the unavoidable high uncertainties affecting the dynamic model that is used to compute the optimal scheduling of drugs injection. The certification paradigm can be viewed as a tool for tuning the treatment parameters and protocols as well as for getting a rational use of limited or expensive drugs. The proposed framework is illustrated using the specific problem of combined immunotherapy/chemotherapy of cancer.

Mirko Fiacchini, Mazen Alamir

In this paper we consider the problem of computing control invariant sets for linear controlled high-dimensional systems with constraints on the input and on the states. Set inclusions conditions for control invariance are presented that involve the N-step sets and are posed in form of linear programming problems. Such conditions allow to overcome the complexity limitation inherent to the set addition and vertices enumeration and can be applied also to high dimensional systems. The efficiency and scalability of the method are illustrated by computing approximations of the maximal control invariant set, based on the 10-step operator, for a system whose state and input dimensions are 30 and 15, respectively.

Mazen Alamir, Sacha Clavel

This paper addresses the problem of identifying parsimonious explicit piece-wise polynomial relationships that might involve a relatively large number of raw features. The algorithm leverages a recently proposed identification algorithm that yields parsimonious implicit relationships enabling to derive normality characterization in the context of anomaly detection and localization. The algorithm proposed in this paper goes a step further by deriving explicit piece-wise representations that are built using the set of polynomials involved in the implicit representations. The framework is illustrated on the problem of identifying parsimonious explicit representations of the inverse model of a 6-axis manipulator robot. Moreover, further experiments on a 4-axis robot are also shown which are designed to investigate the generalization capability of parsimonious models compared to state-of-the-art DNNs structures, when models face unseen contexts of use.

Mazen Alamir

In this paper, a supervised clustering based-heuristic is proposed for the real-time implementation of approximate solutions to stochastic nonlinear model predictive control frameworks. The key idea is to update on-line a low cardinality set of uncertainty vectors to be used in the expression of the stochastic cost and constraints. These vectors are the centers of uncertainty clusters that are built using the optimal control sequences, cost and constraints indicators as supervision labels. The use of a moving clustering data buffer which accumulates recent past computations enables to reduce the computational burden per sampling period while making available at each period a relevant amount of samples for the clustering task. A relevant example is given to illustrate the contribution and the associated algorithms.

Mazen Alamir

In this paper, a new nonlinear identification framework is proposed to address the issue of off-line computation of moving-horizon observer estimate. The proposed structure merges the advantages of nonlinear approximators with the efficient computation of constrained quadratic programming problems. A bound on the estimation error is proposed and the efficiency of the resulting scheme is illustrated using two state estimation examples.

Mazen Alamir, Jean Dobrowolski, Amgad tarek Mohammed

This paper addresses the design of robust dynamic output feedback control for highly uncertain systems in which the unknown disturbance might be excited by the derivative of the control input. This context appears in many industrial problems such as the speed control of the hydraulic turbines and the frequency stabilization in micro grids to cite but few examples. A key feature that has to be carefully addressed in this context is that too agressive feedback might lead to the loss of controllability and/or a significant drop in the closed-loop performance. The paper formulates the problem, underlines its relevance and gives a rigorous solution in which a dynamic output feedback is given together with a realistic set of sufficient conditions on the controller's parameters that enable to assess the behavior of the closed-loop under different circumstances. A numerical example is given to illustrate the relevance of the proposed successful design characterization.

Mazen Alamir

This paper describes the Parametrized Derivative-Free Model Predictive Control pdf-mpc package, a matlab coder-based set of subroutines that enables a model predictive control problem to be defined and solved. the pdf-mpc is made available for free download and use through the website of the author..

Mazen Alamir

Contraction-Based Nonlinear Model Predictive Control (NMPC) formulations are attractive because of the generally short prediction horizons they require and the needless use of terminal set computation that are commonly necessary to guarantee stability. However, the inclusion of the contraction constraint in the definition of the underlying optimization problem often leads to non standard features such as the need for multi-step open-loop application of control sequences or the use of multi-step memorization of the contraction level that may induce unfeasibility in presence of unexpected disturbance. This paper proposes a new formulation of contraction-based NMPC in which no contraction constraint is explicitly involved. Convergence of the resulting closed-loop behavior is proved under mild assumptions.

Mazen Alamir

In this paper, a methodology is proposed that enables to analyze the sensitivity of the outcome of a therapy to unavoidable high dispersion of the patient specific parameters on one hand and to the choice of the parameters that define the drug delivery feedback strategy on the other hand. More precisely, a method is given that enables to extract and rank the most influent parameters that determine the probability of success/failure of a given feedback therapy for a given set of initial conditions over a cloud of realizations of uncertainties. Moreover predictors of the expectations of the amounts of drugs being used can also be derived. This enables to design an efficient stochastic optimization framework that guarantees safe contraction of the tumor while minimizing a weighted sum of the quantities of the different drugs being used. The framework is illustrated and validated using the example of a mixed therapy of cancer involving three combined drugs namely: a chemotherapy drug, an immunology vaccine and an immunotherapy drug. Finally, in this specific case, it is shown that dash-boards can be built in the 2D-space of the most influent state components that summarize the outcomes' probabilities and the associated drug usage as iso-values curves in the reduced state space.

Mazen Alamir

In this paper, a tractable methodology is proposed to approximate stochastic optimal feedback treatment in the context of mixed immuno-chemo therapy of cancer. The method uses a fixed-point value iteration that approximately solves a stochastic dynamic programming-like equation. It is in particular shown that the introduction of a variance-related penalty in the latter induces better results that cope with the consequences of softening the health safety constraints in the cost function. The convergence of the value function iteration is revisited in the presence of the variance related term. The implementation involves some Machine Learning tools in order to represent the optimal function and to perform complexity reduction by clustering. Quantitative illustration is given using a commonly used model of combined therapy involving twelve highly uncertain parameters.

Mazen Alamir

In this paper, it is shown that the solutions of general differentiable constrained optimization problems can be viewed as asymptotic solutions to sets of Ordinary Differential Equations (ODEs). The construction of the ODE associated to the optimization problem is based on an exact penalty formulation in which the weighting parameter dynamics is coordinated with that of the decision variable so that there is no need to solve a sequence of optimization problems, instead, a single ODE has to be solved using available efficient methods. Examples are given in order to illustrate the results. This includes a novel systematic approach to solve combinatoric optimization problems as well as fast computation of a class of optimization problems using analogic circuits leading to fast, parallel and highly scalable solutions.

Houssem Meghnoudj, Bogdan Robu, Mazen Alamir

In this study we focus on the diagnosis of Parkinson's Disease (PD) based on electroencephalogram (EEG) signals. We propose a new approach inspired by the functioning of the brain that uses the dynamics, frequency and temporal content of EEGs to extract new demarcating features of the disease. The method was evaluated on a publicly available dataset containing EEG signals recorded during a 3-oddball auditory task involving N = 50 subjects, of whom 25 suffer from PD. By extracting two features, and separating them with a straight line using a Linear Discriminant Analysis (LDA) classifier, we can separate the healthy from the unhealthy subjects with an accuracy of 90 % $(p < 0.03)$ using a single channel. By aggregating the information from three channels and making them vote, we obtain an accuracy of 94 %, a sensitivity of 96 % and a specificity of 92 %. The evaluation was carried out using a nested Leave-One-Out cross-validation procedure, thus preventing data leakage problems and giving a less biased evaluation. Several tests were carried out to assess the validity and robustness of our approach, including the test where we use only half the available data for training. Under this constraint, the model achieves an accuracy of 83.8 %.

Mazen Alamir

In this paper, a simple heuristic is proposed for the design of uncertainty aware predictive controllers for nonlinear models involving uncertain parameters. The method relies on Machine Learning-based approximation of ideal deterministic MPC solutions with perfectly known parameters. An efficient construction of the learning data set from these off-line solutions is proposed in which each solution provides many samples in the learning data. This enables a drastic reduction of the required number of Non Linear Programming problems to be solved off-line while explicitly exploiting the statistics of the parameters dispersion. The learning data is then used to design a fast on-line output dynamic feedback that explicitly incorporate information of the statistics of the parameters dispersion. An example is provided to illustrate the efficiency and the relevance of the proposed framework. It is in particular shown that the proposed solution recovers up to 78\% of the expected advantage of having a perfect knowledge of the parameters compared to nominal design.

Mazen Alamir

In this paper, a framework with complete numerical investigation is proposed regarding the feasibility of constrained Nonlinear Model Predictive Control (NMPC) design using Data-Driven model of the cost function. Although the idea is very much in the air, this paper proposes a complete implementation using python modules that are made freely available on a GitHub repository. Moreover, a discussion regarding the different ways of deriving control via data-driven modeling is proposed that can be of interest to practitioners.

Mirko Fiacchini, Mazen Alamir

In this paper we consider the problem of computing control invariant sets for linear controlled systems with constraints on the input and on the states. We focus in particular on the complexity of the computation of the N-step operator, given by the Minkowski addition of sets, that is the basis of many of the iterative procedures for obtaining control invariant sets. Set inclusions conditions for control invariance are presented that involve the N-step sets and are posed in form of linear programming problems. Such conditions are employed in algorithms based on LP problems that allow to overcome the complexity limitation inherent to the set addition and can be applied also to high dimensional systems. The efficiency and scalability of the method are illustrated by computing in less than two seconds an approximation of the maximal control invariant set, based on the 15-step operator, for a system whose state and input dimensions are 20 and 10 respectively.

Mazen Alamir

In this paper, a state-dependent control updating period framework is proposed that leads to real-time implementable Model Predictive Control with certified practical stability results and constraints satisfaction. The scheme is illustrated and validated using new certification bound that is derived in the case where the Fast Gradient iteration is used through a penalty method to solve generally constrained convex optimization problems. Both the certification bound computation and its use in the state-dependent updating period framework are illustrated in the particular case of linear MPC. An illustrative example involving a chain of four integrators is used to show the explicit computation of the state-dependent control updating scheme.

Mazen Alamir

In this paper, a method is proposed for on-line monitoring of the control updating period in fast-gradient-based Model Predictive Control (MPC) schemes. Such schemes are currently under intense investigation as a way to accommodate for real-time requirements when dealing with systems showing fast dynamics. The method needs cheap computations that use the algorithm on-line behavior in order to recover the optimal updating period in terms of cost function decrease. A simple example of constrained triple integrator is used to illustrate the proposed method and to assess its efficiency.