Searcharxiv⌕ Search

arXiv subjects

Laurent Fribourg

Publications and source records attributed to Laurent Fribourg.

At least 19 recordsLinked to original sources

A Data-dependent Early Stopping Rule using Rademacher Complexity with L1-norm

Training neural networks requires balancing the trade-off between fitting the training data and achieving robust performance on unseen inputs. This ability, commonly referred to as generalizability, is determined by the gap between the empirical risk on the training set (``empirical loss'') and the expected risk over the data distribution (``generalization error''). Existing approaches typically estimate the generalization error numerically, requiring gradient descent training and an ``early stopping'' strategy. In this work, we introduce an analytic framework that estimates the optimal time of early stopping without the need for training. Several works in the literature also give such analytical estimations, but they are generally based on random matrix theory and often make assumptions on the distribution of the data or the eigenvalue distribution of the covariance matrix. In contrast, our work is based on Rademacher complexity (RC) without needing such probabilistic assumptions. For both theoretical and numerical reasons, it is more relevant to express RC with the L1- norm rather than with the L2-norm. We focus on the case of linear models and the problem of linear regression. Thanks to the ``linear probing'' method, our results can, however, be successfully applied to nonlinear neural networks, as illustrated in the classification MNIST example.

cs.LG↗

One-Step Early Stopping Strategy using Neural Tangent Kernel Theory and Rademacher Complexity

The early stopping strategy consists in stopping the training process of a neural network (NN) on a set $S$ of input data before training error is minimal. The advantage is that the NN then retains good generalization properties, i.e. it gives good predictions on data outside $S$, and a good estimate of the statistical error (``population loss'') is obtained. We give here an analytical estimation of the optimal stopping time involving basically the initial training error vector and the eigenvalues of the ``neural tangent kernel''. This yields an upper bound on the population loss which is well-suited to the underparameterized context (where the number of parameters is moderate compared with the number of data). Our method is illustrated on the example of an NN simulating the MPC control of a Van der Pol oscillator.

cs.LG↗

Proving the Convergence to Limit Cycles using Periodically Decreasing Jacobian Matrix Measures

Methods based on "(Jacobian) matrix measure" to show the convergence of a dynamical system to a limit cycle (LC), generally assume that the measure is negative everywhere on the LC. We relax this assumption by assuming that the matrix measure is negative "on average" over one period of LC. Using an approximate Euler trajectory, we thus present a method that guarantees the LC existence, and allows us to construct a basin of attraction. This is illustrated on the example of the Van der Pol system.

eess.SY↗

Parametric schedulability analysis of a launcher flight control system under reactivity constraints

The next generation of space systems will have to achieve more and more complex missions. In order to master the development cost and duration of such systems, an alternative to a manual design is to automatically synthesize the main parameters of the system. In this paper, we present an approach for the specific case of the scheduling of the flight control of a space launcher. The approach requires two successive steps: (1) the formalization of the problem to be solved in a parametric formal model and (2) the synthesis of the model parameters with a tool. We first describe the problem of the scheduling of a launcher flight control, then we show how this problem can be formalized with parametric stopwatch automata; we then present the results computed by the parametric timed model checker IMITATOR. We enhance our model by taking into consideration the time for switching context, and we compare the results to those obtained by other tools classically used in scheduling.

cs.SE↗

Constructing invariant tori using guaranteed Euler method

We show here how, using Euler's integration method and an associated function bounding the error in function of time, one can generate structures closely surrounding the invariant tori of dynamical systems. Such structures are constructed from a finite number of balls of $\mathbb{R}^n$ and encompass the deformations of the tori when small perturbations of the flow of the system occur.

eess.SY↗

Robust optimal periodic control using guaranteed Euler's method

In this paper, we consider the application of optimal periodic control sequences to switched dynamical systems. The control sequence is obtained using a finite-horizon optimal method based on dynamic programming. We then consider Euler approximate solutions for the system extended with bounded perturbations. The main result gives a simple condition on the perturbed system for guaranteeing the existence of a stable limit cycle of the unperturbed system. An illustrative numerical example is provided which demonstrates the applicability of the method.

eess.SY↗

Generation of bounded invariants via stroboscopic set-valued maps: Application to the stability analysis of parametric time-periodic systems

A method is given for generating a bounded invariant of a differential system with a given set of initial conditions around a point $x_0$. This invariant has the form of a tube centered on the Euler approximate solution starting at $x_0$, which has for radius an upper bound on the distance between the approximate solution and the exact ones. The method consists in finding a real $T>0$ such that the "snapshot" of the tube at time $t=(i+1)T$ is included in the snapshot at $t=iT$, for some integer $i$. In the phase space, the invariant is therefore in the shape of a torus. A simple additional condition is also given to ensure that the solutions of the system can never converge to a point of equilibrium. In dimension 2, this ensures that all solutions converge towards a limit cycle. The method is extended in case the dynamic system contains a parameter $p$, thus allowing the stability analysis of the system for a range of values of $p$. This is illustrated on classical Van der Pol's system.

eess.SY↗

Proceedings 8th International Workshop on Verification and Program Transformation and 7th Workshop on Horn Clauses for Verification and Synthesis

The proceedings consist of a keynote paper by Alberto followed by 6 invited papers written by Lorenzo Clemente (U. Warsaw), Alain Finkel (U. Paris-Saclay), John Gallagher (Roskilde U. and IMDEA Software Institute) et al., Neil Jones (U. Copenhagen) et al., Michael Leuschel (Heinrich-Heine U.) and Maurizio Proietti (IASI-CNR) et al.. These invited papers are followed by 4 regular papers accepted at VPT 2020 and the papers of HCVS 2020 which consist of three contributed papers and an invited paper on the third competition of solvers for Constrained Horn Clauses. In addition, the abstracts (in HTML format) of 3 invited talks at VPT 2020 by Andrzej Skowron (U. Warsaw), Sophie Renault (EPO) and Moa Johansson (Chalmers U.), are included.

cs.LO↗

Robust optimal control using dynamic programming and guaranteed Euler's method

Set-based integration methods allow to prove properties of differential systems, which take into account bounded disturbances. The systems (either time-discrete, time-continuous or hybrid) satisfying such properties are said to be "robust". In the context of optimal control synthesis, the set-based methods are generally extensions of numerical optimal methods of two classes: first, methods based on convex optimization; second, methods based on the dynamic programming principle. Heymann et al. have recently shown that, for certain systems of low dimension, the second numerical method can give better solutions than the first one. They have built a solver (Bocop) that implements both numerical methods. We show in this paper that a set-based extension of a method of the second class which uses a guaranteed Euler integration method, allows us to find such good solutions. Besides, these solutions enjoy the property of robustness against uncertainties on initial conditions and bounded disturbances. We demonstrate the practical interest of our method on an example taken from the numerical Bocop solver. We also give a variant of our method, inspired by the method of Model Predictive Control, that allows us to find more efficiently an optimal control at the price of losing robustness.

eess.SY↗

Guaranteed phase synchronization of hybrid oscillators using symbolic Euler's method: The Brusselator and biped examples

The phenomenon of phase synchronization was evidenced in the 17th century by Huygens while observing two pendulums of clocks leaning against the same wall. This phenomenon has more recently appeared as a widespread phenomenon in nature, and turns out to have multiple industrial applications. The exact parameter values of the system for which the phenomenon manifests itself are however delicate to obtain in general, and it is interesting to find formal sufficient conditions to guarantee phase synchronization. Using the notion of reachability, we give here such a formal method. More precisely, our method selects a portion $S$ of the state space, and shows that any solution starting at $S$ returns to $S$ within a fixed number of periods $k$. Besides, our method shows that the components of the solution are then (almost) in phase. We explain how the method applies on the Brusselator reaction-diffusion and the biped walker examples.

eess.SY↗

Guaranteed optimal reachability control of reaction-diffusion equations using one-sided Lipschitz constants and model reduction

We show that, for any spatially discretized system of reaction-diffusion, the approximate solution given by the explicit Euler time-discretization scheme converges to the exact time-continuous solution, provided that diffusion coefficient be sufficiently large. By "sufficiently large", we mean that the diffusion coefficient value makes the one-sided Lipschitz constant of the reaction-diffusion system negative. We apply this result to solve a finite horizon control problem for a 1D reaction-diffusion example. We also explain how to perform model reduction in order to improve the efficiency of the method.

math.OC↗

Controlled Recurrence of a Biped with Torso

We have recently used a symbolic reachability method for controlling the stability of special hybrid systems called 'sampled switched systems'. We show here how the method can be extended in order to control the stability of more general hybrid systems with guard conditions and state resets. We illustrate the method through the example of a biped robot with 6 state variables, using a proportional-derivative (PD) controller. More specifically, we isolate a state region R such that, starting from a state located in R just after a footstep, the PD-control makes the robot state return to R at the end of the following footstep.

eess.SY↗

Parametric schedulability analysis of a launcher flight control system under reactivity constraints

The next generation of space systems will have to achieve more and more complex missions. In order to master the development cost and duration of such systems, an alternative to a manual design is to automatically synthesize the main parameters of the system. In this paper, we present an approach on the specific case of the scheduling of the flight control of a space launcher. The approach requires two successive steps: (1) the formalization of the problem to be solved in a parametric formal model and (2) the synthesis of the model parameters with a tool. We first describe the problematic of the scheduling of a launcher flight control, then we show how this problematic can be formalized with parametric stopwatch automata; we then present the results computed by IMITATOR. We compare the results to the ones obtained by other tools classically used in scheduling.

cs.SE↗

Guaranteed Control of Sampled Switched Systems using Semi-Lagrangian Schemes and One-Sided Lipschitz Constants

In this paper, we propose a new method for ensuring formally that a controlled trajectory stay inside a given safety set S for a given duration T. Using a finite gridding X of S, we first synthesize, for a subset of initial nodes x of X , an admissible control for which the Euler-based approximate trajectories lie in S at t $\in$ [0,T]. We then give sufficient conditions which ensure that the exact trajectories, under the same control, also lie in S for t $\in$ [0,T], when starting at initial points 'close' to nodes x. The statement of such conditions relies on results giving estimates of the deviation of Euler-based approximate trajectories, using one-sided Lipschitz constants. We illustrate the interest of the method on several examples, including a stochastic one.

eess.SY↗

Verification of an industrial asynchronous leader election algorithm using abstractions and parametric model checking

The election of a leader in a network is a challenging task, especially when the processes are asynchronous, i.e., execute an algorithm with time-varying periods. Thales developed an industrial election algorithm with an arbitrary number of processes, that can possibly fail. In this work, we prove the correctness of a variant of this industrial algorithm. We use a method combining abstraction, the SafeProver solver, and a parametric timed model-checker. This allows us to prove the correctness of the algorithm for a large number p of processes (p=5000).

cs.LO↗

Control Synthesis of Nonlinear Sampled Switched Systems using Euler's Method

In this paper, we propose a symbolic control synthesis method for nonlinear sampled switched systems whose vector fields are one-sided Lipschitz. The main idea is to use an approximate model obtained from the forward Euler method to build a guaranteed control. The benefit of this method is that the error introduced by symbolic modeling is bounded by choosing suitable time and space discretizations. The method is implemented in the interpreted language Octave. Several examples of the literature are performed and the results are compared with results obtained with a previous method based on the Runge-Kutta integration method.

eess.SY↗

Control of nonlinear switched systems based on validated simulation

We present an algorithm of control synthesis for nonlinear switched systems, based on an existing procedure of state-space bisection and made available for nonlinear systems with the help of validated simulation. The use of validated simulation also permits to take bounded perturbations and varying parameters into account. It is particularly interesting for safety critical applications, such as in aeronautical, military or medical fields. The whole approach is entirely guaranteed and the induced controllers are correct-by-design.

eess.SY↗

Distributed Synthesis of State-Dependent Switching Control

We present a correct-by-design method of state-dependent control synthesis for linear discrete-time switching systems. Given an objective region R of the state space, the method builds a capture set S and a control which steers any element of S into R. The method works by iterated backward reachability from R. More precisely, S is given as a parametric extension of R, and the maximum value of the parameter is solved by linear programming. The method can also be used to synthesize a stability control which maintains indefinitely within R all the states starting at R. We explain how the synthesis method can be performed in a distributed manner. The method has been implemented and successfully applied to the synthesis of a distributed control of a concrete floor heating system with 11 rooms and 2^11 = 2048 switching modes.

eess.SY↗