Jacobi curves are deep generalizations of the spaces of “Jacobi fields” along Riemannian geodesic... more Jacobi curves are deep generalizations of the spaces of “Jacobi fields” along Riemannian geodesics. Actually, Jacobi curves are curves in the Lagrange Grassmannians. In our paper we develop differential geometry of these curves which provides basic feedback or gauge invariants for a wide class of smooth control systems and geometric structures. Two principal invariants are the generalized Ricci curvature, which is an invariant of the parametrized curve in the Lagrange Grassmannian endowing the curve with a natural projective structure, and a fundamental form, which is a fourth-order differential on the curve. The so-called rank 1 curves are studied in more detail. Jacobi curves of this class are associated with systems with scalar controls and with rank 2 vector distributions.
ABSTRACT We announce a new sufficient condition for a bang-bang extremal to be a strong local opt... more ABSTRACT We announce a new sufficient condition for a bang-bang extremal to be a strong local optimum for a control problem in the Mayer form. The controls appear linearly and take values in a polyhedron and the state space and the constraints are smooth, finite dimensional manifolds.
We use a control-theoretic setting to model the process of training (deep learning) of Artificial... more We use a control-theoretic setting to model the process of training (deep learning) of Artificial Neural Networks (ANN), which are aimed at solving classification problems. A successful classifier is the network whose input-output map approximates well the classifying map defined on a finite or an infinite training set. A fruitful idea is substitution of a multi-layer ANN by a continuous-time control system, which can be seen as a neural network with infinite number of layers. Under certain conditions it can achieve high rate of approximation with presumably not so high computational cost. The problem of best approximation for this model results in optimal control problem of Bolza type for ensembles of points. The two issues to be studied are: i) possibility of a satisfactory approximation of complex classification profiles; ii) finding the values of parameters (controls) which provide the best approximation. In control-theoretic terminology it corresponds respectively to the verifi...
Jacobi curves are deep generalizations of the spaces of “Jacobi fields” along Riemannian geodesic... more Jacobi curves are deep generalizations of the spaces of “Jacobi fields” along Riemannian geodesics. Actually, Jacobi curves are curves in the Lagrange Grassmannians. In our paper we develop differential geometry of these curves which provides basic feedback or gauge invariants for a wide class of smooth control systems and geometric structures. Two principal invariants are the generalized Ricci curvature, which is an invariant of the parametrized curve in the Lagrange Grassmannian endowing the curve with a natural projective structure, and a fundamental form, which is a fourth-order differential on the curve. The so-called rank 1 curves are studied in more detail. Jacobi curves of this class are associated with systems with scalar controls and with rank 2 vector distributions.
ABSTRACT We announce a new sufficient condition for a bang-bang extremal to be a strong local opt... more ABSTRACT We announce a new sufficient condition for a bang-bang extremal to be a strong local optimum for a control problem in the Mayer form. The controls appear linearly and take values in a polyhedron and the state space and the constraints are smooth, finite dimensional manifolds.
We use a control-theoretic setting to model the process of training (deep learning) of Artificial... more We use a control-theoretic setting to model the process of training (deep learning) of Artificial Neural Networks (ANN), which are aimed at solving classification problems. A successful classifier is the network whose input-output map approximates well the classifying map defined on a finite or an infinite training set. A fruitful idea is substitution of a multi-layer ANN by a continuous-time control system, which can be seen as a neural network with infinite number of layers. Under certain conditions it can achieve high rate of approximation with presumably not so high computational cost. The problem of best approximation for this model results in optimal control problem of Bolza type for ensembles of points. The two issues to be studied are: i) possibility of a satisfactory approximation of complex classification profiles; ii) finding the values of parameters (controls) which provide the best approximation. In control-theoretic terminology it corresponds respectively to the verifi...
Uploads
Papers by Andrei Agrachev