The consequences of approximating Q Values with function approximators are investigated Two criteria of optimality are introduced a global and local criterion and the viability of each is investigated for the case of a linear combination of prechosen basis functions It is found that optimizing the global cost function is unlikely to result in nearly optimal policies when the set of bases is not...

Design methods of adaptive H∞ consensus control of multi-agent systems composed of the first-order and the second-order regression models and nonlinear terms by utilizing neural network approximators, are presented in this paper. The proposed control schemes are derived as solutions of certain H∞ control problems, where estimation errors of tuning parameters, imperfect knowledge of the leader, ...

| This paper proposes a new method to reduce training time for neural nets used as function approximators. This method relies on a geometrical control of Multilayer Perceptrons (MLP). A geometrical initializa-tion gives rst better starting points for the learning process. A geometrical parametriza-tion achieves then a more stable convergence. During the learning process, a dynamic geometrical c...

Reinforcement learning algorithms can be used to optimally solve dynamic decision-making and control problems. With continuous-valued state input variables, reinforcement must rely on function approximators represent the value policy mappings. Commonly numerical approximators, such as neural networks or basis expansions, have two main drawbacks: they are black-box models offering no insight in ...

Abstract This work studies receding-horizon control of discrete-time switched linear systems subject to polytopic constraints for the continuous states and inputs. The objective is approximate optimal strategy cases in which online computation intractable due necessity solving mixed-integer quadratic programs each discrete time instant. proposed approach builds upon an approximated finite-horiz...

Since 1997 there has been a steady stream of advances for the maximum disjoint paths problem. Achieving tractable results usually required focusing on relaxations such as: (i) to allow some bounded edge congestion in solutions, (ii) only consider unit weight (cardinality) setting, (iii) require fractional routability selected demands (the all-or-nothing flow setting). For general form (no conge...

Partial differential equations are used to model fluid flow in porous media. Neural networks can act as equation solution approximators by basing their forecasts on training samples of permeability maps and corresponding two-point flux approximation solutions. This paper illustrates how convolutional neural various architecture, depth parameter configurations manage forecast solutions the Darcy...