Continuous action set learning automata for stochastic optimization

Adaptive Limited Fractional Guard Channel Algorithms: A Learning Automata Approach

“…and its objective is find an action α that minimizes M (α). A CALA, whose action probability distribution at instant n is Gaussian with mean µ n and standard deviation σ n , is introduced in 33 . At instance n, the action chosen by CALA can be represented by a pair (α n , µ n ), where α n is chosen from distribution N (µ n , σ n ) and µ n is the mean of the Gaussian distribution.…”

“…For this algorithm it is shown that with arbitrary large probability, µ n will converge close to α * and σ n will converge close to σ L , if we choose µ and σ L sufficiently small and K sufficiently large 33 . Beigy and Meybodi proposed a CALA whose action probability distribution at instant n is Gaussian with mean µ n and standard deviation σ n 34 .…”

“…µ n will converge close to α * 34 . In these two automata 33,34 , it is assumed that no information about the optimal action is available and at the beginning, the actions are chosen from a near uniform distribution. So the initial distribution is considered as a Gaussian distribution with a random mean and a large value of standard deviation.…”

Using Stochastic AI Techniques to Achieve Unbounded Resolution in Finite Player Goore Games and its Applications

“…But unlike the solutions reported in [16], [17], [19], the solution here is far more consequential because the system does not rely on a Teacher or "Oracle" instructing the LA which way it should move. Thus, our solution will have applications in all the areas mentioned earlier for which the GG has found direct applications [9], [10], and for the areas where the entire field of LA and stochastic learning, has found uni-modal optimization applications from a finite or infinite action set [1]- [3], [5], [6], [18], [20].…”

A Novel Strategy for Solving the Stochastic Point Location Problem Using a Hierarchical Searching Scheme

“…Age-old techniques resort to moving toward the optimal point by using the first derivative of the criterion function, and where these updates are, for example, either proportional to the derivative(s) or involve the second derivative(s). In a related manner, the SPL is also associated to the field of learning automata (LA) in which the action space has an infinite number of actions [52]. In this case, one would not resort to invoking the derivatives of the criterion, but rather to sampling the action space and moving appropriately within the space.…”