Generalized TSE: a new generalized estimator-based learning automaton

Agache, M.; Oommen, B. John

doi:10.1109/iccis.2004.1460420

Cited by 2 publications

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, at every iteration, the estimated reward vector was also used to update the action probability vector, instead of updating it based only on the RE's feedback. In this way, the frequency of choosing the actions with higher reward estimates is increased and the chances of choosing actions with lower estimates are significantly reduced 3 . In doing this, Thathachar and Sastry proposed the family of estimator algorithms for the P -model in [115], and extended these concepts to the S-model in [113].…”

Section: Estimator and Pursuit Algorithmsmentioning

confidence: 99%

“…The vectorized updating rule for the estimator algorithm was proposed by Agache and Oommen in [3]. They utilized it to derive the formula for the generalized probability updating scheme obtained by applying unequal weights for the probabilities with greater reward estimate values.…”

Section: Estimator and Pursuit Algorithmsmentioning

confidence: 99%

“…These two phases are applied alternatively until no further improvements can be obtained. These phases are applied equivalently and alternatively to the cost function in the same manner, until the cost of processing the transactions cannot be reduced 3 When k objects are accessed in a query, we can consider it as a case of having variations, where k is the block count of the present candidate solution. In an approach similar to the first phase, the costs of variations are evaluated by the PE, and the one yielding the greatest improvement over the previous candidate becomes the new current candidate.…”

Section: A Hill Climbing Solutionmentioning

confidence: 99%

“…Please note that this is the total probability of E presenting the pair ⟨O i , O j ⟩ 3. It can be shown that by an appropriate re-mapping of the indices, one can obtain a block matrix in which the elements of the partitioning Ω * are adjacent, even if the original matrix is cluttered.…”

mentioning

confidence: 99%

“…Our OMA-based scheme works analogously 3. Typically, in the literature, all of the "reliable" sensors share the same probability value, p.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Novel Solutions and Applications of the Object Partitioning Problem

Shirvani¹

View full text Add to dashboard Cite

This thesis considers the fundamental problem of "partitioning" which is allpervasive in computer science. It has applications in "Big Data" because the vast amounts of data encountered in "Big Data" applications cannot be processed in a single block, but are better analyzed when it is partitioned in various monolithic units. It also has direct applications in numerous areas including databases, process scheduling, mapping and image retrieval. In this research we consider a specific instantiation of the Object Partitioning Problem, namely the Equi-Partitioning Problem (EPP), in which the partitions are equi-sized. In particular we concentrate on the various Learning Automata (LA)-based solutions. In this regard, the Object Migration Automata (OMA), and its variants have been the benchmark solutions. The thesis thoroughly investigates the structural aspects of the OMA such as its internal grouping-based transitions, structure, initialization methods, convergence criteria and its recorded deficiencies. Rather than solely consider how the transitions can be designed, we specifically investigate how the well-known Pursuit paradigm, that has been recently invoked to enhance the field of LA, can be used to optimize the OMA. To do this, we propose two models for the Environment when it is noisy and noiseless, and show that the LA's convergence can be improved if we can discern the so-called "divergent" queries, namely, those which cause the LA to be sluggish. We have thus devised a technique that recognizes the "divergent" pairs and excludes them from the learning cycle, and to thus effectively pursue the optimal solution. This has resulted in the Pursuit OMA (POMA), whose performance is sometimes nearly two times faster than the original OMA. Other researchers have observed and corrected the deadlock phenomenon in the original OMA, and have thus designed the so-called Enhanced OMA (EOMA). In this thesis, we have shown how the Pursuit phenomenon can be incorporated into the EOMA to yield the Pursuit EOMA (PEOMA), which, in some cases, is more than forty times faster that the traditional OMA. Thereafter, we have observed that all the previous versions of the OMA were not able to include the property of transitivity. To include this phenomenon, we have also extracted the relations among the objects by utilizing the information in the iii Pursuit matrix. The newly-proposed method that incorporates Transitivity, the socalled Transitive PEOMA (TPEOMA) outperforms the PEOMA by a factor of two, and is sometimes about ninety times faster than the OMA. To finally demonstrate the power of the OMA-based paradigm in a real-life application, we have also incorporated it to resolve the outlier detection problem in Noisy Sensors Networks (NSNs). iv ACKNOWLEDGEMENTS I will, first and foremost, thank God, my Creator, for giving me the strength and patience to see this work through to its fruitful conclusion. This has been no small endeavor, which has encountered numerous obstacles, highs and lows. He has helped me all along. I am also g...

show abstract

Section: Estimator and Pursuit Algorithmsmentioning

confidence: 99%

Section: Estimator and Pursuit Algorithmsmentioning

confidence: 99%

Section: A Hill Climbing Solutionmentioning

confidence: 99%

mentioning

confidence: 99%

“…Our OMA-based scheme works analogously 3. Typically, in the literature, all of the "reliable" sensors share the same probability value, p.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Novel Solutions and Applications of the Object Partitioning Problem

Shirvani¹

View full text Add to dashboard Cite

show abstract

Parameter learning from stochastic teachers and stochastic compulsive liars

Oommen

Raghunath

Kuipers

2006

IEEE Trans. Syst., Man, Cybern. B

View full text Add to dashboard Cite

This paper considers a general learning problem akin to the field of learning automata (LA) in which the learning mechanism attempts to learn from a stochastic teacher or a stochastic compulsive liar. More specifically, unlike the traditional LA model in which LA attempts to learn the optimal action offered by the Environment (also here called the "Oracle"), this paper considers the problem of the learning mechanism (robot, an LA, or in general, an algorithm) attempting to learn a "parameter" within a closed interval. The problem is modeled as follows: The learning mechanism is trying to locate an unknown point on a real interval by interacting with a stochastic Environment through a series of informed guesses. For each guess, the Environment essentially informs the mechanism, possibly erroneously (i.e., with probability p), which way it should move to reach the unknown point. When the probability of a correct response is p > 0.5, the Environment is said to be informative, and thus the case of learning from a stochastic teacher. When this probability p < 0.5, the Environment is deemed deceptive, and is called a stochastic compulsive liar. This paper describes a novel learning strategy by which the unknown parameter can be learned in both environments. These results are the first reported results, which are applicable to the latter scenario. The most significant contribution of this paper is that the proposed scheme is shown to operate equally well, even when the learning mechanism is unaware of whether the Environment ("Oracle") is informative or deceptive. The learning strategy proposed herein, called CPL-AdS, partitions the search interval into d subintervals, evaluates the location of the unknown point with respect to these subintervals using fast-converging E-optimal LRI LA, and prunes the search space in each iteration by eliminating at least one partition. The CPL-AdS algorithm is shown to provably converge to the unknown point with an arbitrary degree of accuracy with probability as close to unity as desired. Comprehensive experimental results confirm the fast and accurate convergence of the search for a wide range of values for the Environment's feedback accuracy parameter p, and thus has numerous potential applications.

show abstract

Generalized TSE: a new generalized estimator-based learning automaton

Cited by 2 publications

References 7 publications

Novel Solutions and Applications of the Object Partitioning Problem

Novel Solutions and Applications of the Object Partitioning Problem

Parameter learning from stochastic teachers and stochastic compulsive liars

Contact Info

Product

Resources

About