Home | Personal | Education | Work | Research | Publications | Students | Search


Research Home

Learning Systems

Syntactical Pattern Recognition

Statistical Pattern Recognition

Adaptive Data Structures

Artificial Neural Networks

Database Systems

Data Compression


Data Retrieval and Storage

Following are some of the major results obtained in the area of Learning Systems.
  1. Have proposed Pursuit-based learning algorithms that use Bayesian estimates - for the Continuous Bayesian LA (Proc. AIAI 2011, Proc. IEA/AIE 2011) and the Discretized Bayesian LA (Proc. IEA/AIE 2012, APIN 2013). To the best of our knowledge, as of December 2015, there have been no other similar LA, and these are probably the fastest LA to date.
  2. Have proposed rigorous proofs for the families of Pursuit-based learning algorithms - for the Continuous Pursuit Algorithms (Proc. IEA/AIE 2013, APIN 2014) and the Discretized Pursuit Algorithms (Proc. IEA/AIE 2014, APIN 2015). These results corrected proofs that were flawed for almost three decades.
  3. Have proposed LA-based solutions for optimizing channel selection for cognitive radio networks (Proc. PIMRC 2013, Proc. IEA/AIE 2014), APIN 2015).
  4. Have proposed a hierarchy of twofold resource allocation automata that can supporting optimal web polling. To the best of our knowledge, as of June 2008, there was no other similar hierarchical solution (Proc. IEA/AIE 2008).
  5. Have solved the Stochastic Point Location Problem for non-stationary environments. To the best of our knowledge, as of April 2008, there was no other reported solution for this problem (Proc. IEA/AIE 2007, IEEE T:SMC 2008).
  6. Have used learning automata to efficiently solve the Goore Game (Proc. AI'06 2006, Proc. IEEE-CIG 2007).
  7. Have used a learning automaton solution to the fractional knapsack problem to solve sampling-related resource allocation problems (Proc. IIS:IIPWM 2006, IEEE T:SMC 2007).
  8. Have devised a model for learning automata-based tutorial-like systems (Proc. SummerSim 2006). This talk was a Plenary/Keynote Talk of the Conference. The various components of the Tutorial-like system have also been designed and implemented as follows: Modeling of the Student (Proc. IEA/AIE 2007, IEEE T:SMC 2010); Modeling of the Classroom (Proc. IEEE-SMC 2007, IEEE T:SMC 2010); Modeling of the Domain (Proc. ICMLC 2007, Acta Cybernetica 2010); Modeling of the Teacher (Proc. IEEE-SMC 2007, TCCCI 2012); Modeling the Learning of the Teacher itself (Proc. ISCIS 2007, IEEE T:SMC 2013).
  9. Have used learning automata techniques to solve the fractional knapsack problem and applied the solution to web-searching and allocation problems (Proc. IEEE-CIS 2006, IEEE T:SMC 2006).
  10. Have devised a fault-tolerant routing algorithm for mobile Ad Hoc networks using a stochastic learning-based weak estimation procedure (Proc. WiMob 2006).
  11. Have devised a fast algorithm for routing for MPLS traffic engineering using the principles of stochastic random-races (Proc. INFOCOM 2006, IEEE T:Comp 2007).
  12. Have devised fast and efficient learning automata techniques for dynamic all-pairs shortest paths computations (Proc. ISCC 2005, Proc. FINA 2007, IEEE T:Comp 2006).
  13. Have devised new fixed-structure and variable-structure learning automata solutions for the static mapping and multi-criteria partitioning problems (Proc. SMTPS 2005, Proc. CITSA 2005, Proc. IEEE-CIS 2006).
  14. Have devised fast and efficient learning and pursuit automata techniques for the single source shortest path problem (Proc. IEA/AIE 2004, IEEE T:SMC 2005), and for the shortest path routing tree computation (Proc. ISCC 2004).
  15. Have devised a stochastic learning-based weak estimation strategy for multinomial random variables, and have shown its applications to non-stationary environments (Proc. S+SSPR2004, Pat. Rec. 2006). This has also been used in some data compression problems where the distribution of the data in the files is non-stationary ( Proc. ADVIS 2004, IEEE T:SMC 2006).
  16. Have proposed families of generalized continuous and discretized pursuit learning algorithms. The superiority of these algorithms has been published - these are probably the best learning automata algorithms to-date (IEEE T:SMC 2002). This paper was nominated for the Outstanding Paper of the Year, 2002.
  17. Have proposed families of continuous and discretized pursuit learning algorithms of the reward-inaction and the reward-penalty paradigms. The superiority of these algorithms have also been documented (IEEE T:SMC 2001).
  18. Have proposed learning algorithms for solving the capacity assignment problem for prioritized networks (IEEE T:Comp 2000, IEEE T:SMC 2002). A similar algorithm for assigning capacities and priorities was later developed, but is yet to be published.
  19. Have proposed a learning algorithm for parameter learning from a stochastic liar or a stochastic mentor (IEEE T:SMC 2006). The algorithm need not know whether the teacher is stochastically lying or telling the truth. These results were presented as a Plenary talk at AI'03, the 2003 Australian Joint Conference on Artificial Intelligence, in Perth, Australia, December 2003.
  20. Have proposed learning algorithms for parameter learning from stochastic responses by discretizing the parameter space. A similar algorithm using continuous spaces was later developed (IEEE T:SMC 1997, 1998).
  21. Have proposed, designed and supervised the implementation of a learning-automaton based system for graph partitioning (IEEE T:Comp. 1995).
  22. Have proposed, designed and supervised the implementation of a learning-automaton based system for string taxonomy. This is the first reported solution for it (IEEE T:SMC 1997).
  23. Have proposed, designed and supervised the implementation of a learning-automaton based image retrieval system (Proc. ICARCV 1992).
  24. Have proposed learning automata and genetic solutions for the Keyboard Optimization Problem - the problem of optimizing the position of the keys on a keyboard so as to maximize a certain criterion (IEEE T:SMC 1992).
  25. Have developed discretized pursuit and estimator algorithms. These algorithms have probably been the fastest known learning automata reported for almost a decade (IEEE T:SMC 1990, 1992).
  26. Have worked in the area of random races and its application to learning orderings of actions in a random environment. The theory of such races has been published and their application to economic modeling, scheduling and multi-tasking are being studied (IEEE T:SMC 1993).
  27. Have used the concepts of automata learning to obtain the only known solutions to the Stochastic Minimum Spanning Circle Problem (IEEE T:SMC 1986).
  28. Have discovered many new families of asymptotically optimal discretized learning automata. These are of all the three types: Reward-Inaction, Inaction-Penalty and Reward-Penalty, and can be linear or nonlinear (IEEE T:SMC 1984, 1986).
  29. Worked in the general area of Stochastic Learning. Have discovered variable structure stochastic automata which are ergodic in the mean. These have been proposed for the general learning problem involving multi-actions (IEEE T:SMC 1983, Inf. Sci. 1985).