The quality of optimized decision-making algorithms (estimation, forecasting, classification, hypothesis testing, economic, medical or political decision-making, management, etc.) depends, often critically, on the choice of their parameters (order of models, weight of individual attributes in multi-criteria decision-making, probability of mutations in genetic algorithms, etc.). Therefore, it is desirable to set them, preferably automatically. The work is focused on a general solution to this problem, which is decision-making in the meta-space of parameters.
1. Learn about Bayesian estimation of Markov chains.
2. Familiarize yourself with dynamic programming to the extent necessary for Markov decision-making processes.
3. Familiarize yourself with the methodologies for setting design parameters optimal decision rule.
4. Design your own decision rule setting the parameters mentioned in point 3.
5. Implement the necessary algorithms in the Matlab environment and experimentally evaluate the quality of your solution using a simple example.
Recommended literature (parts selected after agreement with the supervisor)
1. V. Peterka, Bayesian System Identification, in P. Eykhoff "Trends and Progress in System Identification", Pergamon Press, Oxford, 239-304, 1981.
2. M. Puterman, Markov decision processes, John Wiley & Sons, 1994.
3. D. P. Bertsekas, Dynamic Programming, Prentice Hall, 1987.
4. A. E. Eiben, S. K. Smit, Parameter tuning for configuring and analyzing evolutionary algorithms, Swarm and Evolutionary Computation 1 (2011) 19–31.
5. M. Kárný, Towards On-Line Tuning of Adaptive-Agent's Multivariate Meta-Parameter, Pattern Recognition Letters, 150, 170-175, 2021