..::Research
 
My research interests include:
  • Evolutionary Algorithms for singleobjective and multiobjective optimization
    (see DEMO and BBDEMO)

  • Visualization of multidimensional approximation sets
    (see Visualization with prosections) and the empirical attainment function (see Visualization of the EAF)

  • Benchmarking multiobjective optimization algorithms (see the COCO framework on Github and the GBEA suite of real-world problems)

  • Machine learning methods for text processing

  • Outlier detection in access control systems

..::Tutorials on visualization in multiobjective optimization
 
You can download here the tutorials presented at:

..::DEMO
 
DEMO (Differential Evolution for Multiobjective Optimization) is an algorithm based on differential evolution for solving multiobjective optimization problems.

First experiments were made by comparing the performance of DEMO and some other algorithms on five ZDT test problems.

New experiments on DTLZ and WFG test problems were performed to enable a better comparison between DEMO and state-of-the-art algorithms for multiobjective optimization (NSGA-II, SPEA2 and IBEA).

You can also download the DEMO program that was used for these experiments and its source code (upgraded to v1.3 in October 2009). Let me know, if you find any bugs.

To learn more about DEMO see:
  • T. Tušar. Design of an algorithm for multiobjective optimization with differential evolution. (2007) M.Sc. Thesis
    download pdf bibtex

  • T. Tušar and B. Filipič. Differential evolution versus genetic algorithms in multiobjective optimization. In Proceedings of the Fourth International Conference on Evolutionary Multi-Criterion Optimization - EMO 2007, pp. 257-271. (2007)
    download pdf bibtex (© Springer)

  • T. Robič and B. Filipič. DEMO: Differential evolution for multiobjective optimization. In Proceedings of the Third International Conference on Evolutionary Multi-Criterion Optimization - EMO 2005, pp. 520-533. (2005)
    download pdf bibtex (© Springer)

..::BBDEMO
 
BBDEMO (Black Box Differential Evolution for Multiobjective Optimization) is an algorithm based on the BBDE and the DEMO algorithms. The hypervolume improvement is computed as in the COMO-CMA-ES algorithm.

You can download the source code of the BBDEMO algorithm and its results on the new COCO test problems.

To learn more about BBDEMO see:
  • T. Tušar. Algorithm Results on the New COCO Test Problems. Jozef Stefan Institute, IJS-DP 12992, 2019.
    download pdf bibtex

..::Visualization with prosections
 
With prosections (projections of a section) 4D approximation sets can be plotted in 3D in a simple and intuitive way.

First visualization examples were presented at GECCO 2011.

More recent (and sophisticated) visalizations can be found in the TEVC article. All approximation sets and accompanying gnuplot scripts from the article can be downloaded as a zip or tgz file.

To learn more about visualization with prosections see:
  • T. Tušar. Visualizing Solution Sets in Multiobjective Optimization. Ph.D. Thesis, 2014.
    download pdf bibtex

  • T. Tušar and B. Filipič. Visualization of Pareto front approximations in evolutionary multiobjective optimization: A critical review and the prosection method. IEEE Transactions on Evolutionary Computation, 19(2):225-245, 2015.
    download pdf bibtex (Open Access)

  • T. Tušar and B. Filipič. Scaling and visualizing multiobjective optimization test problems with knees. In Proceedings of the 15th International Multiconference Information Society - IS 2012, Volume A, pp. 155-158, 2012.
    download pdf bibtex

  • T. Tušar and B. Filipič. Visualizing 4D approximation sets of multiobjective optimizers with prosections. In Proceedings of the 13th Annual Genetic and Evolutionary Computation Conference - GECCO'11, pp. 737-744, 2011.
    download pdf bibtex (© ACM)

..::Visualization of the empirical attainment function (EAF)
 
The EAF is able to describe the probabilistic distribution of multiple approximation sets and can be therefore used to analyze and compare the performance of stochastic multiobjective optimization algorithms. See Attainment Function Tools and EAF Graphical Tools for more information on the EAF and tools to support its computation (2D and 3D cases) and visualization (2D case).

We have tackled the visualization of 3D EAF values and differences in the exact as well as the approximated case. To learn more about this see:
  • T. Tušar. Visualizing Solution Sets in Multiobjective Optimization. Ph.D. Thesis, 2014.
    download pdf bibtex

  • T. Tušar and B. Filipič. Visualizing exact and approximated 3D empirical attainment functions. Mathematical Problems in Engineering, 2014:Article ID 569346, 18 pages, 2014.
    download pdf bibtex (Open Access)

  • T. Tušar and B. Filipič. Initial experiments in visualization of empirical attainment function differences using maximum intensity projection. In GECCO 2014 companion: Genetic and Evolutionary Computation Conference Companion, pp. 1099-1105, 2014.
    download pdf bibtex (© ACM)

  • T. Tušar and B. Filipič. An approach to visualizing the 3D empirical attainment function. In GECCO 2013 companion: Genetic and Evolutionary Computation Conference Companion, pp. 1367-1372, 2013.
    download pdf bibtex (© ACM)
..::publications ..::about me ..::research