Research Interests

Professor Deb's research is primarily focussed on developing new and efficient algorithms and applying them to optimization problems in engineering design. He is one of the senior researchers in the field of Genetic and Evolutionary Algorithms and has worked on this field for more than 21 years with other experts (of the field) from around the globe. In particular, Dr Deb has significantly contributed to the following areas of research:

  • Multi-Objective Evolutionary Algorithms

  • The field of Multi-Objective Evolutionary Algorithms is one of the hottest fields of research today and Prof Deb is a pioneer in this area of research. Starting from an early multi-objective EA, Prof Deb, along with other researchers across the world, have since then tackled some key research issues, which are now making the topic a new emerging field of research and application. Today the name Kalyanmoy Deb epitomizes Multi-Objective GAs and researchers from all over the world look up to him as one of the authorities for his immense research, experience and success in this field of research, which is also his personal favourite. Among his outstanding contributions to the MOEA community are the following:

    1. Non-dominated Sorting Genetic Algorithm (NSGA)
    Prof. Deb developed and proposed this algorithm with his students way back in 1994. Soon it became a synonym for Multi-Objective GA (although other algorithms also existed then) for its neat idea and outstanding performance. Many researchers wrote to Prof. Deb about its excellent performance and wide usability but complained slightly about its high execution time. Prof. Deb made sure to rectify this problem in his next multi-objective venture.

    2. Innovization: Innovation Through Optimization
    Prof. Deb has recently proposed a new and systematic design procedure by which innovative design principles can be obtained by performing a multi-objective optimization and by making a post-optimality analysis. In its simplest form, two conflicting objectives of design (such as minimization of size and maximization of output) can be considered and a Pareto-optimal front can be obtained using an EMO (such as NSGA-II). Thereafter, the obtained solutions can be analyzed to decipher common relationships among design variables and objectives. Such information are often found to provide useful design principles which cannot be obtained by any other means. For more information, look here.

    3. Non-dominated Sorting Genetic Algorithm-II (NSGA-II)
    This algorithm was born as a result of numerous experiments and outstanding teamwork of the then KanGAL team (in early 2000) under the expert guidance of Prof. Deb. Everyone had something to offer to NSGA-II. After a number of experiments conducted in KanGAL, it was established that on one hand NSGA-II performed better than any other existing MOEA in all test functions and on the other hand several remarkable results were obtained on application of NSGA-II to engineering design problems with a multi-objective flavour. The coming of NSGA-II started a surge of research and experiments in KanGAL and resulted in a number of research papers and theses -- not to say that KanGAL has still not completely recovered from the NSGA-II fever. Students in KanGAL are today trying to apply NSGA-II to far-fetched real world problems (a venture that could hardly have been approved by Prof Deb sometime ago, but for the immense success and growing popularity of NSGA-II). It was this very algorithm and his care for the field of MOEAs that served as an inspiration for Prof. Deb's recent book on Multi-Objective Evolutionary Algorithms which is the only comprehensive source of multi-objective evolutionary optimization. Prof. Deb also envisioned and was one of the organizers of the First Conference on Multi-Objective Evolutionary Optimization (EMO-2001).

    3. Proposer of a New Philosophy of Practical Optimization
    Besides the description of Multi-Objective Optimization algorithms and introduction to Evolutionary Algorithms and Multi-Modal optimization, Prof Deb outlines a completely different philosophy of practical optimization in his recent book on Multi-Objective Evolutionary Optimization. Most optimization books and schools of thought consider multi-objective optimization as a special application case study of single-objective optimization procedures. Since multiple objectives must be converted into a single objective using the classical means of solving multi-objective optimization problems, this made sense. Prof Deb's two-stage ideal multi-objective optimization concept is quite contrary to this traditional concept. Although intuitive in concept, Prof Deb is the first to propose that single-objective optimization is a degenerate case of multi-objective optimization and clearly shows in his book how Single-Objective Optimization, Multi-Modal Optimization, and Goal Programming problems can be solved using his two-stage ideal multi-objective optimization concept.

  • Real-Parameter Genetic Algorithms

  • Prof. Deb proposed and developed one of the earliest and efficient real coded GA (which is widely used today) namely the Simulated Binary Crossover Operator (SBX) in 1995 with his students. SBX imitates the working principle of a binary genetic algorithm in real paradigm. The idea of SBX is simple to conceptualize, almost intuitive and the algorithm is easy to code. The main motivation for developing this real-parameter GA came from his endeavor in trying to make GAs more applicable in Engineering and Scientific Optimization problems, which mostly deal with real parameters. The property of Self Adaptation (which SBX possesses) ensures the attainment of optima even in functions where optima varies with time. In other words, SBX chases the function optima wherever it goes in the search space.
    In 1999, Prof Deb conducted a comprehensive research in Germany which included a survey of the behaviour of the various real coded evolutionary algorithms (namely the various types of Evolution Strategies and Real coded GAs (SBX in particular)) and established a similarity in performance of both. Along with researchers from Evolution Strategy, he established the fact that all Real-Parameter Evolutionary Algorithms (namely Evolution Strategies and the various real-parameter genetic algorithms) are inherently identical techniques to achieve the same end (that is to achieve the optima), although they may have remarkably different means of achieving the end. Currently, further research to this effect is going on in his lab (KanGAL).
    In the last few months Prof. Deb and his students have developed a generic multi-parent recombination operator (Parent Centric Crossover) which performs an order of magnitude better than existing multi-parent recombination schemes (UNDX and SPX) under a steady state generation alteration model (G3 model) when applied on some tough test problems. Further experiments and extensions of this novel approach are awaited from him and his students in time to come.

  • Constrained Nonlinear Optimization

  • Most real-world search and optimization problems involve constraints and often the so called feasible search region is so small compared to the overall search space that a randomly created solution is quite likely to be infeasible. In such problems, it is necessary to have an algorithm which will devise a strategy to search through infeasible regions in the hope of reaching the feasible region. Prof Deb, in 1998, showed with his Constraint-Tournament Strategy, how feasible and infeasible solutions can be relatively emphasized in a population-based evolutionary algorithm to reach the true constrained optima. The interesting aspect of the proposed technique is that it is simple and does not require any extra parameter (such as a Penalty Parameter or Lagrange multiplier) to make correct decisions. The success of the approach in single-objective optimization problems motivated Prof Deb and his students to use a similar constraint handling technique in their multi-objective optimization algorithm - NSGA-II.

  • Other Research Interests

  • In addition to these, Prof Deb is interested in the theoretical properties of evolutionary algorithms, such as convergence analysis, population sizing, linkage issues leading to representation-operator interactions, and test problem design. He is equally interested in applications of evolutionary algorithms in engineering and scientific problems and developing Hybrid Optimization Algorithms with evolutionary and classical search methods. He is also interested in Robotics and Artificial Intelligence Techniques.

  • Past Research Studies

  • Prof Deb is also well known for his earlier research studies on Multi-Modal Optimization (the task of finding multiple optimal solutions (local and global alike)) and Messy Genetic Algorithms developed for better understanding of the working of a genetic algorithm and to solve complex optimization problems.

  • OVERALL RESEARCH MOTTO

  • Evolutionary algorithms had been proposed more than 35 years ago and practiced with full rigor for more than 15 years. However, it is still not clear in which types of problems evolutionary algorithms outperform other contemporary methods. Prof Deb strongly believes the populations approach of evolutionary algorithms makes them unparallelled and unique in solving Multi-Modal, Multi-Objective, and Constrained Optimization Problems. The aim of his future research (with theoretical and application studies) is to identify other search and optimization problems in which evolutionary algorithms have a niche over their traditional counterparts. Prof Deb is also interested in designing evolutionary algorithms with minimal number of user-defined parameters. As is evident from his past research, Prof Deb believes `simplicity is the key to any successful research'.