Invited talk together with B. Invited talk on " Ein auf evolutionaerer Mehrzieloptimierung basierender Ansatz zur Regularisierung neuronaler Netze " A method for neural network regularization based on evolutionary multi-objective optimization , Fachbereich Informatik , Lehrstuhl Systemanalyse Prof.
- Never Have Your Dog Stuffed: And Other Things Ive Learned.
- The Divine Imperative (A Study in Christian Ethics)?
- New Uses of Sulfur.
Invited talk on "Dynamic weighted aggregation: from multi-objective optimization to dynamic optimum tracking". Invited talk on "Evolutionary multi-objective optimization: Methods, analysis and applications". Deb and Prof. Conference on Robotics and Automation , Anchorage, May , PC member, CEC PC member, Int. Mieres , Asturias, Spain. PC member, 8th International Conf. PC member, The First Int. Luce, Martinique. PC member, GrC'07 , Nov. Reviewer, CEC , Sep. Reviewer, Int.
Benidorm , Spain. Last update May Please direct all inquires to Yaochu Jin. Related research topics include: Evolutionary optimization Data driven evolutionary optimization and decision-making Surrogate-assisted evolutionary optimization Robust and dynamic optimization, robustness over time Multi-objective and many-objective optimization Large-scale optimization Machine learning Multi-objective machine learning accuracy versus complexity, diversity, sparsity , and interpretability Secure machine learning robust and resilient machine learning against adversarial behaviors, learning with encrypted data, and cybersecurity intelligence.
NIC : Workshop on Nature Inspired Computing « Guide 2 Research
Advanced machine learning semi-supervised learning, transfer learning and knowledge incorporation, active learning, and ensemble learning Deep learning explainable deep learning, generative adversarial networks, and deep reinforcement learning Ensemble learning and drop-out learning Real world applications include Design optimization and control of complex systems, e.
The second motivational factor is the computational tractability. For instance, for a given optimization problem, multiple robust versions exist depend on the structure of the uncertainty set, therefore maintaining tractability is important. The classification models for RO includes local vs. Based on the nature of the problem, this technique is also known as min-max or worst-case approach.
It provides a good guaranteed solution for most possible realizations of the uncertainty in the data. It is also useful if some of the parameters belong to the estimation process and contains estimation errors. One important concept in defining and interpreting robustness and the resulting models is constraint robustness model robustness [ 5 ]. The application of RO in engineering is known as robust design optimization or reliability-based design optimization where the solutions remain feasible for all possible values of the uncertain inputs.
RO methodology is applicable to every generic optimization problem in which numerical data can be separated from the structure of the problems. The challenge of RO is that it gives the same weight and values to all of the uncertain parameters. The advantages of RO formulation are cost saving and increment of stability, qualitative and quantitative robustness. The practical usage of RO is that it does not significantly increase the complexity of the considered optimization problems in most cases [ 6 , 7 ].
Dynamic optimization DO , also known as dynamic programming is a process of finding the optimal control profile of one or more control parameters of a system. It is used to find the possible number of solutions for a given problem. There are several approaches of DO such as based on the calculus variations, deal with optimization discrete time and extend the static optimization. Basically, the process of DO implementation involves a system controller, a performance criterion and an algorithm to execute the control.
Two key attributes of DO are optimal substructure and overlapping sub-problems [ 8 ]. Four major steps on development of DO algorithm are: Characterize the structure of an optimal solution. The advantage of this paradigm: it performs the optimization recursively by dividing the problems into a collection of simpler sub-problems. Each sub-problem is solved only once using either top-down or bottom-up approach. To facilitate its lookup, a technique called memorization is applied where the solutions of subproblems are indexed based on its input parameter values, thereby solving computation time at the expense of modest expenditure in storage space.
Practically, the concept of DO is universal and flexible which can be applied to the execution of any effort [ 9 ]. Artificial intelligence AI has been viewed as a regulation in computer science. It has been developing and examining frameworks which work logically. Bio-inspired computation, metaheuristics and computational intelligence are the common examples of algorithms from numerous parts of AI.
Bio-inspired computation utilizes the computing power to demonstrate the living marvels. Computational intelligence which emphasizes on strategy and outcome can be broadly divided into five dominant fields: swarm intelligence, evolutionary computation, artificial neural networks, artificial immune system and fuzzy systems. This chapter will be focusing on a few swarm intelligence-based algorithms which are inspired by their natural processes. Swarm intelligence SI is evaluated as an adaptive strategy which takes collective intelligence as a behavior without centralized control structure on how an individual should behave.
The rules of SI are simple, self-organizing, co-evolution and being widely applied in the domains of optimizing, searching methods, research in DNA computing improvement, heating system planning etc. SI paradigm includes bird flocking, cuckoo search, animal herding and fish schooling etc. However, the two dominant subfields of SI are ant colony optimization, inspired by pheromone-trail of the ant behavior and particle swarm optimization, inspired by flocking and swarming behavior [ 10 ].
However, providing a complete review to all the swarm-based algorithms is rather impossible. The next sub-sections present the inspiration, working, metaphor and heuristic of eight popularly known swarm-based methods. These methods have been introduced and implemented in the last decade.
The main challenges of the field and their future trends have also been discussed. Bat algorithm BA [ 11 ] helps in simplicity and flexibility. It is found to be very efficient in handling nonlinear and multi objective issues. Bats have a special high-level capability of bio-sonar echolocation which is used to find their prey, obstacles, roosting crevices detection and discriminate different types of insects.
The efficiency of BA depends on the features below: Automatic zooming: this capability is performed based on the automatic switch from explorative direction to the local insensitive exploitation. Microbats are the famous examples among all the bat species. The echolocation attribute of microbat is used to model BA. Literature has reported a diverse range of BA applications such as loading pattern of nuclear core in engineering optimization, nonlinear economic dispatch problem, design of a power system stabilizer, size optimization for the skeletal structures which consist of truss and frame, multilevel image thresholding which is an image processing technique.
In the context of inverse problem and parameter estimation, bat calculations have been utilized in solving numerical improvement, advancing the brushless DC wheel engines, and enhancing topological shape in microelectronic applications [ 12 , 13 , 14 , 15 ]. There are some successful implementations of BA in SO. In their work of stochastic resonance for MR images enhancement [ 16 ], proposed a neuron model that tapped on the BA multi-objective optimization property to tune the parameters. In their work, the BA is utilized to maximize both the image performance indices contrast enhancement factor and the mean opinion score.
Their results show that the method has improved the gray-white matter differentiation, which has been found useful to diagnose MR images. In another work by [ 17 ], BA is adapted with inclusion of two operations— 1 iterative local search, and 2 stochastic inertial weight to improve its performance in terms of accuracy, speed and convergence stability. It is claimed that BA is easy to fall into local optima and has unstable optimization results due to low global exploration ability.
The authors overcome the weaknesses of BA when their iterative local search algorithm disturbs the local optimum and do some local re-search, such that the BA has better ability to get out of the local optima. Adding with their stochastic inertial weight to disrupt the velocity updating equation, it enhances the diversity and flexibility of bat population.
They proved their results based on 10 classic benchmark functions, CEC benchmark suite, and two 2 real-world problems, in which they concluded with improved performance. A robust tuning of power system stabilizer is demonstrated to be possible by using BA [ 18 ]. In such scenario of RO application, the stability of the power system is highly critical.
What is Nature Inspired Computing?
This paper proposed BA to optimize the gain and the pole-zero parameters of the stabilizer. The optimization was performed with objective function based on eigenvalue shifting to guarantee the stability of nonlinear plant for a wide range of operating conditions.
- Unconventional Computation: 5th International Conference, UC 2006, York, UK, September 4-8, 2006. Proceedings.
- Basics in Medical Education?
- Deathstalker War (Owen Deathstalker, Vol. 3)!
- Inspiration from the Nature;
- The Book: A Global History!
- Nature-Inspired Computation in Engineering.
A dynamic perceptive BA [ 19 ] is used to optimize particle filter for multiple targets tracking. This is an example of DO in which the authors proposed a multiple-maneuvering-target tracking algorithm and combined it with the BA to optimize particle filter typically used in a modern radar tracking system. This dynamic control of adjusting the particle filter, adding with a joint probabilistic data association has enabled an improved accuracy in target tracking even under a complex environment.
In other relevant applications, BA has been reported in data mining techniques of classifications and clustering. BA has been applied in grouping microarray information, minimization of make span and mean flow time to study half breed flow shop booking issues [ 20 ]. In the application of image processing, BA has been utilized for full body human stance estimation.
Related Nature-Inspired Computation in Engineering
Copyright 2019 - All Right Reserved