Methods and Tools for Big Data Solar Physics
The focus of this talk is on big data analytics of the Solar Observatory (www.nasa.gov/sdo/), which is a flagship of NASA's current "Living with a Star" program. The audience will first learn about the importance of solar data analysis, then about the complexity of data maintained on the servers in our Data Mining Lab (dmlab.cs.montana.edu/). After that, three ongoing research projects will be introduced: (1) the Content-based Image Retrieval (CBIR) system for solar data, (2) the development of machine learning techniques for automated validation of NASA's software modules dedicated to the recognition of popular types of solar phenomena, and (3) the search for spatio-temporal co-occurrence patterns among different types of solar activity. Finally, we will briefly talk about the future of our solar databases and data mining projects.
Multi-Objective Evolutionary Fuzzy Rule-Based Classifier Design
Design of fuzzy rule-based systems involves conflicting objectives such as interpretability maximization and accuracy maximization. For example, linguistic interpretability of fuzzy rule-based systems can be improved by decreasing the number of fuzzy rules, the number of antecedent conditions in each fuzzy rule, and the complexity of fuzzy partition of each input variable. However, those interpretability improvement efforts often degrade the accuracy of fuzzy rule-based systems. That is, very simple fuzzy rule-based systems with high linguistic interpretability are usually not highly accurate. In the 1990s, conflicting objectives were combined into a single integrated objective function, which was optimized by a single-objective optimization algorithm. Currently those objectives are handled as different objectives and simultaneously optimized by an evolutionary multi-objective optimization (EMO) algorithm. A large number of non-dominated fuzzy rule-based systems are obtained from a single run of an EMO algorithm. In this talk, we focus on fuzzy rule-based classifier design. This talk starts with a brief review of well-known highly-cited studies related to fuzzy rule-based classifier design from two view points: Accuracy maximization and interpretability maximization. Next fuzzy rule-based classifier design is explained as multi-objective optimization problems to which EMO algorithms are applied. Then we discuss how to choose a single final solution from a large number of obtained non-dominated fuzzy rule-based classifiers . Finally we explain parallel distributed implementation of evolutionary algorithms for drastically decreasing the computation time of evolutionary fuzzy rule-based classifier design .
 M. Fazzolari, R. Alcalá, Y. Nojima, H. Ishibuchi, and F. Herrera, “A review of the application of multiobjective evolutionary fuzzy systems: Current status and further directions,” IEEE Trans. on Fuzzy Systems, vol. 21, no. 1, pp. 45-65, February 2013.
 H. Ishibuchi and Y. Nojima, “Repeated double cross-validation for choosing a single solution in evolutionary multi-objective fuzzy classifier design,” Knowledge-Based Systems, vol. 54, pp. 22-31, December 2013.
 H. Ishibuchi, S. Mihara, and Y. Nojima, “Parallel distributed hybrid fuzzy GBML models with rule set migration and training data rotation,” IEEE Trans. on Fuzzy Systems, vol. 21, no. 2, pp. 355-368, April 2013
Self-Learning Control of Nonlinear Systems based on Iterative Adaptive Dynamic Programming Approach
The optimal control of nonlinear systems often requires solving the nonlinear Hamilton-Jacobi-Bellman (HJB) equation instead of the Riccati equation as in the linear case. The discrete-time HJB (DTHJB) equation is more difficult to work with than the Riccati equation because it involves solving nonlinear partial difference equations. Though dynamic programming has been a useful computational technique in solving optimal control problems for many years, it is often computationally untenable to run it to obtain the optimal solution, due to the backward numerical process required for its solutions, i.e., the well-known "curse of dimensionality". A self-learning control scheme for unknown nonlinear discrete-time systems is developed for this purpose. An iterative adaptive dynamic programming algorithm via globalized dual heuristic programming technique is developed to obtain the optimal controller with convergence analysis. Neural networks are used as parametric structures to facilitate the implementation of the iterative algorithm, which will approximate at each iteration the cost function, the optimal control law, and the unknown nonlinear system, respectively. Simulation examples are provided to verify the effectiveness of the present self-learning control approach.
Developments in medical image retrieval
Medical imaging produces enormous amounts of data in health care institutions and is essential in diagnosis and treatment planning. Being able to use the knowledge stored in past cases respecting patient privacy is the goal of many medical image retrieval approaches that use a mix of free text, structured data and image analysis (or CBIR, content-based image retrieval) to navigate in the large institutional archives or make links with the literature and resources available on the Internet. In this talk I will present several developments in medical image retrieval that is in the process to show real medical usefulness. These steps will include the analysis of multi-dimensional data, the links between semantics and image data and also the next steps that will likely include combinations of genomics and image data. Using visual data out of their context seems to be one of the main problems of many past approaches.
Discovering Components of the Primary Language
Linguistic Geometry (LG), developed since 1972, is a type of game theory scalable to solving complex real world problems that are considered intractable by conventional approaches. I will introduce participants to the modern applications of LG related to the US national defense that generate, in real time, courses of action exceeding the level of human commanders. Then I will look back in history and link LG to the ancient warfare, such as the battles of Alexander the Great and Hannibal, to reveal its impact on development of human intelligence. I will provide relationship of LG to the hypothesis of the Primary Language of the human brain (as introduced by J. von Neumann in 1957) and its components. I will suggest that the Primary Language includes at least two major components critical for humanity, LG and the Algorithm of Discovery; they should have common age and means. This suggestion is based on the hypothesis that there is a universal Algorithm of Discovery, i.e., the algorithm for developing new algorithms, driving all the innovations and, certainly, the advances in all sciences. Presentation of the preliminary results on discovering the Algorithm of Discovery will complete this talk. (More details about LG and the Algorithm of Discovery will be given in the tutorial.)
Towards Human-Like Intelligence: From Self-Organizing Neural Networks to Integrated Cognitive Architecture
Towards Human-Like Intelligence: From Self-Organizing Neural Networks to Integrated Cognitive Architectures
Human intelligence involves a complex interplay of functions, notably self-awareness, knowledge, reasoning, learning, and problem solving. This talk will present a family of self-organizing neural networks, collectively known as fusion Adaptive Resonance Theory (fusion ART) (Tan et al., 2007) for simulating high level cognitive functions. By extending the original ART models consisting of a single pattern field into a multi-channel architecture, fusion ART unifies a number of important neural models developed over the past decades, including the original ART networks for unsupervised learning, Adaptive Resonance Associative Map (ARAM) for supervised learning and pattern recognition (Tan, 1995), and Fusion Architecture for Learning and Cognition (FALCON) (Tan et al., 2008), for reinforcement learning and real-time decision making.
Following the notion of embodied cognition (Anderson, 2003), this talk will further show how fusion ART, encompassing a universal set of neural coding and adaptation principles, can be used as a building block of autonomous systems, integrating self-awareness, memory, emotion, planning, and deliberative behaviour (Tan et al., 2010; Wang et al., 2013). Several case studies will be presented, illustrating how such cognitive autonomous systems may be used in the domains of command and control (Feng et al., 2008), first-person shooting game (Wang et al., 2009), adaptive Computer Generated Forces (CGF) (Teng et al., 2013), and modelling of human-like characters in virtual environment (Kang et al., 2012).
Anderson, M. (2003). Embodied cognition: A field guide. Artificial Intelligence, 149: 91-130.
Subagdja, B. and Tan, A.-H. (2012). iFALCON: A Neural Architecture for Hierarchical Planning. Neurocomputing, 86: 124-139.
Feng, Y.-H., Teng, T.-H. and Tan, A.-H. (2008) Modelling Situation Awareness for Context-aware Decision Support. Expert Systems with Applications, 36(1): 455-463.
Kang, Y., Subagdja, B., Tan, A.-H., Ong, Y.-S. and Miao, C.-Y. (2012) Virtual Characters in Agent-Augmented Co-Space. In Proceedings, Eleventh International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2012) (Demo Track), pp. 1465-1466, Valencia, Spain.
Tan, A.-H. (1995). Adaptive Resonance Associative Map. Neural Networks, 8(3): 437-446.
Tan, A.-H., Carpenter, G. A. and Grossberg, S. (2007). Intelligence Through Interaction: Towards A Unified Theory for Learning . In LNCS 4491, Part I, pp. 1098-1107.
Tan, A.-H., Feng, Y.-H. and Ong, Y.-S. (2010) A Self-Organizing Neural Architecture Integrating Desire, Intension and Reinforcement Learning. Neurocomputing, 73(7-9): 1465-1477.
Tan, A.-H., Lu, N. and Xiao, D. (2008). Integrating Temporal Difference Methods and Self-Organizing Neural Networks for Reinforcement Learning with Delayed Evaluative Feedback. IEEE Transactions on Neural Networks, 9(2): 230-244.
Teng, T.-H., Tan, A.-H. and Teow, L.-N. (2013). Adaptive computer-generated forces for simulator-based training. Expert Systems with Applications, 40(18): 7341-7353.
Wang, D., Subagdja, B., Tan, A.-H. and Ng, G.-W. (2009) Creating Human-like Autonomous Players in Real-time First Person Shooter Computer Games. In proceedings, Twenty-First Annual Conference on Innovative Applications of Artificial Intelligence (IAAI'09), Pasadena, California.
Wang, W., Subagdja, B., Tan, A.-H., and Starzyk, J.A. (2012) Neural Modeling of Episodic Memory: Encoding, Retrieval, and Forgetting. IEEE Transactions on Neural Networks and Learning Systems, 23(10): 1574-1586.
State-Of-The-Art Many Objective Evolutionary Algorithms And Its Applications In Weather Forecasting
Gary Yen, Ph.D., FIEEE, FIET
Oklahoma State University School of Electrical and Computer Engineering
Evolutionary computation is the study of biologically motivated computational paradigms which exert novel ideas and inspiration from natural evolution and adaptation. The applications of population-based heuristics in solving multiobjective optimization problems have been receiving a growing attention. To search for a family of Pareto optimal solutions based on nature-inspiring problem solving paradigms, Evolutionary Multiobjective Optimization Algorithms have been successfully exploited to solve optimization problems in which the fitness measures and even constraints are uncertain and changed over time.
When encounter optimization problems with many objectives, nearly all designs performs poorly because of loss of selection pressure in fitness evaluation solely based upon Pareto optimality principle. This talk will survey recently published literature along this line of research- evolutionary algorithm for many-objective optimization and its real-world applications in data mining area. The list includes, but not limited to, multiobjective evolutionary algorithm based on decomposition (MOEA/D), e-dominance based multiobjective evolutionary algorithm (e -MOEA), preference order based genetic algorithm (POGA), territory defining evolutionary algorithm (TDEA), hypervolume estimation algorithm (HypE), grid-based evolutionary algorithm (GrEA), and fuzzy-based Pareto optimality evolutionary algorithm. At the end of my presentation, I will touch upon a successfully application in weather forecasting.
Fuzzy Models and Granular Fuzzy Models:
Pursuing New Avenues of Computational Intelligence
Department of Electrical & Computer Engineering
University of Alberta, Edmonton, Canada
Systems Research Institute, Polish Academy of Sciences
Fuzzy models (controllers, classifiers, predictors…) have been at the forefront of the developments of the technology of fuzzy sets. We have been witnessing emergence of new constructs within this realm including interval-valued and type-2 fuzzy models.
Granular Computing offers a unified and coherent environment of processing information granules (regardless of their formal settings, say fuzzy sets, rough sets, intervals) and in this way could be of interest to study in the context of system modeling.
In particular, we discuss three fundamental and overarching concepts of Granular Computing essential to the analysis and synthesis of fuzzy models: (i) the principle of justifiable granularity supporting a way of forming information granules, (ii) exploitation of information granularity regarded as an essential design asset when forming models that are more in rapport with the complexity of real-world systems, and (iii) realization of higher type and higher order information granules in the description of hierarchical and distributed modeling architectures. In system modeling, information granules (say, fuzzy sets) of higher order and higher type form one of the interesting conceptual and methodological pursuits. We highlight key motivating factors behind the emergence of type-2 and order-2 information granules and reveal apparent linkages between type-n information granules and hierarchical architectures of fuzzy models.