CIS Publication Spotlight [Publication Spotlight]

Derong Liu, Chin-Teng Lin, Kay Chen Tan, Graham Kendall, Angelo CangelosiIEEE Comput. Intell. Mag 2015 · DOI

The August 2015 issue of the IEEE Computational Intelligence Magazine features the CIS Publication Spotlight, a curated selection of high-impact research from the society's flagship journals. This spotlight distills complex breakthroughs in neural networks, fuzzy systems, and evolutionary computation into accessible insights, addressing the challenge of interdisciplinary information overload.

The CIS Publication Spotlight highlights converging research from across the field of computational intelligence.

Preserving Data Integrity and Optimizing Control

One of the primary challenges in modern data science is performing component analysis on bivariate data without losing the physical meaning or internal correlation of the sources. Researchers introduced the Correlation Preserving Transform (CPT), which utilizes augmented complex statistics and conjugate eigenvectors to maintain data channel order and mutual correlation—a significant step forward from standard uncorrelating techniques.

In the realm of control theory, the spotlight features work on achieving optimal control in "black box" nonlinear systems. By using Shunting Inhibitory Artificial Neural Networks (SIANN) to categorize data and Recurrent Neural Networks (RNN) to reconstruct unknown dynamics, researchers have mathematically proven that these systems remain stable even in the presence of model errors.

Bridging the Human-Machine Communication Gap

A standout achievement in the spotlight is the GALiWeather system. This system bridges the gap between raw meteorological data and human-readable descriptions. Using "Computing with Perceptions" and Natural Language Generation (NLG), GALiWeather automates textual weather reporting for the Galician region. In expert evaluations, the system earned a score of 4.83 out of 5, indicating its output is nearly indistinguishable from forecasts written by professional meteorologists.

Complementing this is research into Observer-Biased Fuzzy Clustering. This approach suggests that data clustering should incorporate the "position" of the observer. By using an observer metaphor as a point of interest, algorithms can produce more meaningful, "zoomed-in" partitions of data, making the results more relevant to specific user needs.

Bio-Inspired Vision and the Complexity of Evolution

In robotics, the BITPIC architecture addresses "information overwhelm" by mimicking biological eyes. Instead of processing an entire field of view, the system uses a limited "fovea" that moves actively to prioritize relevant data. This "Ecological Active Vision" allows robots to solve search-and-reach tasks more efficiently by filtering out visual noise.

The spotlight also explores the limits of evolutionary algorithms through Many-Objective Optimization. Testing on knapsack problems with up to 10 objectives revealed that standard algorithms like NSGA-II struggle as complexity increases, unless objectives are highly correlated. This finding underscores the urgent need for new diversity-improvement strategies in complex optimization tasks.

Proven Results and Future Standards

The featured research provides concrete proofs in game theory, specifically for the game of Amazons. Using enhanced solvers, researchers proved that a 5x6 board is a first-player win, while a 6x4 board favors the second player.

Despite these successes, the spotlight identifies critical gaps. A survey on parameter control in evolutionary algorithms revealed methodological shortcomings, calling for more rigorous standards. Furthermore, the issue concludes with a call for standardized Computational Intelligence Software to help transition these theoretical breakthroughs into production-ready applications.

We value your feedback on these research highlights.

Summary updated

BibTeX
@article{liu2015cis,
  author = {Derong Liu and Chin-Teng Lin and Kay Chen Tan and Graham Kendall and Angelo Cangelosi},
  title = {{CIS Publication Spotlight [Publication Spotlight]}},
  journal = {IEEE Comput. Intell. Mag.},
  year = {2015},
  doi = {10.1109/mci.2015.2472117},
}
Figure