Editorial: Neural inspired computational intelligence and its applications

Ying Tan, Yuhui Shi, Yan PeiFrontiers in Neuroscience 2024 · DOI · OpenAlex

This editorial provides a comprehensive overview of the special Research Topic "Neural inspired computational intelligence and its applications," published in Frontiers in Neuroscience. The collection bridges the gap between biological neural mechanisms and artificial intelligence, demonstrating how "bio-inspiration" leads to more robust, efficient, and generalizable algorithms.

The Challenge of Modern AI

While Artificial Intelligence (AI) has seen explosive growth, current models often struggle with efficiency, generalization to unseen data, and the complexity of real-world constraints. Biological systems, conversely, have evolved elegant solutions for navigation, recognition, and learning. The authors of this editorial argue that Computational Intelligence (CI) must look "under the hood" of biological processes to overcome current bottlenecks, such as the difficulty of estimating object rotation in robotics or the lack of biological plausibility in neural network context processing.

Methodological Innovations

The editorial summarizes five distinct methodological approaches that leverage biological principles:

  • Circular Latent Spaces for Robotics: Plebe and Da Lio propose a neural network for rotation estimation where the latent space is structured as a circle. This mimics the "ellipsoid body" in the brain of Drosophila (fruit flies), where physical shifts in neural activity correspond directly to angular rotations. This geometric alignment allows the model to map latent shifts directly to angular rotations in world space.
  • Dendritic Plasticity Modeling: Baronig and Legenstein investigate "apical dendrites" in pyramidal neurons. They model how local synaptic plasticity and calcium ($Ca^{2+}$) events allow individual neurons to integrate top-down contextual information, facilitating continual learning without the "catastrophic forgetting" typical of standard AI.
  • Feature Importance in Vision: Zhang et al. use Deep Convolutional Neural Networks (CNNs) to dissect face perception, analyzing how networks prioritize different facial features compared to human biological vision.
  • Constrained Swarm Intelligence: Liu et al. adapt swarm intelligence—algorithms inspired by the collective behavior of social insects—to handle high-dimensional optimization problems in supply chain management involving complex "cross-dimensional constraints."
  • Mechanistic Theory Building: Haiping Huang identifies eight fundamental challenges (e.g., causal learning, adversarial robustness, and internal models of the brain) to build a "bottom-up" roadmap for a comprehensive theory of intelligence.

Key Results and Insights

Superior Generalization

The Drosophila-inspired rotation model outperformed three state-of-the-art CNN backbones commonly used in robotics. Remarkably, it achieved accurate rotation estimation for object categories it had never encountered during training, proving that structuring a network’s internal representation to match the geometry of the task (circularity for rotation) is key to zero-shot generalization.

Biological Alignment in Vision

Corroborating human psychological studies, the CNN research suggests that features like eyebrows may be more critical for identification than the eyes themselves. This suggests a shared hierarchy in biological and artificial vision systems.

Efficiency in Optimization

The swarm intelligence approach proved highly effective at solving complex supply chain problems that typically paralyze standard optimization algorithms due to high dimensionality and cross-dimensional constraints.

Limitations and Future Directions

While these bio-inspired models are promising, the authors acknowledge that they must be scaled to work with Big Data, the Internet of Things (IoT), and Cloud Computing to be viable in "smart city" applications.

A significant future challenge remains the "mechanics of subjective experience." While we can model the mechanics of synaptic plasticity or rotation estimation, bridging the gap between mechanistic neural firing and the actual feeling of intelligence remains a primary hurdle. Addressing the eight challenges identified by Huang—including causal learning and internal brain models—will be essential for the next generation of artificial and biological intelligence research.

Summary updated

Abstract

the rotation angle of unknown objects from RGB images. Inspired by the rotational representation in the ellipsoid body of Drosophila, the proposed model embeds the understanding of rotational transformations into its architecture. The network's latent space is structured in a circular manner to effectively capture the cyclic nature of rotation. The rotation operator acts as a shift in the circular latent space's units, establishing a direct correspondence between shifts in the latent space and angular rotations of the object in the world space. Remarkably, the model accurately estimates the difference in rotation between two views of an object, even for categories of objects it has never seen before. Furthermore, it outperforms three state-of-the-art convolutional networks commonly used as the backbone for vision-based models in robotics. This innovative approach not only enhances our understanding of rotational transformations but also sets a new benchmark for performance in this domain. The fourth article, entitled "An efficient swarm intelligence approach to the optimization on high-dimensional solutions with cross-dimensional constraints, with applications in supply chain management," by Hsin-Ping Liu, Frederick Kin Hing Phoa, Yun-Heh Chen-Burger, Shau-Ping Lin, explores the application of bio-inspired swarm intelligence to complex optimization problems. The proposed approach addresses high-dimensional solutions with cross-dimensional constraints, demonstrating its effectiveness in supply chain management scenarios. This research highlights the versatility and practical utility of bio-inspired computational methods in solving real-world problems, particularly in the context of e-commerce and supply chain optimization.Finally, the fifth article, entitled "Eight challenges in developing theory of intelligence," by Haiping Huang, identifies and discusses key challenges in developing a comprehensive theory of intelligence. The article emphasizes the importance of bottom-up mechanistic modeling in understanding both natural and artificial intelligence. The identified challenges include representation learning, generalization, adversarial robustness, continual learning, causal learning, internal model of the brain, next-token prediction, and the mechanics of subjective experience. Addressing these challenges provides a roadmap for future research in both artificial and biological intelligence, emphasizing the need for interdisciplinary collaboration and innovative approaches.In summary, these articles showcase the cutting-edge advancements in neural-inspired computational intelligence and its applications. The insights and innovations presented not only summarize the current state of the field but also highlight the broad application prospects and future directions. From enhancing robotic vision systems to optimizing complex supply chain networks, and from understanding the intricacies of face perception to developing a comprehensive theory of intelligence, the potential applications are vast and varied. We hope that the insights and innovations presented in this special issue will inspire further research and development in the exciting and rapidly evolving fields, paving the way for new breakthroughs and practical solutions in computational intelligence.
BibTeX
@inproceedings{tan2024editorial,
  author = {Ying Tan and Yuhui Shi and Yan Pei},
  title = {{Editorial: Neural inspired computational intelligence and its applications}},
  booktitle = {Frontiers in Neuroscience},
  year = {2024},
  doi = {10.3389/fnins.2024.1515039},
}
Figure