Special issue on neural computing and applications 2021

Kai Liu, Jingjing Cao, Yimin Yang, Wun‐She Yap, Rui Tan, Zenghui WangNeural 2022 · DOI · OpenAlex

Introduction

This editorial presents a curated selection of high-impact research from the 2021 International Conference on Neural Computing for Advanced Applications (NCAA 2021), published in a special issue of Neural Computing and Applications. Out of 144 original submissions, only 12 papers (8.3%) were selected for this collection, representing the vanguard of machine learning theory and its diverse applications ranging from genomics to urban logistics.

Addressing Core Challenges in Neural Computing

The field of neural computing faces several persistent hurdles that this special issue aims to address. These include the high computational overhead of complex models, the difficulty of training robust models on small or high-dimensional datasets (particularly in genomics), and the challenge of applying theoretical models to dynamic, real-world environments like building energy management or multi-agent robotics.

Methodological Innovations

The selected papers introduce several novel techniques across various domains:

  • Optimization & Classification: The Sparse Discriminant Twin Support Vector Machine (SD-TSVM) utilizes Fisher regularization and 1-norm coefficients to enhance model sparsity and classification accuracy.
  • Fuzzy Logic: New granular Takagi–Sugeno (T-S) fuzzy rule-based models improve interpretability through strategic space partitioning.
  • Computer Vision: A graph filtering approach is introduced to mitigate high-frequency noise in multi-view data, alongside a non-iterative clustering framework for contour-respecting superpixel decomposition.
  • Generative Modeling: The Gene-CWGAN, a conditional Wasserstein GAN with gradient punishment, addresses data scarcity in gene expression analysis by generating high-quality synthetic samples.
  • Natural Language Processing (NLP): FastMCN optimizes self-attention mechanisms for medical concept normalization, significantly reducing computational costs for specialized literature.

Diverse Benchmarking and Real-World Datasets

The validity of these methods was established using a broad range of datasets:

  • General Machine Learning: Nine real-world datasets validated the SD-TSVM.
  • Optimization: 29 benchmark functions were used for multi-objective optimization (MOPs).
  • Energy & Physics: Practical building load datasets and the Lorenz chaotic time-series were used for forecasting tasks.
  • Biomedical: Five public gene expression profile datasets and large-scale biomedical literature supported the genomic and NLP research.
  • Robotics: Simulated environments facilitated research into multi-taxi cruising and wheeled mobile robot formations.

Key Insights: Sparsity, Generative Models, and Efficiency

Several critical insights emerge from this collection:

  1. Functional Sparsity: 1-norm regularization in models like SD-TSVM does more than reduce size; it improves discriminative power by focusing on the most relevant features.
  2. Augmenting Small Data: Generative models like Gene-CWGAN provide a viable path for training robust classifiers in domains like rare disease research where data is naturally scarce.
  3. Optimized Attention: The success of FastMCN suggests that the "attention" mechanisms powering modern AI can be tailored for specialized, resource-constrained domains without sacrificing accuracy.
  4. Structure-Aware Processing: Applying signal-processing filters to graph-based data representations significantly improves the quality of multi-view clustering by removing high-frequency noise.

Conclusion and Future Outlook

While the special issue highlights significant progress, it also underscores the need for further research into the scalability of graph-based methods, the transition of analytical models into unpredictable urban environments, and the ongoing quest for model interpretability in deep learning. These 12 papers provide a robust foundation for the next generation of neural computing applications.

Summary updated

BibTeX
@inproceedings{liu2022special,
  author = {Kai Liu and Jingjing Cao and Yimin Yang and Wun‐She Yap and Rui Tan and Zenghui Wang},
  title = {{Special issue on neural computing and applications 2021}},
  booktitle = {Neural},
  year = {2022},
  doi = {10.1007/s00521-022-07732-1},
}

Loading related content...

Figure