6th International Workshop on New Approaches for


Multidimensional Signal Processing


NAMSP 2025

Technical University of Sofia, Sofia, Bulgaria

July 24-26, 2025

Plenary Speakers


Prof. Dr. Wenfeng Wang, International Academy of Visual Art and Engineering, London, UK


Title of Lecture:

Spatial Meta-Learning: Theory, Algorithm and Applications


Abstract: We will introduce how metalearning process can be integrated with spatial knowledge and quickly adapt to new tasks. Although deep learning has advantages in data processing capabilities and automatic feature extraction, it still faces many problems such as poor robustness and generalization, difficulty in learning and adapting to unobserved tasks. In recent years, the development of meta learning in deep learning has provided a new perspective for solving the above problems. Meta learning can learn new abilities by training a small number of samples, quickly adapt to new environments, and thus improve the model's generalization ability. Due to the limitations of slow learning speed and susceptibility to local minima in BP neural networks, model independ-ent meta learning algorithms are introduced to address the limitations of BP neural networks themselves. The focus of this article is on the SMAML algorithm in optimized meta learning, which supplements the derivation details missing in the doctoral thesis of MAML by Chelsea B. Finn from the University of California. Afterwards, the SMAML-BPNN algorithm was proposed, with BP neural network as the base learner, SMAML algorithm as the meta learner, and EXCEL table data as input data. The experimental classification results were specifically presented from various dimensions, with an accuracy improvement of up to 5 percentage points. In the experimental part, the model was applied to spatial prediction, predicting the distribution of the average soil carbon dioxide intensity every ten years and presenting it in the form of a grid. Finally, further reflection was given to the previous paper. CBAM attention mechanism was added to the extension prototype network algorithm, and detailed experimental results were presented through comparative experiments. The results showed that the prediction results were indeed optimized before and after the addition of attention mechanism. However, the accuracy of the experi-mental results was not as high as expected, and further improvement and reflection are needed on the research of algorithm parameter update methods and dataset partitioning in the future.


Biographical Notes:
Professor Dr. Wenfeng Wang is currently the editor in chief of International Journal of Electrical and Electronics Engineering (IJEEE) and International journal of Applied Nonlinear Science (IJANS). He is also a professor in Shanghai Institute of Technology. He is the director of International Academy of Visual Art and Engineering in London and the JWE Technological Research Center in Shanghai. He is also a tenured professor in IMT Institute in India and the director of Sino-Indian Joint research center of artificial intelligence and robotics. He was selected in 2018 as a key tallent in Ningbo Institute of Materials Technology and Engineering, Chinese Academy of Sciences. He is a reviewer of many SCI journals, including some top ones - Water Research, Science China-Information Sciences, Science of the Total Environment, Environmental Pollution, IEEE Transactions on Automation Science and Engineering and etc. He served as a keynote speaker of AMICR2019, IACICE2020, OAES2020, 3DIT-MSP&DL2020, NAMSP2021, ICCAES 2021, CSAMCS 2021 and etc.