| Full text | |
| Author(s): |
Zielinski, Kallil M. C.
;
Scabini, Leonardo
;
Ribas, Lucas C.
;
Bruno, Odemir M.
Total Authors: 4
|
| Document type: | Journal article |
| Source: | 2024 14TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION SYSTEMS, ICPRS; v. N/A, p. 7-pg., 2024-01-01. |
| Abstract | |
Network classification has become increasingly significant in understanding complex systems across various scientific fields. Life-like Network Automata (LLNA) has emerged as a powerful method for capturing the dynamic behavior of networks through Time-Evolution Patterns (TEPs). Despite LLNA's efficiency, the current method relies on a vast rule space, particularly with a Moore neighborhood of size 8, presenting a computational challenge and requiring a more efficient approach to rule selection without compromising classification accuracy. This paper aims to investigate the influence of varying neighborhood sizes on the performance of the LLNA-DTEP method and to assess the feasibility of reducing the computational load while maintaining high classification accuracy. An exhaustive search of all possible LLNA rules was conducted for different neighborhood ranges from 1 to 8 (Moore's neighborhood). For each rule, a feature vector was built based on histograms from the TEPs, which then was used in a Support Vector Machine (SVM) classifier to determine classification efficiency. The findings revealed that a reduced neighborhood range could significantly decrease the rule space and computational time. However, the impact on classification accuracy varied across four different datasets, with some showing robustness to changes in neighborhood size and others exhibiting notable sensitivity. The study shows that while reducing the neighborhood range in LLNA significantly reduces computational requirements, the choice of neighborhood size is a critical factor that must be tuned to each dataset's specific characteristics. (AU) | |
| FAPESP's process: | 23/04583-2 - Pattern recognition in images based on artificial neural networks and complex systems: from handcrafted descriptor extraction to automated learning |
| Grantee: | Lucas Correia Ribas |
| Support Opportunities: | Regular Research Grants |
| FAPESP's process: | 23/10442-2 - Deep learning for pattern recognition on multi-sensor and multidimensional data |
| Grantee: | Leonardo Felipe dos Santos Scabini |
| Support Opportunities: | Scholarships in Brazil - Post-Doctoral |
| FAPESP's process: | 21/08325-2 - An analysis of network automata as models for biological and natural processes |
| Grantee: | Odemir Martinez Bruno |
| Support Opportunities: | Regular Research Grants |
| FAPESP's process: | 24/00530-4 - Texture Feature Aggregation and Learning with Vision Transformers and its Applications on Biological and Medical Images |
| Grantee: | Leonardo Felipe dos Santos Scabini |
| Support Opportunities: | Scholarships abroad - Research Internship - Post-doctor |
| FAPESP's process: | 18/22214-6 - Towards a convergence of technologies: from sensing and biosensing to information visualization and machine learning for data analysis in clinical diagnosis |
| Grantee: | Osvaldo Novais de Oliveira Junior |
| Support Opportunities: | Research Projects - Thematic Grants |