Hierarchical Interpretable Construction Algorithm for Data Analysis

In low-computation modeling tasks, the Incremental Constructive Algorithm (ICA) shows good performance. However, it relying on horizontal network expansion for model building limits its mapping ability. Thus, it struggles to capture deep feature relations in complex data and can not meet accuracy ne...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	2025 30th International Conference on Automation and Computing (ICAC) s. 1 - 5
Hlavní autoři:	Dai, Wei, Duan, MingZi, Nan, Jing
Médium:	Konferenční příspěvek
Jazyk:	angličtina
Vydáno:	IEEE 27.08.2025
Témata:	Accuracy Computational efficiency Computational modeling Convergence Data analysis Geometry geometry control strategy Heuristic algorithms hierarchical deep structure Interpretable constructive algorithm Network architecture Prediction algorithms Stacking
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	In low-computation modeling tasks, the Incremental Constructive Algorithm (ICA) shows good performance. However, it relying on horizontal network expansion for model building limits its mapping ability. Thus, it struggles to capture deep feature relations in complex data and can not meet accuracy needs in complex predictions. This paper proposes the Hierarchical Incremental Construction Algorithm (HICA) by combining ICA interpretability control strategy and DNNs deep stacking structure. HICA has hierarchical propagation. It uses the previous layer residual as the next layer predicted value and the previous hidden layer output matrix as input. It can dynamically add hidden layers and iteratively replace actual outputs with expected ones for better generalization. Nodes are selected by an interpretability-based geometric control strategy. Experiments prove HICA accuracy advantage and lower computational cost.
DOI:	10.1109/ICAC65379.2025.11196560