Deep Forest Regression Based on Dynamic State Transition Optimization Algorithm

As a deep algorithm of non-neural network structure, deep forest regression (DFR) can be used to build soft measuring models of difficult-to-measure key parameters. However, as a kind of deep learning, the optimization of hyperparameters has become an inevitable problem in DFR. To solve above proble...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	Chinese Automation Congress (Online) S. 3786 - 3791
Hauptverfasser:	Xia, Heng, Tang, Jian, Qiao, Junfei
Format:	Tagungsbericht
Sprache:	Englisch
Veröffentlicht:	IEEE 06.11.2020
Schlagworte:	Data models deep forest regression dynamic state transition algorithm Forestry Heuristic algorithms hyper-parameters optimization Optimization Stochastic processes Support vector machines Training
ISSN:	2688-0938
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	As a deep algorithm of non-neural network structure, deep forest regression (DFR) can be used to build soft measuring models of difficult-to-measure key parameters. However, as a kind of deep learning, the optimization of hyperparameters has become an inevitable problem in DFR. To solve above problem, an improved dynamic state transition algorithm (DSTA) is used to optimize the hyper-parameters of the model. To achieved more accurate optimization process, the error change rate is used to fine-tuning the state factor during the iteration process, which is further improved with gradient-based refinement. Finally, simulation experiments are performed on the benchmark data set, and satisfactory simulation results show the effectiveness of the proposed approach.
ISSN:	2688-0938
DOI:	10.1109/CAC51589.2020.9327803