Stochastic approximation for learning rate optimization for generalized relevance learning vector quantization

Steeneck, Daniel W.; Bihl, Trevor J.

doi:10.1109/naecon.2017.8268804

Cited by 2 publications

(5 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…General Relevance Learning Vector Quantization Improved (GRLVQI) is an extension of Kohonen's Learning Vector Quantization (LVQ) [33] which is in the family of self-organizing Neural Network (NN) approaches using nearest Prototype Vector (PV) optimization. As a classifier, LVQ associates a PV with a given class (typically multiple PVs per class).…”

Section: Post-classification Grlvqimentioning

confidence: 99%

“…When an observation is input to the network, the PV closest to the observation "fires" and the prediction accuracy is based on whether or not the firing PV(s) are associated with the correct class for the observation. GRLVQI extends LVQ by incorporating cost functions, learning methods, and logic and operation improvements [33].…”

Section: Post-classification Grlvqimentioning

confidence: 99%

“…There are five fundamental hyperparameters for GRLVQI processing, including the gradient descent learning rate, relevance learning rate, conscience rate(s), and the number of class PVs. The correct model construction requires expertise or appropriate heuristics [33].…”

Section: Post-classification Grlvqimentioning

confidence: 99%

See 2 more Smart Citations

DNA Feature Selection for Discriminating WirelessHART IIoT Devices

Rondeau

Temple

2020

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

The proliferation of Wireless Highway Addressable Remote Transducer (WirelessHART) communications in support of Industrial Internet of Things (IIoT) applications is accompanied by increased vulnerability concerns that amplify the need for improved pre-attack security and post-attack forensic methods. This paper summarizes demonstration activity aimed at applying Time Domain Distinct Native Attribute (TD-DNA) fingerprinting and improving feature selection to increase computational efficiency and the potential for near-real time operational application. Assessments include both pre-classification and post-classification dimensional reduction using TD-DNA fingerprint features extracted from experimentally collected WirelessHART signals.Results show that pre-classification selection methods are superior, with average percent correct classification differential of 8% < %CD < 1% being maintained using selected feature subsets containing only 24 (10%) of the 243 full-dimensional features.

show abstract

Section: Post-classification Grlvqimentioning

confidence: 99%

Section: Post-classification Grlvqimentioning

confidence: 99%

See 1 more Smart Citation

DNA Feature Selection for Discriminating WirelessHART IIoT Devices

Rondeau

Temple

2020

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

show abstract

“…In doing so, this extends the initial work of [18] by allowing GRLVQI to find optimal settings in an iterative fashion for its four continuous hyperparameters: gradient descent learning rate ( ), relevance learning rate ( ), conscience rate 1 ( ), and conscience rate 2 ( ). Additionally, the work of [13] is extended by creating a more efficient approach to optimizing GRLVQI settings and by creating a more robust approach to GRLVQI optimization by averaging replication results.…”

Section: Introductionmentioning

confidence: 90%

“…[16] [17], and 2) stochastic estimation, e.g. [18]. Although various experimental design approaches have been used to find optimal hyperparameter settings, see [19], these do not avoid the limitations in specifying the high and low settings to examine.…”

Section: Introductionmentioning

confidence: 99%