&lt;title&gt;Hybrid neural network and rule-based pattern recognition system capable of self-modification&lt;/title&gt;

Glover, C.W.; Silliman, Mike; Walker, M.; Spelt, P.F.; Rao, Nageswara S. V.

doi:10.1117/12.21079

Cited by 12 publications

(5 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, CNNs are generally used to process images, while RNNs are used in natural language processing and weather forecasts. In addition, multiple ANNs have incorporated various types of layers and modules in their structures, such as the generative adversarial network (GAN) (Zhang, Xu, et al 2017), as well as hybrid neural networks (HNN) (Glover et al 1990) with both convolutional and recurrent layers, which arose from the need to process data with both spatial and temporal features.…”

Section: Introductionmentioning

confidence: 99%

Multi-Frequency Data Acquisition Model and Hybrid Neural Network for Precise Electromagnetic Wellbore Casing Inspection

Ooi

Khater

Ozakin

et al. 2022

Day 2 Tue, November 01, 2022

View full text Add to dashboard Cite

Casing integrity inspection tools are indispensable in identifying defects that threaten the structural integrity of oil wells. In particular, electromagnetics-based (EM-based) inspection tools are commonly used for multi-casing corrosion imaging. These tools measure the scattered EM fields inside the inspected casings and generate estimations of metal loss properties. However, the interpretation of EM measurements is difficult due to their intrinsic nonlinearity with respect to defect characteristics. In this paper, a new machine learning-based inspection framework is developed to generate accurate cross-sectional images of casings to characterize metal loss location and shape. A hybrid neural network (HNN) consisting of a main structure that integrates both convolutional and recurrent layers, as well as a parallel cross-frequency module with convolutional filters predicts the cross-sectional images of the inspected casings. Metal losses on the inner surface of the inspected casing, as well as fully-penetrating losses, are detected using high-frequency signals. On the other hand, low-frequency signals enable the detection of metal losses on the outer surface, in addition to the two previous kinds of losses. The resulting inspection scheme requires only four receiver (RX) coils for each frequency of signals to accurately predict both the azimuthal location and size of defects.

show abstract

Section: Introductionmentioning

confidence: 99%

Multi-Frequency Data Acquisition Model and Hybrid Neural Network for Precise Electromagnetic Wellbore Casing Inspection

Ooi

Khater

Ozakin

et al. 2022

Day 2 Tue, November 01, 2022

View full text Add to dashboard Cite

show abstract

“…EVERAL researchers have investigated the design of hybrid systems that combine expert and connectionist subsystems [40], [47], [7], [12], [11], [20], [48]. The typical result of a transformational type of hybridization [49] is a Knowledge-Based Neural Network (KBNN) system with theory refinement capabilities, usually involving four phases: 1) the rule-base representation phase, where initial domain knowledge is extracted and represented in a symbolic format (e.g., a rule-based system); 2) the mapping phase, where initial domain knowledge is mapped into an initial connectionist architecture; 3) the learning phase, where this connectionist architecture is trained by a set of domain examples, and 4) the rule-extraction phase, where the trained and, thus, modified connectionist architecture is mapped back into an updated rule-based system.…”

Section: Introductionmentioning

confidence: 99%

Symbolic interpretation of artificial neural networks

Taha

Ghosh

1999

IEEE Trans. Knowl. Data Eng.

172

View full text Add to dashboard Cite

Hybrid Intelligent Systems that combine knowledge-based and artificial neural network systems typically have four phases involving domain knowledge representation, mapping of this knowledge into an initial connectionist architecture, network training, and rule extraction, respectively. The final phase is important because it can provide a trained connectionist architecture with explanation power and validate its output decisions. Moreover, it can be used to refine and maintain the initial knowledge acquired from domain experts. In this paper, we present three rule-extraction techniques. The first technique extracts a set of binary rules from any type of neural network. The other two techniques are specific to feedforward networks, with a single hidden layer of sigmoidal units. Technique 2 extracts partial rules that represent the most important embedded knowledge with an adjustable level of detail, while the third technique provides a more comprehensive and universal approach. A rule-evaluation technique, which orders extracted rules based on three performance measures, is then proposed. The three techniques area applied to the iris and breast cancer data sets. The extracted rules are evaluated qualitatively and quantitatively, and are compared with those obtained by other approaches.

show abstract

“…A variant of the N-learners problem has been first discussed in [25] in the context " of sensor fusion in a hybrid system. Potential applications of the N-learners problem include sensor fusion [11,16], hybrid systems [13,25], information pooling and group decision models [14,20], and majority systems [8].…”

Section: Introductionmentioning

confidence: 99%