A Review of Neural Networks Architectures, Designs, and Applications

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

A Review of Neural Networks Architectures, Designs, and Applications

***

Abstract: Artificial neural networks (ANNs) are cutting-edge computing techniques that have been widely applied to resolving a variety of challenging issues in the real world. The extraordinary data processing characteristics of ANNs, which aremostlyconnectedtohighparallelism,faultandnoisetolerance,learning,andvastnonlinearityskills,arewhatmakethem soalluring.ToserveasatoolkitandresourceforANNs modelers,thisworkoffersanoverviewofafewANNarchitecturesin the areas of recognition, prediction, and control. The review mechanism depends on comparing the most recent research in thesedisciplinesintermsofthefieldthatwasimplemented,thetoolsthatwereused,theresearchmethodology,andthekey goalsthatwereachieved.

Keywords Neural networks architectures, deep neural networks, graph neural networks and fully convolutional neural networks.

I. INTRODUCTION

Therearenumerousfactorsdrivingtherecentresurgenceofinterestinneuralnetworks,includingtheemergenceoftraining techniquesforadvancednetworktopologiesthatgobeyondtheconstraintsofearlyneuralnetworks[1].

Modelingbrainprocessesismadeeasierbyfastdigitalcomputers[2].Itisnowpossibletoproduce specialized hardwarefor neural networks thanks to technology [3]. While advancements have been made, new directions in neural network research haveemergedasaresultoflimitationsinstandardcomputerseriesthathavespedupneuralnetworkanalysis[4].

Artificial Neural Networks (ANNs) are recently developing machine modeling techniques that are widely used to solve challenging real-world issues across a variety of fields [5]. ANNs are defined as structures made up of interconnected, adaptable processing units (also known as artificial neurons or nodes) that have the ability to process and represent information in large-scale concurrent computations [6]. Even though ANNs are blatant abstractions of their biological counterparts, their purpose is to solve complicated issues by using the functionality data of physical networks rather than trying to mimic the structure of natural systems. The fascinating history of ANNs is what makes them so appealing [7]. Managingattributesincludenonlinearity,highparallelism,solidity,toleranceforflawsanderrors, intelligence,andtheability tomanagefluidandinaccurateinformation[8].

Both in its creation and in its use, neural network research is a highly interdisciplinary field [9]. A brief overview of a few neural networksnowinuse demonstratesthe breadth oftheir potential applications,fromsuccessful corporateapplications tosuccessfulsciencethatholdsgreatpromiseforthefuture[10]:

1. Generating signals: Eventually, noise and non-seismic signals contaminate registered seismic signals from various sources,suchasoceanwaves,wind,traffic,noise,electricnoise,etc.[11],[12].Inthegeneralareaofsignalprocessing for neural networks, there is a tone of software. The telephone line noise reduction was one of the first commercial applications[13].

2. Monitormanagement:Everyonewhohasattemptedorobservedthesemaneuversatfirstisawareofthedifficultiesof backing a trailer. A skilled driver, however, does the task with remarkable ease [14]. An effective vector control method called sliding mode control (SMC) aims to have the managed system's direction always point to a certain multiplicityandfinallyborderwithinaspecifictinyareasurroundingthespecificareamanifold[15],[16].

3. Understanding Pattern: The broad area of pattern recognition encompasses a number of key issues [17]. Numerous neural network applications have been developed for the automatic recognition of handwritten characters, whether theyarenumbersorletters[18],[19].Bothanessentialareaofmachinelearningresearchandacrucialaspectofour dailylivesisaudiopatternrecognition[20].

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

4. Therapeutic: The idea behind the programmer is to teach an artificial neural memory network to retain a lot of medical records, each of which contains details on a given case's symptoms, diagnosis, and course of treatment [21],[22]. One of the safest methods to lower breast cancer mortality was early detection and treatment. For the purpose of finding breast cancer, mammography and ultrasound are also employed. The bioelectric signal produced by the human heartbeat serves as a reflection of knowledge about many physiological characteristics. A biological eventinthehumanbodycausespulsemanifestation[23].

5. Speech development and voice recognition: One of the most distinctive features of humans is their ability to speak society and creation. Speech represents a great deal of individualized knowledge [24]. There were several technologies built utilizing the latest development technologies a speech signal's generated information [25]. The Learning to read English literature aloud is a difficult endeavor because how a letter is phonetically pronounced dependsonthecontextinwhichthemessageisused[26].Itisnormaltocreateatableofexceptionstoasetofrules dictatinghowspecificgroupsoflettersaretypicallypronouncedwhentacklingthis issue[27].Additionally,progress is being made in the challenging, speaker-neutral field of voice recognition. Many sound systems have limited vocabularyorgrammarorneedtoberetrainedfordifferentspeakers[28,29].

6. Business and Company: Neural networks are employed in a range of business settings [30]. Although the laws that form the fundamentals of mortgage subscription are easily comprehended, it is difficult to adequately describe the procedure by which experts make choices in marginal circumstances [31]. Artificial or computer intelligence can perceivesensoryinformationinavarietyofways,oneofwhichisthroughneuralnetworks.Thefirstneuralnetworks weremodeledafterneuronalbiologicalnetworks[32].

The primary difficulty with all of these activities is predicting the upcoming events. So-called case logs are where data is normally kept after prior processes have been completed [33]. These logs are a trustworthy source for predictive training models, which implies that historical occurrences are an important indicator of how a mechanism will develop in the future [31].Thereisa weightlayer ina neural network withonelayer.Separationisalsopossible betweeninputunitsthatreceive signalsfromtheoutsideenvironmentandoutput unitsfromwhichthenetresponsecanbegleaned[34].Aninterconnection betweentheinputandoutputunitsofoneormorenodelayers(orlevels)isknownasamultilayernetwork[8].Theweightfor two neighboring unit levels is typically present (input, hidden, or output) [35]. More complex issues can be handled by multilayer networks than by single-layer ones. A layer that competes has lots of neural networks in it. In architectural diagrams,suchnetworksaretypicallynotconnectedtoneuronsinthe competitivelayer[36].Inorderforsomeapplications to be successfully completed with decent results, certain ANNs techniques are essential. This study intends to give a preliminary grasp of thesetechniques. Therefore, thegoal istoproducea thoroughexplanation of the most well-knownand significant ANN strategies, together with information on the implementation domains, tools, and platforms that are appropriateforeachmethodology.Asaresult,theresearchersareassistinginthequickadvancementofthesekeyprocedures. The remainder of this work is structured as follows. In section II, a survey of neural networks. All referenced and reviewed studiesarecontrastedanddiscussedinsectionIII.TheconclusionofthisworkisstatedinsectionIV.

II. NEURAL NETWORKS SURVEY

Inthepastfewyears,alotofacademicshavediscussedneuralnetworksandtheframeworksusedtoconstructthem.Thekey findingsofvariousrecentstudiesarediscussedinthissection.

TDGebhardandcolleagues[37]outlined indetailthechallengesthatmachinelearningcanattempttoaddressinthecontext of looking for compact binary coalescences (CBC) gravity waves, and addressed their shortcomings when taking the place of matched or Bayesian parameter evaluation methods. Then use evolutionary neural networks (CNNs) to extend the present binary classification-based strategy to account for the varying input lengths. Additionally, upcoming challenges and subtle occurrences in the data generation process are emphasized, which could result in inaccurate comparisons. Finally, our architecture'sempiricalresultsdemonstratethatdeepneuralnetworksareapotentadditiontothecurrentpipelineforquick andeffectivetriggercreation.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

J.Zhou,H.Liu,andothers[38]proposedanovelmethodforfaultdetectionemployingrecurrentneuralnetworksintheform of an auto encoder (RNN). The Gated Recurrent Unit (GRU)-based denoising auto-encoder used in this method predicts numerous vibration values for rolling bearings in the future. These GRU-NP-DAEs, which are nonlinear auto encoders, are well-suited for generalizing each fault pattern. Then, reconstruction errors between the following cycle and output data produced by various GRU-NP-DAEs are used as input data to diagnose unstable conditions and identify fault type. The effectivenessanddominanceofthesuggesteddiagnosticapproachoveralternativestate-of-the-artproceduresareattestedby historicaldatasetsforspinningmachines.

N.Choma andothers[39]usedGNN networkstoimprovesignal identificationin the Icecube neutrinoobservation. TheIceCube detector array is represented as a table, with the sensors serving as the vertices and the edges serving as a learned attribute of the sensors' spatial coordinates. The author claims that our GNN is adaptive and that measurement is limited to theinput signal supportsinceonlya portionoftheIce-sensor Cubesareused duringa givenobservation. The authorshows how our GNN design is useful for describing ice cube events since it outperforms both the conventional physics-based approachandthetraditional3Dconvergenceofneuralnetworks.

V.-A. Le and others [40] a novel 3D overhead crane system with an adaptable sliding hierarchical mode control system has been developed.Theprocessofbuildinga controllerbeginswith thehierarchical configurationoftwoslidingsurfacesofthe firstorder,whicharerepresentedbytwoactuatedandunsaturatedbridgecranesubsystems.Asaresultofdisturbancesinthe 3D overhead crane dynamic model, unknown parameters are provided to characterize radial base function networks whose weights are retrieved from a Lyapunov function in this situation. As a result, the controller parameters are intelligently generated.Thesuggestedremedyrendersthecranesystemstableunderunforeseencircumstances,whenitischallengingto constructsuchambiguousandill-definedcharacteristics[55–56].

A new neural network identification and evaluation method developed by H. Niu et al. [41] is centered on an attack that detectsabnormaltrafficflowcausedbyaclassofattacksoncontactlinksinanetworkedcontrolsystemfeedbackloop(NCS). The network attack identification residual is created by modeling the present network flow in the bottleneck node as a nonlinearfunctionandemployinganN.N.Observer.Theresidualisthenusedtodeterminewhetheritexceedsasetthreshold when the contact network assault occurs begins after the finding, another N.N. The attack's flow injection is employed to approximation. The author develops an attack detection strategy for the physical system using optimal event-led adaptive dynamicprogramming.Thenetworkprocessorpausesandreceivesinsufficientpackets.

Thesuggestedapproachwilldetectandassessnetworkthreatsaswellasthephysicaldevice'ssensors.

The Incident Cause (EABSET) exponential mitigation approach was created by Y. Fan et al. [42] to achieve the worldwide stabilization of delayed nerve networks (MNNs). The issue is raised for two reasons: first, the methods for maintaining the weights of state-dependent links may be difficult to use, and second, the current event trigger’s procedures may be conservativeindecreasingtriggertimes.The stabilizationconundrumisinitiallydevelopedinanetworkedcontrolsystemto addresstheseproblems.Then,anexponentialattenuationtermisneededforthegiventhresholdfunction.Itwilldecreasethe frequencyofdata packet transmission andlengthenthe timebetween twoconsecutivelytriggered events.Some appropriate requirements are obtained utilizing the input delay approach, temporal and component-based Lyapunov functionality, and matrixnorminequalities.

S.Xieetal.[43]investigatedalargerrangeofcommunicationpatternsusingneuralnetworklensesthatwererandomlylinked. Theconceptofastochasticnetworkgenerator,whichencompassestheentireprocessofnetworkdevelopment,isintroduced by the author first. Encapsulation provides a unified perspective on the random wiring networks and the search for neural architecture(NAS).Theauthorthenusesthreetraditionalrandomgraphmodelstogeneraterandomlywirednetworkgraphs. The results are unexpected: several iterations of these random generators produce network instances with computable comparedtotheImageNetbenchmark,precision.

D. K. Jain and others, [44] suggested a system for categorizing each image into one of six categories of facial expression. The Deep Neural Networks (DNNs) model is made up of deep residual blocks and single layers of convolution. In the suggested model,animagemarkwasfirstmadeoneachfaceasaprelude.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

Second, the suggested DNN model incorporates the images. Dataset Expanded Cohn- Kanade (C.K.+) and Female Japanese Facial Expression were used to train this model (JAFFE). Overall findings indicate that the present DNN model will perform betterthanthenewemotiondetectionmethods.Eventheproposedmodelimprovesonourpreviousmodel'sprecision.

R.Ptuchaandothers[45]presentedawhollyconvolutionalnetworkarchitecturethatgeneratestextsymbolsofanylength.A canonical display of the input blocks is a preprocessing phase norm, negating the need for an expensive, repeating symbol rectification. The author frequently includes a probabilistic error rate to rectify wrong blocks of terms if a lexicon is known. Our multi-State convolution method is the first method to demonstrate cutting-edge results on both lexicon-based and subjectivehandwritingrecognitionbenchmarks.

M.S.Ayhanetaldescription.'sofanintuitivetechniqueformeasuringthestate-of-the-artdiagnosticinstabilityinDeepNeural Networks (DNNs) for the diagnosis of diabetes [46] was centred on test time data growth. The author claims that the computationofconfusionderivedisfairandthat even experienced doctors frequentlystruggletoidentifycasesofuncertain diagnosis.ThisopensthedoorforanadaptiveambiguitytreatmentforDNN-baseddiagnosticsystems.

K. Akyol employed the clinical dataset provided by the University of Bonn to evaluate the suggested model, which was providedwithanensemblestackingtechnique.Theproposedseizuredetectionmodel'sviabilitywasdemonstratedusingboth the proposed model and a deep neural network model. The proposed model is competitive with the Deep Neural Networks (DNNs)basemodel,accordingtoexperiments.ANonlinearPredictiveModelTechnique(NMPC)forAtmospheric

Pressure Plasma Jets (APPJs) was presented by A. D. Bonzanini et al. [48] with examples of implementations in plasma medicine. The goal of NMPC is to maintain patient comfort and safety standards while controlling the cumulative thermal impactsofplasmainasubstratum.Withasimple,explicitcontrollaw,deepneuralnetworksareemployedtoapproximatethe NMPC implicit law. By projecting the neural network output onto a set that ensures that the state stays within a properly definedinvariantgroup,themaximumpossibleconstraintfulfillmentismadepossible.

Closed-loopsimulationsandin-the-momentcontroltestsshowhownonlinearcontrolcostscanbesuccessfullymanagedina fastsamplingcyclewhilemaintainingtheindicatedestimates.

BaturDinleretal.[49]attemptedtoformasizableKurdishvocabularydatasetanddeterminethemodel'sidealparametersfor the identification of speech segments based on consonant, vowel, and silence (C/V/S) discrimination. To achieve this, the phonemeborderswereportrayedusingthreehybridfunctionvectortechniques,threewindowtypes,andfourwindow sizes. ArecurrentGatedRecurrentUnit(GRU)networkwithsixdifferentC/V/Sdiscriminatoryclassificationalgorithmswasusedto determine phoneme boundaries. In Kurdish acoustic signals, the author demonstrated that the GRU model had high speech segmentation performance. By utilizing hybrid characteristics, window widths, window forms, and classification models for Kurdishspeakersvirtually,theexperimentalresultsofthisstudyshowtheimportanceofsegment-detection.

R.J.Wesleyandothers[50] after beingcombinedina processknownas"reconstructedspace,"segmentedspeechphonemes are examined using cutting-edge filters (RPS). Since they were created from start with embedded voice data, these characteristics for extracting Convolutional filters are ideal for various data networks. A geometric explanation of the dynamicsoftheobservable structurecanbefoundinthereconstructionofphasespace.thenprovidea studydemonstrating the use of a convolutional neural network to distinguish between attributes arising from the texture and shape of this geometric representation (CNN). CNNs are widely employed in picture tasks, although they were not used in step space portraiture,likelyduetotheintegration'shigherdimensionality.

Anjos and others [51] designed a serious game controlled in real-time by children's voices to aid kids in controlling the development of European Portuguese sibilant sounds (E.P.). By utilizing a sibilant classifier for a consonant Kid can practice creating these sounds more frequently because the game doesn't require adult supervision, which may speed up voice improvement. The author suggests that E.P. sibilant phonemes should be recognized by deep convolutional neural networks andincorporatedintoourchallengingspeechandlanguagetherapygames.UtilizingtheMelfrequencycepstralcoefficientsor Mellogfilterbankstocomparetheeffectivenessofvariousartificialneuralnetworks.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

Byshowinghowthemanybehaviorsincludedinaphaseinfluencedtheforecast,M.Harletal.[53]contributedamethodology thatfurtherexemplifiesaforecast.ThisthesisisthefirsttoemployGGNNPBPMandthefirsttodescribedecisionsusingGNN networks.Itusesadatasetofprocesseventstodemonstrateourapproach.

III. COMPARISON AND DISCUSSION

It is clear from the previous section that researchers have used a variety of approaches and strategies across a range of disciplines. Researchers have underlined key elements that are relevant to comparing their proposals. A comparison of the queriesdescribedinSectionIIisshowninTable1.Inordertovalidatetheobjectivesoutlinedintheirtechniqueinthefieldof theneuralnetwork,thecomparisoncontainsfourcrucialqualitiesthatmatchtheirpatterns.Thecomparisonofimplemented field, tools used, research methodology, and significant goals achieved the chart makes it clear that the sources [37], [39] directlydependon thesignal processing field.References[40,42]jobascontrollers,butinsteadofstaying researchers,they worked in the fields of speech production, speech recognition, business, and pattern recognition. The researcher employs a crucial technique depending onthefield ofstudy Neural Networks,DeepNeural Networks,and Deep Convolutional Neural Networks.Recurrentneuralnetworks,convolutionalneuralnetworks,gatedgraphneuralnetworks,andothertechniquesare alsoemployed.Bothresearchershavestrongstructures,frames,andfunctionsasaresultofapplyingthesemethodologiesand strategies.However,thetrendamongresearchershasbeentowardcontemporarycommercialandmedicaldomains.

IV. CONCLUSION

Inthisresearch,activemethodsforconstructedneuralnetworkarchitectureswerediscussed.Variousactivemechanismsplay a crucial part in the sectors where neural networks are implemented, one can infer from the examined studies. These industriesincludespeechproductionandrecognition,patternrecognition,medical,business,andsignalprocessing.Powerful technologies are also employed in this field, including Sibilant Consonant Classifier, Control Problem Formulation, Phoneme SetanditsProperties,andBayesianDeepNeuralNetworks.ThemethodsusedareDeepNeuralNetworks,DeepConvolutional NeuralNetworks,FullyConvolutionalNeuralNetworks,RecurrentNeuralNetworks,GraphNeuralNetworks,andGatedGraph NeuralNetworks.Today'sacademicsarefocusingmoreonbusiness,speechrecognition,andmedicalapplications.Framesfor effective neural networks consequently, several systems have been created, including those that make prediction more understandable, speech segment detection for the Kurdish language, a framework based on an increase in test-time data, a noveltechniqueofoutputprediction,andamulti-stagein-depthlearningapproach.Additionally,itmaybeconcludedthatthe best method for signal processing is Convolutional Neural Networks. Additionally, both fully convolutional neural networks and deep neural networks are appropriate for use in both commercial and medical applications. While Deep Convolutional NeuralNetworksareeffectiveforimplementationsofSpeechProductionandRecognition.

V. REFERENCES

1. R. W. P. D. S. Levine and D. L. Long, Fundamentals of neural network modeling: Neuropsychology and cognitive neuroscience:MitPress,1998

2. L. da Fontoura Costa and G. Travieso, "Fundamentals of neural networks: By Laurene Fausett. Prentice-Hall, 1994, pp. 461,ISBN0-13-334186-0,"ed:Elsevier,1996.

3. I. A. Basheer and M. Hajmeer, "Artificial neural networks: fundamentals, computing, design, and application," Journal of microbiologicalmethods,vol.43,pp.3-31,2000.

4. A. S. Eesa, Z. Orman, and A. M. A. Brifcani, "A novel featureselection approach based on the cuttlefish optimization algorithmforintrusiondetectionsystems,"Expertsystemswithapplications,vol.42,pp.2670-2679,2015.

5. Y. Chen, L. Shen, R. Li, X. Xu, H. Hong, H. Lin, et al., "Quantification of interfacial energies associated with membrane fouling in a membrane bioreactor by using BP and GRNN artificial neural networks," Journal of Colloid and Interface Science,vol.565,pp.1-10,2020.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

6. G. Lan, Y. Wang, G. Zeng, and J. Zhang, "Compressive strength of earth block masonry: Estimation based on neural networksandadaptivenetwork-basedfuzzyinferencesystem,"CompositeStructures,vol.235,p.111731,2020.

7. M. R. Mahmood and A. M. Abdulazeez, "Different Model for Hand Gesture Recognition with a Novel Line Feature Extraction,"in2019InternationalConferenceonAdvancedScienceandEngineering(ICOASE),2019,pp.52-57.

8. F.H.Guenther,"Neuralnetworks:Biologicalmodelsandapplications,"Oxford:InternationalEncyclopediaoftheSocial& BehavioralSciences,pp.10534-7,2001.

9. J. A. Ahmed and A. M. A. Brifcani, "A new internal architecture based on feature selection for holonic manufacturing system,"InternationalJournalofMechanical,Aerospace,Industrial,MechatronicandManufacturingEngineering,vol.2,p. 1431,2015.

10. S.RusselandP.Norvig,Artificialintelligence:amodernapproach:PearsonEducationLimited,2013.

11. W.Zhu,S.M.Mousavi,andG.C.Beroza,"Seismicsignaldenoisinganddecompositionusingdeepneuralnetworks,"IEEE TransactionsonGeoscienceandRemoteSensing,vol.57,pp.9476-9488,2019.

12. A. K. Mukhopadhyay and S. Samui, "An experimental study on upper limb position invariant EMG signal classification basedondeepneuralnetwork,"BiomedicalSignalProcessingandControl,vol.55,p.101669,2020.

13. N. K. N. Aznan, A. Atapour-Abarghouei, S. Bonner, J. D. Connolly, N. Al Moubayed, and T. P. Breckon, "Simulating brain signals: Creating synthetic eeg data via neural-based generative models for improved ssvep classification," in 2019 InternationalJointConferenceonNeuralNetworks(IJCNN),2019,pp.1-8.

14. Y. Zhou and Z. Zeng, "Event-triggered impulsive control on quasisynchronization of memristive neural networks with time-varyingdelays,"NeuralNetworks,vol.110,pp.55-65,2019.

15. S.Wang,Y.Cao,T.Huang,Y.Chen,P.Li,andS.Wen,"Slidingmodecontrolofneuralnetworksviacontinuousorperiodic samplingevent-triggeringalgorithm,"NeuralNetworks,vol.121,pp.140-147,2020.

16. J. Ni and P. Shi, "Global predefined time and accuracy adaptive neural network control for uncertain strict-feedback systemswithoutputconstraintanddeadzone,"IEEETransactionsonSystems,Man,andCybernetics:Systems,2020.

17. A. Raphael, Z. Dubinsky, D. Iluz, and N. S. Netanyahu, "Neural Network Recognition of Marine Benthos and Corals," Diversity,vol.12,p.29,2020.

18. S.Zhang,W.Huang,andC.Zhang,"Three-channelconvolutionalneuralnetworksforvegetableleafdiseaserecognition," CognitiveSystemsResearch,vol.53,pp.31-41,2019.

19. D.A.Zebari, D. Q.Zeebaree, A.M.Abdulazeez,H.Haron, andH.N.A. Hamed,"Improved ThresholdBasedand Trainable Fully Automated Segmentation for Breast Cancer Boundary and Pectoral Muscle in Mammogram Images," IEEE Access, 2020.

20. Q.Kong,Y.Cao,T.Iqbal,Y.Wang,W.Wang,andM.D.Plumbley,"Panns:Large-scalepretrainedaudioneuralnetworksfor audiopatternrecognition,"IEEE/ACMTransactionsonAudio,Speech,andLanguageProcessing,vol.28,pp.2880-2894, 2020.

21. A.J.Weisman,M.W.Kieler,S.B.Perlman,M.Hutchings,R.Jeraj,L.Kostakoglu,etal.,"ConvolutionalNeuralNetworksfor Automated PET/CT Detection of Diseased Lymph Node Burden in Patients with Lymphoma," Radiology: Artificial Intelligence,vol.2,p.e200016,2020.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

22. D. Q. Zeebaree, H. Haron, A. M. Abdulazeez, and D. A. Zebari, "Machine learning and Region Growing for Breast Cancer Segmentation,"in2019InternationalConferenceonAdvancedScienceandEngineering(ICOASE),2019,pp.88-93.

23. Z. Chen, A. Huang, and X. Qiang, "Improved neural networks based on genetic algorithm for pulse recognition," ComputationalBiologyandChemistry,vol.88,p.107315,2020.

24. W. Zhang, Y. Liu, X. Wang, and X. Tian, "The dynamic and taskdependent representational transformation between the motorandsensorysystemsduringspeechproduction,"CognitiveNeuroscience,pp.1-11,2020.

25. H. Abbasi, A. J. Gunn, L. Bennet, and C. P. Unsworth, "Deep convolutional neural network and reverse biorthogonal wavelet scalograms for automatic identification of high frequency microscale spike transients in the post-hypoxicischemic EEG," in 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC),2020,pp.1015-1018.

26. P. Nantasri, E. Phaisangittisagul, J. Karnjana, S. Boonkla, S. Keerativittayanun, A. Rugchatjaroen, et al., "A Light-Weight ArtificialNeuralNetworkforSpeechEmotionRecognitionusingAverageValuesofMFCCsandTheirDerivatives,"in2020 17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology(ECTI-CON),2020,pp.41-44.

27. S. Parthasarathy and C. Busso, "Semi-supervised speech emotion recognition with ladder networks," IEEE/ACM transactionsonaudio,speech,andlanguageprocessing,vol.28,pp.2697-2709,2020.

28. M. Sajjad and S. Kwon, "Clustering-Based Speech Emotion Recognition by Incorporating Learned Features and Deep BiLSTM,"IEEEAccess,vol.8,pp.79861-79875,2020.

29. T.-S.Nguyen,S.Stüker,J.Niehues,andA.Waibel,"Improvingsequence-to-sequencespeechrecognitiontrainingwithonthe-fly data augmentation," in ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP),2020,pp.7689-7693.

30. V. Vashisht, S. Kamya, and M. Vashisht, "Defect Prediction Framework using Neural Networks for Business Intelligence Technology Based Projects," in 2020 International Conference on Computer Science, Engineering and Applications (ICCSEA),2020,pp.1-5.

31. P. Philipp, R. Jacob, S. Robert, and J. Beyerer, "Predictive Analysis of Business Processes Using Neural Networks with Attention Mechanism," in 2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC),2020,pp.225-230.

32. S. Leoshchenko, A. Oliinyk, S. Subbotin, and T. Zaiko, "Using Recurrent Neural Networks for Data-Centric Business," in DataCentricBusinessandApplications,ed:Springer,2020,pp.73-91.

33. B. Singh, N. Kumar, and B. K. Panigrahi, "Steepest Descent Laplacian Regression Based Neural Network Approach for OptimalOperationofGridSupportiveSolarPVGeneration,"IEEETransactionsonCircuitsandSystemsII:ExpressBriefs, 2020.

34. Z.Guo,Y. Shen,A. K.Bashir, M.Imran, N.Kumar,D.Zhang, etal.,"Robustspammer detectionusing collaborativeneural networkininternetofthingapplications,"IEEEInternetofThingsJournal,2020.

35. S. S. Basha, S. R. Dubey, V. Pulabaigari, and S. Mukherjee, "Impact of fully connected layers on performance of convolutionalneuralnetworksforimageclassification,"Neurocomputing,vol.378,pp.112-119,2020.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

36. L. Nanni, A. Lumini, S. Ghidoni, and G. Maguolo, "Stochastic Selection of Activation Layers for Convolutional Neural Networks,"Sensors,vol.20,p.1626,2020.

37. T. D. Gebhard, N. Kilbertus, I. Harry, and B. Schölkopf, "Convolutional neural networks: A magic bullet for gravitationalwavedetection?,"PhysicalReviewD,vol.100,p.063015,2019.

38. H.Liu,J.Zhou,Y.Zheng,W.Jiang,andY.Zhang,"Faultdiagnosisofrollingbearingswithrecurrentneuralnetwork-based autoencoders,"ISAtransactions,vol.77,pp.167-178,2018.

39. N. Choma, F. Monti, L. Gerhardt, T. Palczewski, Z. Ronaghi, P. Prabhat, et al., "Graph neural networks for icecube signal classification," in 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), 2018, pp. 386-391.

40. V.-A.Le,H.-X.Le,L.Nguyen,andM.-X.Phan,"Anefficientadaptivehierarchicalslidingmodecontrolstrategyusingneural networksfor3Doverheadcranes,"InternationalJournalofAutomationandComputing,vol.16,pp.614-627,2019.

41. H. Niu, C. Bhowmick, and S. Jagannathan, "Attack detection and approximation in nonlinear networked control systems usingneuralnetworks,"IEEEtransactionsonneuralnetworksandlearningsystems,vol.31,pp.235-245,2019.

42. Y. Fan, X. Huang, H. Shen, and J. Cao, "Switching event-triggered control for global stabilization of delayed memristive neuralnetworks:anexponentialattenuationscheme,"NeuralNetworks,vol.117,pp.216-224,2019.

43. S. Xie, A. Kirillov, R. Girshick, and K. He, "Exploring randomly wired neural networks for image recognition," in ProceedingsoftheIEEEInternationalConferenceonComputerVision,2019,pp.1284-1293.

44. D. K. Jain, P. Shamsolmoali, and P. Sehdev, "Extended deep neural network for facial emotion recognition," Pattern RecognitionLetters,vol.120,pp.69-74,2019.

45. R. Ptucha, F. P. Such, S. Pillai, F. Brockler, V. Singh, and P. Hutkowski, "Intelligent character recognition using fully convolutionalneuralnetworks,"Patternrecognition,vol.88,pp.604-613,2019.

46. M.S.Ayhan,L.Kuehlewein,G.Aliyeva,W.Inhoffen,F.Ziemssen,andP.Berens,"Expert-validatedestimationofdiagnostic uncertaintyfordeepneuralnetworksindiabeticretinopathydetection,"MedicalImageAnalysis,p.101724,2020.

47. K. Akyol, "Stacking ensemble based deep neural networks modeling for effective epileptic seizure detection," Expert SystemswithApplications,vol.148,p.113239,2020.

48. A.D.Bonzanini,J.A.Paulson,D.B.Graves,andA.Mesbah,"TowardSafeDoseDeliveryinPlasmaMedicineusingProjected NeuralNetwork-basedFastApproximateNMPC⋆,"inProceedingsofIFACWorldCongress,2020.

49. Ö. Batur Dinler and N. Aydin, "An Optimal Feature Parameter Set Based on Gated Recurrent Unit Recurrent Neural NetworksforSpeechSegmentDetection,"AppliedSciences,vol.10,p.1273,2020.

50. R.J.Wesley,A.N.Khan,andA.Shahina,"PhonemeClassificationinReconstructedPhaseSpacewithConvolutionalNeural Networks,"PatternRecognitionLetters,2020.

51. I. Anjos, N. Cavalheiro Marques, M. Grilo, I. Guimarães, J. Magalhães, and S. Cavaco, "Sibilant consonants classification comparisonwithmulti‐andsingle‐classneuralnetworks,"ExpertSystems,p.e12620,2020.

52. N. Mehdiyev, J. Evermann, and P. Fettke, "A novel business process prediction model using a deep learning method," Business&informationsystemsengineering,vol.62,pp.143-157,2020.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN:2395-0072

53. M.Harl,S.Weinzierl,M.Stierle,andM.Matzner,"Explainablepredictivebusinessprocessmonitoringusinggatedgraph neuralnetworks,"JournalofDecisionSystems,pp.1-16,2020.

54. G. Park and M. Song, "Predicting performances in business processes using deep neural networks," Decision Support Systems,vol.129,p.113191,2020.

55. S. B. Sadkhan, A. Q. Hameed, H. A. Hamed, " A proposed identification method for multi-user chirp spread spectrum signals based on adaptive Neural-Fuzzy Inference System (ANFIS)", 2016 Al-Sadeq International Conference on MultidisciplinaryinITandCommunicationScienceandApplications(AIC-MITCSA.

56. S. B. Sadkhan., A. Q. Hameed, H. A. Hamed," Digitally modulated signals recognition based on adaptive neural-fuzzy inferencesystem(ANFIS)",InternationalJournalofAdvancementsinComputingTechnology(IJACT)