Dissertation Methodology Example: Convoluted Neural Networks

Date:  2021-06-21 12:09:10
5 pages  (1179 words)
Back to categories
logo_disclaimer
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
University/College: 
Carnegie Mellon University
Type of paper: 
Dissertation methodology
logo_disclaimer
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

CHAPTER 3: METHODOLOGY 4
1.Aim of the Study 4
1.1.Purpose statement 4
1.2.Objectives statement 5
1.3.Questions 5
2.Research Approach 6
2.1.True experimental research 6
2.2.Justification for True Experimental Research 6
2.3.Intended outcomes 7
3.Population and Sampling 7
3.1.Population 7
3.1.1.Genetic Algorithms (GA) 7
3.1.2.Particle Swarm Optimization (PSO) 7
3.2.Sampling  7
4.Data Collection 8
4.1.Type of data to be collected 8
4.2.Data Collection Tools 8
5.Procedures 8
6.Data Analysis 8
Ethical Considerations 9
Trustworthiness 9
Limitations 9
References 10

CHAPTER 3: METHODOLOGY

Aim of the Study

Convoluted Neural Networks (CNNs) have gained popularity in the recent past as the simulation of human interpretation of photos. The two main fields where simulation is done is the object level- what the photograph aimed to capture- and the scene- the event or location where the photo was taken. Major breakthroughs have been made in the interpretation of image data using multi-layered CNN models that combine information from different sources to increase the reliability of event recognition frameworks. The combination of object and scene level CNN information aims to improve the overall accuracy of detection, increasing the capabilities of event recognition frameworks.

However, the existing literature has been discovered to have several shortcomings, one of which is that it doesnt cover the performances of different individual models with respect to one another. Information on the best possible combination of different CNN models for enhanced performance is also diminished by minimal attention given to this scope of deep architecture. With the insufficient coverage of combining different CNNs, the analysis of the impacts on the fusion scheme performances while shifting from one dataset to another is also an insightful area to be investigated. Understanding the impact of utilizing more than one datasets can open up neural networking to hitherto unexplored areas of discovery.

Purpose statement

The purpose of this research was to evaluate the effectiveness of genetic and particle swarm-based optimization of the fusion of CNN models for event recognition.

Objectives statement

The objectives of the study, which wound up also being the greatest contribution of the study to literature in the field, are stated below:

The report aims to indicate that the combination of Genetic Algorithm and Particle Swarm-based late fusion schemes which combine different deep models performs better than the separate use of models.

Creating a better combination of well-known architectures that are pre-trained on object and places datasets was another aim of the study. Higher performance and ground point of reference for future work was especially targeted.

To achieve the objectives stated above, the study also conducted experiments on three of the most challenging and well-known datasets to determine the optimal weighted combination of different CNN models that leads to a higher performance.

Questions

The research process was guided by a number of questions pertinent to the topic of Convoluted Neural Networks. They enabled the researchers to be confined to a set of topical areas that add to the knowledge availability in the field. The main questions covered by the study included:

What is Genetic Algorithm?

What is Particle Swarm Optimization?

How do the two models of deep architecture compare?

How does the object level of CNN interpretation of image data differ from the scene level?

How do different CNN models compare in creating higher performance in the interpretation of image datasets?

How can the models be combined to provide an optimal level of performance?

In seeking the answer to these questions, the objectives of the study, which includes gaining a deeper understanding of convoluted neural networks, were achieved. Another achievement gained from the pursuance of the objectives stated above was that the researchers were able to determine optimal combination of different CNN models, an industrially applicable accomplishment.

As two distinct methods of CNN, Genetic Algorithm (GA) and Particle Swarm Optimization are addressed in different subtopics in every section. This allows the researchers to focus on every critical area of the two before coming up with a strategy for their fusion.

Research ApproachBefore stating the method that was employed in conducting the research, it is imperative that a background to the research topic (CNN) is given. CNN

True experimental research

True experimental research pertains to the forms of experimentation where the subjects of research are treated in a varying manner, with one set of variables being controlled and the other or others being observed to make conclusions on their relationship [1].

Justification for True Experimental Research Applications of and brief definition for discipline origin

Intended outcomes

The approach will necessitate the formulation of relational questions in the practical research activities, produce data on relationships, and require relational analytics of data collected.

Population and Sampling

Population

Genetic Algorithms (GA)

General information and definition and statement of variables consistent with the purpose statement and the research questions

Particle Swarm Optimization (PSO)

General information and definition and statement of variables consistent with the purpose statement and the research questions

Sampling

CNN datasets and models defined

Purposively select Alexnet, GoogleNet, VGGNet16, and ResNetExplanation of each subject:

AlexnetGoogleNet,

VGGNet16,

ResNetData Collection

Type of data to be collected

Data is obtained from the analysis of existing CNN models data for PSO and GA, especially the performance of the combination of various models and the effects of transferring weights from one dataset to another. <From my evaluation of the material provided, it appears like no new data is to be obtained on CNNs. If my assumption is false, it would be nice to avail the information on the collection of the data> The type or types of data to be collected from the experiments, with detailed analysis of the equations used to define optimization. The instrument to be used in the collection of the data will also be stated and justified. (

Data Collection Tools

Experimental observations

If established instruments will be utilized, this section will detail each data-collection instrument. The relevant information pertaining to each instrument will include the source or developers of the instrument and any other salient information.

Procedures

Based directly on the research questions, this is the how-to section of the study. It details how the data will be collected based on the questions of interest.

Data Analysis

The steps involved in conducting an analysis of qualitative data. It will describe how the data will be organized and transcribed, the coding procedures of the transcripts or text files, and the specific quantitative software used for the analysis.

Ethical Considerations

This section should describe how you will maintain ethics of the study. Preserve anonymity and keep the documents secure.

Trustworthiness

This section shall highlight the steps taken to ensure objectivity and impartiality in the collection, analysis, and presentation of research findings. It demonstrates the aspects of the studys validity and reliability, the accuracy of the findings, and interpretation of results.

Limitations

Any limitations, restrictions, or constraints that may affect the dissertation outcomes (for example, the use of secondary CNN data)

 

References

Arnott D. (n.d.) Experimental Research. Monash University

Kennedy J. & Eberhart R. (1995) Particle Swarm Optimization. IEEE, vol. 3, pp. 1942-1948, DOI: 0-7803-2768-3/95

Ouellette R., Browne M. & Hirasawa K. (2004) Genetic Algorithm Optimization of a Convolutional Neural Network for Autonomous Crack Detection. IEEE, vol. 2, no. 4, pp. 516-521

Krizhevsky A., Sutskever I. & Hinton G. E. (2012) ImageNet Classification with Deep Convolutional Neural Networks. Neural Information Processing Systems Conference.

Bai Q. (2010) Analysis of Particle Swarm Optimization Algorithm. Computer and Information Science, vol. 3, no. 1, pp. 180-184,

 

logo_essaylogo_essay

Request Removal

If you are the original author of this essay and no longer wish to have it published on the ProEssays website, please click below to request its removal: