Project Plan

Project Proposal

 

Title Deep Learning: Mining of Visual Content
Weekly Report https://thinkspace.csu.edu.au/11709982asadalikhan/weekly-progress-reports/.
Project Area Data Mining
Type of Project Research
Technology Deep Learning
Technique Mining
Domain Visual Content
Keywords Visual Data Mining, Static Structural Mining, Convolutional Neural Network, Dynamic Structural Mining, Deep Fakes.
Rationale 1.     Problem Domain Deep learning is a subset of machine learning and is integral for artificial intelligence. Autonomous cars, robotics, bioinformatics, medical diagnostics, Entertainment Industry, Deep fakes etc.

As per Memon et al. (2010) a network can have many nodes and links which can increase the complexity. This makes it difficult to understand and therefore it is desired to decrease this complexity by forming subgroups. Many applications can be on advantage because of this technology (Chap 1).

According to Guo et al. (2016) There are different type of neural networks available and the most noted approach in deep learning is Convolutional Neural Networks (CNNs) (pp 27-48). With different neural networks available, thus it is wise to study and research for the concluding if the most noted CNNs are more productive than other regular neural networks.

Convolutional networks and regular networks are not very useful in dynamic content mining because they can not deal with memory and time.

Deep fakes are new technology which is mostly being used for negative reasons e.g., political rumors, pornography, evidence forgery. There are some positive usages as well like it help in filming when the actors are not available or detection of frauds but still negatives weigh more than positives which makes it essential to study.
  2.     Purpose and Justification Mining of Visual Content is integral for artificial intelligence and technologies like robotics and Bioinformatics etc. The reason why I am doing research on this particular topic, is because of its broad application in present and future technologies.
  3.     Background Information As per Zemmari & Benois-Pineau (2020) In computer vision, Visual content mining has been a crucial aspect. Objects in videos and images are labelled and find correctly using Visual mining. It is involved in retrieval and indexing of multimedia, natural language processing, Bioinformatics and diagnosis, robotics, restoration of images and videos etc. (pp 1-3). Visual content mining is based on two subcategories Static Structural Mining and Dynamic Structural Mining. Neural Networks in static mining have few drawbacks such as having no concept of sequence in memory nor in time, which is why these are not useful in Dynamic content mining. To solve this problem there are networks such as Recurrent neural network (RNN), Hidden Markov models and long-short term memory networks (LSTM). Selection of a model is always extremely important in machine learning for the outcomes. There are many different neural networks such as Modular Neural Network, a research will be conducted if Convolutional neural networks are better or not in comparison to others. According to Schirrmeister et al. (2017) Convolutional networks have Pros and Cons is comparison to different models of machine learning. Benefits include their suitability for end-to-end learning and the scalability that they can have in bigger datasets. There are cons as well and the most important is providing false result with conviction (pp 5391-5420). There are currently available some studies available but the comparison between these networks still need elaboration, thus researching on it will be very beneficial for Data Science professionals, students, and researchers.

Deep fakes are growing concern in the society and as according to Paterson and Hanley (2020) the political warfare on digital networks and social media has risen and technologies like ‘deep fakes’ are being used to spread rumors and fake news (pp 439-454). Maddocks states (2020) most of deep fakes that are available on the internet at different platforms are pornographic and still the major attention is only on deep fakes related to politics (415-423). Solution to counter deep fakes is using machine learning techniques and neural networks to identify fakes.
Research Questions 1.     How can the complexity of neural network can be reduced in Static Structural Mining?

2.     Are Convolutional Neural Networks (CNNs) being better than regular neural Networks?

3.     What approaches can be used for mining of dynamic content as convolutional neural networks are not suitable?

4.     How the deep fakes represent great danger to the society?
Brief Description of Proposal The proposal is about Deep learning: Mining of Visual Content. There are static structural and dynamic structural mining methods which use different neural networks such as CNNs, RNNs, and LSTM etc. There can be complexities in these networks and one can be efficient in comparison to other. Additionally, where CNNs are not suitable, there are other approaches like RNNs available. This technology and techniques have many applications in modern era but along with innovations and benefits, it also brought some negativities and deep fake is the most problematic of all.

This research is going to be beneficial in understanding different aspect of Mining of Visual content through deep learning.
Theoretical Framework This research is based on theoretical framework as I will be discussing different concepts and theories to provide grounds for my research.
Methodology 1.     Analysis of source of Information Source of information is very important for any research project as it can impact the end results of the research as well as the credibility. My source of information will be books, articles, conference papers, etc. from credible publishers such Springers and IEEE etc. I will use APA referencing tool for referencing and crediting the work of others. End Note is a great tool for managing the references and formatting citations which keeps the researcher focused on the research more.
  2.     Research Method(s) ·       Qualitative Research Method.

·       Conclusive Research Method.

As there are questions in this research project which need to be answered at the end of the research and a conclusion is to be provided, thus I have opted for conclusive research method.

The reason for qualitative is as we are not dealing with the number and when we are not dealing with quantitative data, then qualitative research in the best option.
  3.     Data Collection or System design methods This research is based on literature, therefore methods used for literature review will be opted for the searching out the written material on the selected topic. It will assist in finding out any trends revealed by this research area already. It can identify where more investigations are required.

This literature review will be carried out by extant literature search, quality of available literature will be assessed, and screening will be done for inclusion. Extraction and analysis will be performed.
  4.     Ethical Issues Code of Ethics by Australian Computer Society cover all Ethical issues related to an IT professional.

Ethical research conduct makes it necessary for the researcher to collect and store the data securely. It can range from annotated bibliography to surveys, literatures, or interviews.

Through out the research work, I will follow APA7 for citations and reference to give due credit to all the authors, editors, and publishers and keep the credibility of my work also. I will be honest in analysis of literature and shall report the answers to the research questions.

It is important that every research in any field should enhance the quality of life. I will work on this project in away that it will help the professionals, students, researchers of this field which in return will be able to enhance every one’s quality of life.

For a researcher it is essential to work competently, and I shall comply with it.

Ethics always compel every professional to enhance professionalism. I shall increase the integrity of the institution I am in, Australian Computer Society and all the teachers, members, colleagues, and my supervisor for this research.
  5.     Compliance Requirement One of the most mandatory characteristics of a professional is to comply with ethics, code of conduct and rules set by workplace, industry or by legislative authorities. In this research project, the researcher needs to comply with Code of Ethics set by Australian Computer Society. This will cover all industrial and governmental compliance requirements.

 

 

Project Plan

  1. Deliverables:
  • Project Proposal and Plan
  • Weekly Progress Report
  • Annotated Bibliography
  • Final Report
  • Final Representation and Seminar
  1. Work Breakdown Structure (WBS)

WBS

  1. Risk Analysis

 

Risk ID Description Likelihood Consequence Treatment
1 Problem Domain is not well defined. LOW MEDIUM Revise and develop a more aligned and detailed oriented problem domain.
2 Purpose and justification are not well argued. LOW LOW Research and justify with primary studies literature.
3 Timeline is not well thought. LOW HIGH Understand every task in the timeline. Before creating a timeline, investigate the timelines of similar projects for guidance.
4 Lack of academic literature on the proposed topic. LOW MEDIUM Use advance search tools to find more relevant literature.
5 Research Questions are not answered, or conclusion is vague. MEDIUM HIGH Revise methodologies and search more relevant literature.
6 Personal knowledge and skills are lacking. MEDIUM HIGH Read more literature and improve your skills. Ask professionals for help. Attend workshops and seminars in university.
7 Size of Project does not match duration of the course. LOW MEDIUM Scope of the project can be adjusted accordingly.
8 Outside factors e.g., health or mental stress or Covid-19. LOW HIGH Take care of your health and follow all precautions related to Covid-19.
9 Violation of Ethics and Compliance requirement. LOW HIGH Need to be careful about it otherwise credibility of the research will be badly impacted. Follow personal and professional ethics and compliance requirements.

 

  1. Duration: 19th July- 2021 to 05th Oct- 2021.
  2. Gantt Chart

 

 

 

 

Reference

  1. Memon, N., Xu, J. J., Hicks, D. L., & Chen, H. (Eds.). (2010). Data Mining for Social Network Data.
  2. Guo, Y., Liu, Y., Oerlemans, A., Lao, S., Wu, S., & Lew, M. S. (2016). Deep Learning for Visual understanding: A review. 187. 27-48. https://doi.org/10.1016/j.neucom.2015.09.116.
  3. Zammari A., & Benois-Pineau J. (2020). Deep Learning in Mining of Visual Content. Springer. https://doi-org.ezproxy.csu.edu.au/10.1007/978-3-030-34376-7_1.
  4. Schirrmeister, R. T., Springen, J. T., Fiederer, L.D.J., Glasstetter, M., Eggensperger, K., Tangermann, M., Hutter, F., Burgard, W., & Ball, T. (2017). Deep learning with convolutional neural networks for EEG decoding and visualization. Human Brain Mapping, 38(11), 5391-5420. https://doi-org.ezproxy.csu.edu.au/10.1002/hbm.23730.
  5. Paterson, T., & Hanley, L. (2020). Political warfare in the digital age: cyber subversion, information operations and ‘deep fakes’. Australian Journal of International Affairs, 74(4). 439-454. https://doi-org.ezproxy.csu.edu.au/10.1080/10357718.2020.1734772.
  6. Maddocks, S. (2020). ‘A Deepfake Porn Plot Intended to Silence Me’: exploring continuities between pornographic and ‘political’ deep fakes. Porn Studies, 7(4). 415-423. https://doi-org.ezproxy.csu.edu.au/10.1080/23268743.2020.1757499.