Artificial Intelligence (MDA522 Assignment 2) Assignment Help
Artificial Intelligence (MDA522 Assignment 2) Assignment Help
GenAI
In this assignment, the use of GENERATIVE AI (GenAI) IS PROHIBITED.
Purpose of the Assignment:
This assignment seeks to equip students with many desirable skills and knowledge in artificial intelligence (AI) and related disciplines, and to fulfill the intended learning outcomes of the unit. In particular, students are expected to use AI models, techniques and tools to solve complex real-life problems which are topical and relevant to modern industry. Students are grouped into 4 members each. All groups will work on the same general topic in a competitive manner, use similar tools to acquire similar set of skills. The task is done by the groups in a competitive manner in the sense that it will judge as the winner the group whose deep learning model detects the target with the highest predictive accuracy and lowest error. The project also has a research component besides practical programming in Python. By working in groups, students also acquire teamwork skills.
Project/Task Background:
This is a follow-up to Assignment 1 to enable students to further develop their skills in the applications of AI in computer vision. The project in this assignment is titled “Development of AI Algorithms to Detect Explosives in Waste Receptables Using Vision Transformers”. Some cities have witnessed the concealment of improvised explosive devices (IEDs), or explosives, in public waste receptacles by criminals to destroy people and property. This menace is causing mass removal of dustbins from public places and public transport in cities by city councils the world over as a strategy to solve the problem [1]. The consequence is littering to cause stench, pests, contamination of water bodies, spread of diseases and increased greenhouse gases.
The current solutions to the problem are blast-resistant and transparent dustbins. However, none of these approaches is optimal. First, none of them inherently catches the criminals to deter the act. Second, the awful appearance of rubbish is visible with transparent bins, while blast-resistant dustbins are bulky and costly.
AI algorithms used in computer vision have been historically dominated by convolutional neural networks (CNNs). Currently, the application of Vision Transformer (ViT) in computer vision is increasing. ViT is a model for image classification that employs a transformer-like architecture over patches of the image. A ViT converts an image into a sequence of non-overlapping patches, similar to how language models (e.g. transformers, LSTM and RNNs) handle text. Each patch is then transformed into a vector using a Large Language Model (LLM) and processed within a transformer architecture. ViTs are capable of grasping global information within images, transcending the limitations of local feature extraction performed by CNNs.
This assignment seeks to motivate students to develop a novel approach to protect dustbins from illicit use as we maintain them for their traditional role. This is a new research area so there may not be much available literature or datasets. The project is an intersection of AI, IoT and environmental security. However, the task will focus on the design of AI algorithms.
Methodology:
Students are expected to build AI/ML/DL models to be implemented in waste receptacles or rubbish bins to detect if rubbish thrown into a waste bin is a real waste or a potential explosive. They should search and review existing algorithm and techniques used. Then, attempt to design a new algorithm/model that is better in some sense than the existing ones, if any. Object detection is an image processing task under computer vision. Tasks involved include:
1) Search for an existing database of images of waste/garbage, and features of waste and detonations or IEDs to build a new database.
2) Through an extensive search select the appropriate AI algorithms and Python libraries used for image classification.
3) Use the tools and database constructed to train AI/DL/ML models based on ViTs to classify waste into either:
(a) 2 groups: real waste/garbage or IED. This will build a deep learning model for binary classification.
(b) More than 2 groups: real waste/garbage sorted into classes and IED being another class. This will build a deep learning model for multi-class classification.
Project Objectives and Deliverables:
The project falls under object detection in computer vision, and the following are the two deliverables or milestones expected in this project:
• Use the existing database constructed in [1], extend the database in [1] or construct a new database of IED images which is good enough, regarding veracity, volume and value, to solve the target problem.
• Construct of a deep learning model based on ViTs, using Keras in Jupyter Notebook, which can detect IEDs (explosives) thrown into a waste receptacle. This is the target problem.
Needed Tools:
• Laptop (students should come to class with their laptops)
• Python, Numpy, SciPy, Matplotlib, scikit-learn, Keras/Tensorflow, etc. running in Jupyter Notebook in Anaconda Python distribution.
Submission Instructions:
1) Each group shall submit only one report to the appropriate Moodle link for marking and grading.
2) Deliverables: the report must be in the following three (3) formats:
a. Groupk_Ass2.html: This is the programming/practical part which should be submitted in a single Jupyter Notebook under the name:
File → Download as → HTML (.html)
By executing from Jupyter Notebook. The Jupyter Notebook must use the Markdown markup language to explain the solution to the task concisely but clearly.
b. Groupk_Report_Ass2.doc: This is the file containing the project report. All figures and results in the Jupyter Notebook must be captured in this report.
c. Groupk.ppt: This is the file that summarizes the project report to be presented in class in Week 11.
where k is the Group number.
Note: Moodle cannot any file exceed 200 MB. Such files should be stored on a cloud platform and loaded into the Jupyter Notebook in a way that can be run without any further ado.
3) Format of group report (i.e. Groupk_Report_Ass2.doc). Each group report must be formatted as IEEE Journal paper with a cover page showing names and Student ID of students in group. The cover page should also contain a table showing the contributions of each student to the project and the reports. The sections of the main report should use the titles and numbering:
a. Title
b. Abstract
c. I. Introduction/background
d. II. Literature review (or related works)
(This section analyses works completed and published that relate the topic under review. Cited works should include at least 5 peer-reviewed journal papers)
e. III. Materials and Methods (aka Research Methods)
A. Data Used for the Model Training
(This subsection should include the source of the data and its
description/features)
B. Method/Algorithms/Models Used (the VIT model/s)
C. Experimental Setup
D. Model Training and Experiments
f. IV. Results and discussions
[This section documents a thorough comparison of the performance of any alternative methods used, or alternative values for model parameters, etc. How results could be improved, etc.]
g. V. Conclusions
h. Acknowledgment
i. References
(List of ONLY works cited in the body of the project report and presentation slides. IEEE referencing and bibliographic style must be perfectly adhered to.)
• Note that some of the sections in the report should not be numbered, namely Title, Abstract, Acknowledgement (if any) and References.
• Use illustrations in the form of figures and tables (copied from Jupyter Notebook) wherever they deem fit. Learn from the related works that you study as you do the project. If you read one paper per week, you would have studied at least 11 papers by the time the project is completed.
Final grade = 0.3*Presentation Grade + 0.7*(grade for project report and Jupyter Notebook grade)
Marking Guide for the Oral Presentation
Each group shall present a summary of its work in class in Week 11. The presentation shall be graded using the criteria in the table below.
Marking Criteria: Rubric for the Main Report
Your submission will be marked using the criteria in this table.
Note: Research question should not stand a lone as a section, but integrated into the abstract, introduction, literature review, results and conclusions.
References
[1] A Gyasi-Agyei, “Detection of explosives in dustbins using deep transfer learning based multiclass classifiers,” Appl. Intell. , vol. 54, no. January 2024, p. 2314–2347, 2024.
[2] Gyasi-Agyei A (2023) I2Net: Database of IED Images for IED Detection in Public Waste Receptacles