ARTIFICIAL INTELLIGENCE HT1 Assignment 3 (COIT20277) Assignment Help

Assignment 3 – Written and Programming Assessment

Due date: 11:55 pm AEST, Friday of Week 12 (31 May 2024)

Weighting: Mode: 45%

Group (2-3 students)

Length: Written Report (40 marks) – Maximum 2,000 words (Group Submission) Reflection (5 marks) – Maximum 500 words (Individual Submission)

Full Mark: 45

Objectives

This assignment is designed to reinforce the knowledge and skills acquired from Week 6 to Week 11. The assessment task relates to Unit Learning Outcomes 1, 3 and 4.

Problem Description

This assignment is related to age estimation across ages from face images. It is a challenging yet significant task with many real-world applications, such as security, entertainment, marketing, healthcare, and human-computer interaction. Traditional approaches to age estimation often relied on manual feature extraction followed by machine learning algorithms. However, with the advancements of deep learning and convolutional neural networks (CNNs) in recent times, age estimation has seen remarkable progress.

At the same time, transfer learning which we have discussed in Week 9 and Week 10, is a practical technique where a CNN model trained on a large dataset is fine-tuned on a smaller dataset for a specific task. This technique can improve the performance of CNNs, including for age estimation.

Objectives

In this assignment, you will apply transfer learning to predict the age of individuals based on face images trained on your choice of deep learning models. Specifically, this assignment aims to achieve the following learning objectives:

1. Deepen Understanding of Transfer Learning: You will gain a deeper understanding of transfer learning and its utilisation in leveraging pre-trained models for age estimation tasks. You will attempt to adapt pre-trained CNN architectures for age estimation by training on a dataset of face images.

2. Data Preprocessing and Model Modification: You will perform data preprocessing in preparing the face image dataset for model training. You will practice modifying the output layer of pre-trained CNN models to predict continuous age values instead of discrete classes or categories.

3. Training and Evaluation: You will implement the training pipeline for the modified CNN models chosen and evaluate their performances using appropriate loss metrics such as mean absolute error (MAE) or mean squared error (MSE).

4. Analysis and Interpretation of Results: Through experimentation, you will analyse the performance of the trained models and interpret the results obtained.

Face Image Dataset:

A dataset of face images is provided with the assignment, in the form of a zip archive in the same folder as this assignment specification. The dataset has the following main characteristics: • Contains a total of 996 face images across 57 ages.

• Images of the same ‘age’ are stored in the same sub-folder identified by the ‘age’. For example, ‘00’ is the name of the sub-folder that contains face images of individuals under a year old, ‘01’ refers to the one-year-old, ‘02’ refers to the two-years-old, so on and so forth.

• For some individuals, the dataset includes face images across several ages to facilitate model training and evaluation.

• The filename of each face image is constructed by concatenating the folder name, an underscore (_), and a number between 1 and the number of images in the folder. • All face images are in the JPEG format.

Tasks

Your group will complete the following tasks in this assignment:

1. Data Preprocessing: Prepare the dataset for model training, which might include resizing, normalisation, and other image processing operations. Detail the data preprocessing steps applied, and the decisions made in the report.

2. Model Selection and Modification: Choose a minimum of two pre-trained CNN models to apply transfer learning. Compare the chosen models and make necessary modifications to their architectures for age prediction (output layer, loss function, etc).

3. Model Training: Discuss the model training process, including the creation of the training and testing sets, hyperparameter tuning, and optimization techniques. Report the training and validation metrics obtained on each trained CNN model and discuss their metrics in the written report.

4. Model Evaluation: Explain the performance metrics used to evaluate each trained model’s accuracy, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), etc. Visualise each trained model’s predictions when compared to actual ages on the testing set. Analyse the results in relation to overfitting, underfitting, and generalizability.

5. Model Interpretation: Interpret the evaluation results, key findings, and potential future improvements. Discuss the factors that might influence the CNN models’ performance and potential limitations.

6. Written Report (Group Submission): A maximum of 2,000 words, which should cover: o Data Preprocessing

o Model Selection and Modification

o Model Training and Evaluation

o Model Interpretation

o Conclusions and Insights

7. Reflection (Individual Submission): A maximum of 500 words, which should include: o In this assessment, what were the tasks that you have contributed to in the group? How has the completion of these tasks help reinforced your own understanding of deep learning, CNNs and the transfer learning approach?

o Evaluate member(s) in your team in relation to the tasks they were assigned. o Describe any challenges encountered during group work and how they were addressed.

Submission

The assignment submission has two parts:

1. The Written Report and the Python program (.ipynb) are two separate files. Please do not combine them into a single zip archive. Only one member in the group needs to submit the two files via the group submission link on Moodle on behalf of the group.

2. The Reflection should be submitted individually by each student. A separate submission link on Moodle will be available, just next to the group submission link.

Marks will be deducted for not following these instructions. Late submission will incur a late penalty (5% of the total mark per day) as per the university’s Assessment Policy and Procedure.

Marking Criteria (maximum 45 marks)

Criteria	Sub-Criteria	Marks	Description
Data Preparation and Preprocessing	Code clarity & adequate comments	5	Well-structured and documented code for data preparation and preprocessing.
Model Selection and Modification	Model selection	5	Choose a minimum of two pre-trained CNN models to apply transfer learning. Output the architectural details of the chosen pre-trained models.
Model Selection and Modification	Model Modification	7.5	Make necessary modifications to the pre-trained model architectures to achieve age prediction (output layer, loss function, etc).
Model Training	Model Training	5	Creation of the training and testing sets, carry out hyperparameter tuning, and perform gradient descent optimization.
Model Evaluation and Interpretation	Model Evaluation	5	Compute the performance metrics used to evaluate each trained model’s accuracy. Visualise each trained model’s predictions when compared to actual ages for the testing set.
Model Evaluation and Interpretation	Model Interpretation	2.5	Analyse the model evaluation results and interpret key findings.
Written Report	Organization & clarity	2.5	Well-organized, logical, and easy-to understand report structure.
	Content & comprehensiveness	5	Covers all required aspects of the assignment, including analysis, results, and discussion, including references.
	Visualizations & presentation	2.5	Effective use of visualizations and formatting to clearly convey findings.
Reflection	An individual reflection on the experience acquired in completing the assignment.	5	Reflection on what challenges, lessons and knowledge that have been gained.

Note: You must follow the APA 7 or the latest version for citation and referencing guidelines when writing your report.

Marking Rubric

Criteria

Excellent

(100%)

Good (70%)

Satisfactory

(40%)

Unsatisfactory

(10%)

Data

Preparation and

Preprocessing

Correct,

well

documented,

and efficient

Mostly correct

and

documented

Some errors or

inefficiency

Incorrect or

incomplete

Model Selection and

Modification

Clear

explanation

of purpose

and choice

Basic

understanding

Limited

understanding

Incorrect or

incomplete

Model Training

Clear, well

commented,

and correct

Mostly clear

and correct

Some errors or

missing

information

Incorrect or

incomplete

Model

Evaluation and

Interpretation

Correct,

well

documented,

and efficient

Mostly correct

and

documented

Some errors or

inefficiency

Incorrect or

incomplete

Written Report

Covers all

required

aspects of the

assignment,

including

analysis,

results, and

discussion.

Covers most

required

aspects of the

assignment,

including

analysis,

results, and

discussion.

Covers some

required

aspects of the

assignment,

including

analysis,

results, and

discussion.

Incorrect or

incomplete

Reflection

Effectively

summarizes

the student’s

learnings

from the

assignment.

Cover the key

aspects

mentioned

above, but it

might lack

depth or detail

in some areas.

Attempt to

address the

required points but might be

incomplete or

lack clarity.

Missing crucial

aspects or is not

relevant to the

task.

Academic Integrity

Students must write the report, the reflection, and the Python program themselves. You may be asked to prove that you have written these items.You should keep evidence that you have written them yourself, for example, early drafts of your report and the Python code.

ALL assignments will be checked for plagiarism (materials copied from other students and/or material copied from other sources) using TurnItIn. If you are found to have plagiarised material or if you have used someone else’s words without appropriate referencing, you will be penalised for plagiarism which could result in zero (0) marks for the whole assignment. If you falsify references/information you will also be penalised. In some circumstances, a more severe penalty may be imposed such as having a plagiarism incident raised. Please refer to CQ University’s policy and procedure on academic integrity for details.

Once the assessment is marked, the Unit Coordinator (or nominee) may request additional written information and/or an oral discussion to clarify the student’s understanding of the submitted work. Failure to comply and/or to demonstrate an understanding of the assignment’s submitted items could result in 0 marks for the assignment.

The following list of pertinent courses includes some of the topics that knowledgeable and experienced teachers at Gradespire specifically address: