COMP4436 AIoT Assignment I - Comparative Analysis of ML, DL and SNN Algorithms in AIoT Applications
[2024-25] COMP4436 AIoT Assignment I
Comparative Analysis of ML, DL and SNN Algorithms in AIoT Applications
Objective: The goal of this assignment is to implement and compare the performance of various machine learning (ML) and deep learning (DL) algorithms, including a spiking neural network (SNN), in the context of an Artificial Intelligence of Things (AIoT) application. By using a dataset consisting of images of cats and dogs (or a dataset of your choice), you will gain a comprehensive understanding of supervised and unsupervised learning techniques and their applications in AIoT environments.
Relation to AIoT: This assignment is closely related to AIoT as it showcases how machine learning and deep learning models can be integrated into intelligent systems that process visual data from connected devices. In AIoT, devices such as cameras, sensors, and other smart devices generate vast amounts of data. The ability to classify and analyze this data in real-time using advanced algorithms, such as CNNs and SNNs, exemplifies the intersection of AI and IoT technologies. By applying these models to image data, the assignment demonstrates how intelligent decision-making can be achieved in AIoT applications, leading to improved automation, efficiency, and user experiences in smart environments.
Data Collection: Utilize a dataset containing images of cats and dogs (or a dataset of your choice). (https://www.kaggle.com/datasets/samuelcortinhas/cats-and-dogs-image-classification)
Algorithms: You should implement 5 algorithms of your choice, selecting one from each category and using the same dataset for all algorithms: Supervised ML Algorithm, Unsupervised ML Algorithm, Supervised DL Algorithm, Unsupervised DL Algorithm, Spiking Neural Network (SNN)
Evaluation Metrics: Algorithms should be compared based on various metrics, including but not limited to:
Accuracy: Proportion of correctly classified instances, Precision: Ratio of true positives to the total predicted positives, Recall: Ratio of true positives to the total actual positives, F1-Score: Harmonic mean of precision and recall, Runtime Efficiency: Time taken for each algorithm to complete the training and evaluation process.
Visualization: Generate comparative graphs and tables to present results, facilitating easier analysis of the performance of each algorithm. This will include plots of accuracy across different algorithms.
Expected Outcome: The assignment will culminate in a comprehensive report featuring detailed graphs and tables comparing the results of all implemented algorithms. This report will highlight the superior performance of the spiking neural network in classifying images of cats and dogs, demonstrating its potential advantages in AIoT applications.
Submission: Please use the attached report template as a reference. The report should be 6-8 pages long, formatted in two columns, and include elements such as graphs, flowcharts, equations, tables, and references. Ensure that all source code files, datasets, and relevant materials are uploaded to Blackboard. Your Python code should be compatible with VS Code and include a README file that explains how to execute the code.
Reference paper (for report writing): https://arxiv.org/abs/2001.09636
BONUS Points: 2.0 points awarded if you achieve an SNN accuracy of over 85% with correct implementation and 0.25 points awarded if you propose a unique title for your report. ZERO points will be assigned if any evidence of copying is discovered, regardless of an appeal.
*** The End ***
