Today’s advanced Large Language Model (LLM) technologies, such as OpenAI o3, Google Gemini, and DeepSeek-R1, have revolutionized the way we learn, and are best suited for “learning by doing” — a hands-on experience that prioritizes practical experiments. The method is superior to the traditional theory-first approach. At the heart of this approach is the importance of working on a meaningful project—one that sparks passion, excites curiosity, and motivates us to dive in and see it through to completion.
To begin, students must form teams of five members by Week 3 and select a team leader. The team will independently take on a deep learning AI project that both excites and challenges them, with minimal supervision. The team leader will be responsible for submitting all group materials, including the member list, project proposal, and final report. This structure encourages collaboration, initiative, and ownership of the learning process.
In this project, you have the freedom to choose a topic that truly ignites your passion, and select a project type that aligns with your interests.
The only criteria, it must be interesting, challenging, and ultimately useful to you.
Your passion will be your compass, guiding you on a journey of exploration and innovation. -->
Students typically undertake one of FIVE types of projects:
- Create Your Own Data and use an existing approach ⭐
- The main focus is on collecting data and making it useful for existing DL methods. A good example is that you collect some images and annotate them, and then train an existing image classifier to perform a specific task on these images
- Create Your Own Method on an existing dataset ⭐⭐
- Taking an existing dataset and adapting an existing method to make it your own DL method. You modify parameters, work with existing neural networks, apply what you've learned in the lecture, and aim to make them more efficient.
- Beat the Classics - Implement a DL method and compare to non-DL baseline ⭐⭐⭐
- Challenge the state-of-the-art using algorithms without DL or with DL algorithms that aim to surpass this baseline. Compare the results and demonstrate the ability to outperform the baseline, referred to as "beat”.
- Create Your Own AI-Powered Application ⭐⭐⭐
- Create a practical AI-powered tool or service addressing a real-world problem using advanced AI and deep learning technologies. Demonstrate creativity and technical proficiency in a functional solution.
- Beat the Stars - Improve the state of the art ⭐⭐⭐⭐
- By selecting a research paper, the goal is to demonstrate how to outperform the current state-of-the-art papers. It's important to note that due to the rapid pace of paper publication, it might be difficult to stay up-to-date with the latest advancements. Nonetheless, the objective is to select a recent paper as a baseline and attempt to beat it.
Group Project Assessment:
- Project Team Formation (Week 3)
- Create a team of 5 people and choose a team leader.
- The team leader is responsible for submitting the group project assessment materials, such as member list, proposal, and final report.
- Send the list of team members and the elected team leader to the Instructor at eelmpo@cityu.edu.hk
- Deadline: Feb 3, 2025
- Project Proposal (Week 4)
- A 5-page project proposal (not include references).
- Submit the project proposal in PDF format to CANVAS proposal assignment.
- Proposal must contain:
- Project Title
- Student Name, Student ID and Email Address of each member
- Summary with goals of the project in about 300 words.
- Other suggested content: The group proposal outlines the problem, objectives, methodology, dataset, baseline selection, experimental setup, timeline, evaluation plan, collaboration plan, risks, contingency plans, ethical considerations, and references for the group project. It serves as a roadmap, ensuring clarity and alignment among team members.
- References
- Deadline: Feb 14, 2025
- Oral Presentation (Week 12 and 13)
- The oral presentation assesses students' communication skills, including their ability to clearly convey project objectives, methodology, and findings. Their response to questions gauges their understanding of the project and role within the team.
- Every group is also required to make a 10-minute Power Point presentation of their group project to the entire class.
- The presentation must include:
- A short description of the project and its objectives
- An explanation of the implemented algorithm and relevant theory
- A demonstration of the working program – i.e., results obtained when running the program
- Final Report, PPT, Source Code and Demo Video (Week 14)
- The final project report is expected to be a minimum of 25 pages in length. To assist you in preparing your report, two reference documents have been provided: the Final_Report_Template and the DeepSeek Technical Report:
- Students are allowed to use one report style of these two templates. However, the structure outlined in the template is intended to serve as a flexible guide rather than a strict blueprint for your final report. While the chapters and sections presented in the template are commonly found in research theses or technical reports, the specific nature of your research may require adjustments to this structure. Additionally, the order of items within chapters can be adapted as needed. The template follows the traditional technical report structure, which aims to present a coherent line of argument across six key chapters: Introduction, Literature Review, Research Design, Results, Discussion, and Conclusions.
- Your final report must include Appendix A , titled "Individual Contributions of the Group Project." This appendix is a dedicated section where students provide detailed information about each team member's contributions to the group project. Specifically, it should describe each member's responsibilities, the tasks they completed, and the outcomes they achieved. This appendix is critical for evaluating individual performance within the group project and is necessary to meet the professional accreditation requirements of the course.
- Demo video is required to be a 3-4 minute summary of the project.
- Students are also required to submit the Python source code of any implementation and PPT of the oral presentation for assessment.
- The final report must include an Appendix A for “Individual Contributions of the Group Project”, in which students provide detailed information about each team member's contributions to the group project. This includes describing their responsibilities, the tasks they completed, and the outcomes they achieved. This appendix is important for assessing individual's performance in the group project, which is necessary for meeting the requirements of professional accreditation of the course.
- All the PPT, Final Report and Source Code are required to submit to CANVAS Group Project Final Report
- Deadline: April 25, 2025
Project Hints
- Your passion is your compass.
- A high-quality project would be one that has the potential to be published or nearly published. It is anticipated that some students will continue working on their projects even after completing the course, with the aim of submitting their work to conferences or journals. To gather inspiration, you can explore recent research papers in the field of deep learning, particularly from conferences like ICML and NeurIPS.
- Students are strongly encouraged to leverage advanced AI tools like Elicit, SciSpace, Consensus, and NotebookLM to identify compelling research topics, streamline literature reviews, and deepen their understanding of academic articles. These tools can significantly enhance the efficiency and quality of their research process.
- Once you have identified a topic that interests you, it is beneficial to search for existing research on related subjects using academic search engines such as Google Scholar.
- Another crucial aspect of project design is identifying suitable datasets for your chosen topic. If the data requires significant preprocessing or you plan to collect it yourself, keep in mind that this is just one part of the overall project work and can often consume a substantial amount of time. Nonetheless, it is important to maintain a solid methodology and engage in thorough discussions of the results, so be mindful of pacing your project accordingly.
- For Application type group projects, here are some suggested AI-Powered Applications topics. Use these ideas as a starting point to learn how to build an AI application, then modify and enhance them to create your own unique solution.
Presentation Schedule:
Section A (Week 12)
Group 1: AI Powered Auto-presentation Generator
- DANG Xiangyu (Leader), KE Linfeng, MA Runlang, ZHANG Hao,SONG kunyang
Group 2: Emotion Recognition System Based on Multi-Modal Data Fusion
- Zhang Yujie (Leader), Wu Yalin, Liu Yunju, Hao Yangge, Zhao Zewei
Group 3: HK Remote Sensing Road Extraction
- SHI Junshen (Leader), GUO Dongyao, SHI Ruijie, ZHAO Haijian, BO Runqing
Group 4: Robot Image Recognition & Grasping Technology Based on Deep Learning
- Peng Haoyang (Leader), Cheng Wenxin, Zhang Yu, Huang Wenying, Chen Tianyu
Group 5: EEG Emotion Recognition and Detection Based on DEAP Dataset
- Zhang Zicheng (Leader), Luo Di, Wu Ruike, Fan Xingyu, Gu Ziyang
Group 6: An AI tool for fruit recognition
- He Yunze (Leader), Xie Xifeng, Li Caiting, Zhang Yuwei, Lyu Liangyi
Group 7: Real-Time ECG Signal Classification Using Hybrid Deep Learning Models for Early Cardiac Disease Detection
- HE Zhili (Leader),YUAN Jun, WANG Zihao, Hu Zhixiang, YANG Wenyu
Group 8: Real-time Gesture Detection System Based on YOLO
- SHI Yi (Leader), GU Jiali, WEN Lumin, SHEN Bofei, GAO Rui
Group 9: Video Motion Recognition and Prediction Based on Transformer Architectures
- Rui Jingyu (Leader), FANG Yizhou, ZHANG Jiahua, CHENG Zhongwu, Yang Ruoy
Group 10: Bio-Inspired Cross-Temporal Interaction Network for Lightweight Aerial Tracking
- Yuchong WANG (Leader), Faxue LIU, Qiance ZHAO, Yuzhuo CHANG, Zhongpei YANG
Group 11: Development and Optimization of a Flower Species Recognition System Based on
Convolutional Neural Networks
- LIN Congjie (Leader), Guo Zhenhui, YANG Tianhao, Li Bingzheng, WANG Jiawei
Group 12: Sentiment Analysis using Deep learning
- LIU Zhen (Leader), WANG Wenrui, LIANG Jianyi, QU Zhirui, Hou Yangguang
Group 14: GANs-based model to generate emotionally expressive facial videos
- Wang Yu (Leader), MA Jinwen, SHI Dailong, LIU Yifei, RUAN Yujie
Group 15: Smart Waste Sorting Assistant (EcoBot)
- HUANG Zuliang (Leader), CAO Shuning, CHEN Bo, WANG Mingchuan, ZHONG Chenhui
Section B (Week 13)
Group 16: Deep Learning-Based Vehicle Recognition System
- Feng JIANG（Leader), Beijia Fu, Ziheng ZHU, BinYan Zhang, Dewei Kong
Group 17: Application of BEV Sensing for Urban Road Traffic Scenarios
- DONG Luoyu (Leader), DONG Yixiao, LIANG Lizheng, LIU Zefeng, LYU Run
Group 18: Neural Style Transfer for Image and Video Stylization
- XU Zixin (Leader),YANG Xinyao, ZHONG Yangmu, YAO Yongbin, Li Xucen
Group 19: RSNA Pneumonia Detection Based on Improved YOLO11
- Cheng Wei (Leader), Zhang Min, Chi Hongye, Deng Borong, Li Zhelun
Group 20: BART-Based Automated Text Summarization
- GAI Tianyu (Leader), LIU Yan, WU Fan, TAN Haoran, WANG Nanting
Group 21: Beat the Classics in Medical Image Classification
- LI Rongrong (Leader), ZHANG Ziyao, WANG Ruchen, LU Sitao, ZHANG Zihan
Group 22: Robust Depth Estimation for Autonomous Systems in Complex Environments
- Zhou Yifan (Leader), He Xiao, Liu Tingkang, Zhou Dayi, Yao Jiaying
Group 23: Modeling the Spread and Eradication of Asian Giant Hornets
- SI Yuzhou (Leader), SU Zixuan, BAI Jiayi, ZHOU Wenxin, YAN Jinshi
Group 24: InvestAgent
- ZHENG Yaru (Leader), XIAO Yixuan, WANG Kexin, ZHENG Yuhui
Group 25: Diffusion Model for Object Counting
- Long Yang (Leader), Ma Di, Song Guangyu, Xiong Yujue
Group 26: VoxAI: AI-Powered Voice Generation and Virtual Avatar System
- WANG Zhi (Leader), HUANG Haoyang, LI Jialin, Wei Zhirui, Lin Yechao
Group 27: Medical diagnosis system based on large language model
- Yan Ping (Leader), Dong Yifan, Zhang Mingqi, Ge Lanhua, Li Xinyi
Group 28: Med-healthcare bot based on Large Language Model
- Zhaojin GOU (Leader), Ji Yang, ZENG Ying, SHU Haorui, Yang Liu
Group 29: Deep Reinforcement Learning for Automated Stock Trading
- Tian Ganlin (Leader), Wei Li, Liang Yaohe, Lu Yalin and Tian Wenhao.
Group 30: Movie Review Estimation
- Omar Mohamed Salama Sayed MOUSTAFA (Leader), Han Taixuan, LUO Zeyu
