Self-Learning Mini-Projects in Advanced Topics of
Machine Learning
Mini-Projects Presentations
As part of our advanced ML self-learning initiative, each team will select ONE advanced topic from
the list below and prepare a full mini-project presentation.
This task is designed to push you beyond what we covered in training and help you explore deeper
areas of ML through theory + implementation + presentation.
Instructions:
1. Each team must choose ONLY one topic from the list below.
• No two teams can choose the same topic.
2. Teamwork:
• Each team = 2 members.
• Work must be distributed equally between both members.
• Each member is responsible for a clear role (theory/explanation vs coding/implementation).
3. Presentation Requirements:
Each team must prepare a complete presentation that includes:
• Theory & Concept Explanation.
• Aim of the method/idea.
• Use cases and real-world applications.
• Importance & limitations.
• Full code implementation (with explanations).
• Dataset used (with justification).
• Graphs, figures, and final results.
• Clear demonstration that the idea works.
4. Presentation Format:
• Each member presents the part he/she worked on.
• Slides + Code demo (notebook or script).
Topics to Choose From:
1) Model Interpretability & Explainability (using SHAP or LIME)
• Goal: Understand and explain how ML models make decisions.
• Task:
o Choose any dataset (classification/regression).
o Train a model (e.g., Random Forest, XGBoost, Neural Network).
o Apply SHAP or LIME to explain model predictions.
o Show feature importance, local explanations, visualizations.
o Discuss why interpretability is important and its limitations.
2) Fairness, Bias & Ethical ML
• Goal: Detect and mitigate bias in ML systems.
• Task:
o Choose a dataset where fairness issues may arise (e.g., loan approval, gender bias, age
bias).
o Train a model and evaluate fairness metrics (e.g., demographic parity, equal
opportunity).
o Demonstrate examples of bias in predictions.
o If possible, apply a debiasing technique.
o Discuss ethical implications of biased AI systems.
3) Self-Supervised Learning
• Goal: Learn data representations without labels.
• Task:
o Choose either image data (e.g., MNIST, CIFAR) or tabular data.
o Implement a self-supervised method, such as:
▪ Autoencoders (reconstruct input).
▪ Contrastive Learning (SimCLR-lite: predict similarity of augmented pairs).
▪ Masked prediction (predict missing parts).
o Show how features learned in SSL can be transferred to a downstream task (like
classification).
o Discuss how SSL reduces dependency on labeled data.
4) Meta-Learning (“Learning to Learn”)
• Goal: Build models that can adapt quickly to new tasks with little data.
• Task:
o Focus on image-based meta-learning (not NLP yet).
o Implement either:
▪ Prototypical Networks (few-shot classification).
▪ MAML (Model-Agnostic Meta-Learning) on a simple dataset (e.g., Omniglot,
MNIST).
o Show how the model performs when given a new class with only a few examples.
o Discuss why meta-learning is important for few-shot learning, personalization,
robotics, etc.
Evaluation Criteria
Each team’s work will be evaluated based on:
1. Presentation Skills – clarity, confidence, flow, and ability to engage.
2. Quality of Implementation – correctness, complexity, and completeness of code.
3. Self-Learning – how well the team explored and understood advanced concepts
independently.
4. Organized Work – structure of slides, logical flow, and clarity of explanations.
5. Resources Used – quality of references, tutorials, papers, or documentation that supported
your project.
6. Teamwork & Management – equal contribution from both members and smooth
collaboration.
Final Notes:
▪ Do your best effort on this project, treat it like a mini research + coding assignment.
▪ Be creative: show interesting graphs, comparisons, and insights.
▪ Your work will be evaluated not only on results but also on how well you explain and justify
your approach.
Submission Deadline:
▪ The project must be submitted no later than the middle of next week.
▪ You will be notified of the exact day and time of presentations in advance.