Skip to content

Final Projects

The final project is a team-based research project where you'll explore an open problem in trustworthy machine learning.

Project Overview

Team Size: 3-4 students
Timeline: 8 weeks (Weeks 8-15)
Weight: 30% of final grade

Project Components

Component Weight Due Date
Proposal 5% Week 11
Progress Report 5% Week 13
Final Presentation 10% Week 15
Final Report 10% Finals Week

Project Guidelines

Scope and Topics

Your project should address a research question in one or more areas of trustworthy ML:

  • Fairness & Bias: Novel detection methods, mitigation algorithms, evaluation metrics
  • Robustness: New attack techniques, defense mechanisms, certified methods
  • Interpretability: Explanation methods, evaluation frameworks, human studies
  • Privacy: Differential privacy mechanisms, federated learning innovations
  • Safety & Alignment: Value learning, reward modeling, verification methods

Requirements

  • Novelty: Propose new methods or provide new insights
  • Implementation: Working code with experiments
  • Evaluation: Rigorous experimental validation
  • Writing: Clear technical exposition

Deliverables

Project Proposal (2 pages)

Due: Week 11

Required Sections: 1. Problem Statement: What challenge are you addressing? 2. Related Work: Brief survey of relevant papers (5-10 papers) 3. Approach: Your proposed method or analysis 4. Evaluation Plan: Datasets, metrics, baselines 5. Timeline: Milestones for remaining weeks

Progress Report (1 page)

Due: Week 13

Required Sections: 1. Progress Summary: What you've completed 2. Preliminary Results: Initial findings or implementation 3. Challenges: Issues encountered and solutions 4. Updated Timeline: Revised plan for final weeks

Final Presentation (15 minutes)

Date: Week 15

Presentation Structure: - Problem motivation (2-3 min) - Technical approach (5-6 min)
- Experimental results (4-5 min) - Conclusions and future work (2-3 min) - Q&A (5 min)

Final Report (8-10 pages)

Due: Finals Week

Required Sections: 1. Abstract: Problem, approach, key findings 2. Introduction: Motivation and problem statement 3. Related Work: Comprehensive literature review 4. Method: Detailed technical description 5. Experiments: Setup, results, analysis 6. Discussion: Limitations, implications, future work 7. Conclusion: Summary of contributions

Project Ideas

Fairness Projects

  • Intersectional Fairness: Methods for multi-attribute fairness
  • Dynamic Fairness: Fairness that adapts over time
  • Fairness in NLP: Bias detection in language models
  • Causal Fairness: Using causal inference for fair decisions

Robustness Projects

  • Universal Adversarial Perturbations: Domain-specific attacks
  • Certified Defense: Improving scalability of verification
  • Real-World Robustness: Robustness to natural distribution shifts
  • Robust Federated Learning: Security in distributed training

Interpretability Projects

  • Counterfactual Explanations: Generating actionable explanations
  • Explanation Evaluation: New metrics for explanation quality
  • Interactive Explanations: Human-in-the-loop explanation systems
  • Interpretable Deep Learning: Inherently interpretable architectures

Privacy Projects

  • Local Differential Privacy: Privacy without trusted curator
  • Private Representation Learning: Privacy-preserving embeddings
  • Membership Inference Defense: Protecting against privacy attacks
  • Federated Learning Privacy: Novel privacy-utility trade-offs

Evaluation Criteria

Technical Quality (40%)

  • Correctness: Sound methodology and implementation
  • Novelty: Original insights or approaches
  • Rigor: Thorough experimental validation
  • Reproducibility: Clear implementation details

Presentation (30%)

  • Clarity: Clear problem statement and solution
  • Organization: Logical flow and structure
  • Delivery: Effective oral presentation skills
  • Visual Design: Quality figures and slides

Writing (30%)

  • Technical Writing: Clear, precise exposition
  • Related Work: Comprehensive literature coverage
  • Analysis: Thoughtful discussion of results
  • Formatting: Professional presentation

Resources

Datasets

  • Fairness: Adult, COMPAS, CelebA, Folktables
  • Robustness: CIFAR-10/100, ImageNet, MNIST
  • Privacy: See federated learning benchmarks
  • General: Papers with Datasets collections

Computing Resources

  • Google Colab Pro: For small-scale experiments
  • Course Cluster: For larger computational needs
  • Cloud Credits: Limited AWS/Azure credits available

Collaboration Tools

  • GitHub: Version control and collaboration
  • Overleaf: Collaborative LaTeX writing
  • Slack: Team communication
  • Office Hours: Weekly project consultations

Timeline

Week 8-10: Team Formation & Topic Selection

  • Form teams and explore project ideas
  • Read relevant papers and identify gaps
  • Discuss ideas during office hours

Week 11: Proposal Submission

  • Submit 2-page project proposal
  • Receive feedback from instructors

Week 12-13: Implementation & Experiments

  • Implement core methodology
  • Run initial experiments
  • Submit progress report

Week 14: Final Push

  • Complete experiments and analysis
  • Prepare presentation slides
  • Draft final report

Week 15: Presentations

  • Present findings to class
  • Provide peer feedback
  • Finalize written report

Past Project Examples

Successful Projects (Previous Years)

  • "Fairness-Aware Multi-Task Learning" - Novel algorithm with theoretical analysis
  • "Robust Vision Transformers" - Comprehensive robustness evaluation
  • "Explaining Neural Recommendation Systems" - Human evaluation study
  • "Privacy-Preserving Graph Neural Networks" - DP mechanisms for graph data

For questions about projects, contact the teaching team during office hours or via email.