Take your AI to the next level by creating a more sophisticated model that can distinguish between 7-10 different categories instead of just 3-4. This challenge teaches you about the complexities of handling larger datasets and more nuanced classifications.
Learning Objectives
- Design and implement a complex multi-class classification system with 7+ categories
- Understand the relationship between dataset size and model performance
- Develop strategies for maintaining high accuracy with increased complexity
- Learn to identify and address category confusion in complex models
Challenge Steps
-
Choose Your Domain:
Select a topic area with many related but distinct categories. Examples: 10 different hand gestures, 8 types of weather conditions, 7 species of local birds, or 9 common household objects.
-
Plan Your Categories:
Carefully select categories that are distinct enough to classify but related enough to be interesting. Consider which categories might be confused and plan how to address this.
-
Collect Comprehensive Training Data:
Gather 75-100 samples per category (750-1000 total samples). Ensure maximum variety in lighting, angles, backgrounds, and conditions.
-
Train and Evaluate:
Train your model and test it thoroughly. Calculate accuracy for each individual category and identify problem areas.
-
Analyze Confusion Patterns:
Create a confusion matrix showing which categories your AI confuses most frequently. Develop hypotheses about why certain confusions occur.
-
Iterate and Improve:
Add targeted training data to address specific confusions. Experiment with different approaches to problematic categories.
-
Document Your Process:
Write a detailed report explaining your design decisions, challenges faced, and solutions implemented. Include your confusion matrix and accuracy statistics.
Pro Tips
Start with 5 categories and add more gradually. This helps you identify at what point complexity becomes problematic. Pay special attention to categories that are visually or aurally similar—these will be your biggest challenges.
Success Criteria
- Model successfully classifies at least 7 distinct categories
- Overall accuracy reaches 75% or higher across all categories
- Confusion matrix clearly shows which categories are most problematic
- Documentation explains specific strategies used to improve performance
- Demonstration shows the model working in real-time with various inputs
Identify a genuine problem in your school, community, or home that could be solved with AI. Conduct user research, develop requirements, and create a polished application specifically designed to address this real need. This activity mirrors the professional AI development process.
Learning Objectives
- Conduct user research to identify genuine needs and requirements
- Apply design thinking methodology to AI application development
- Create a complete, polished application with user interface and instructions
- Test with real users and iterate based on feedback
- Present a professional solution to stakeholders
Development Process
-
Problem Discovery:
Interview 5-10 potential users (classmates, teachers, family members, community members) to identify problems they face. Look for problems that involve categorization, recognition, or pattern detection.
-
Requirements Gathering:
For your chosen problem, document: Who are the users? What specific task needs to be accomplished? What counts as success? What constraints exist (speed, accuracy, ease of use)?
-
Solution Design:
Sketch your application's user interface. Plan the user flow from start to finish. Decide what information users need at each step. Design error messages and help text.
-
Development:
Build your AI model with real-world testing in mind. Create a polished interface with clear instructions. Add features like confidence thresholds, reset buttons, and result history.
-
User Testing:
Have at least 5 people from your target audience test your application. Observe how they use it without providing help. Collect feedback on usability, accuracy, and value.
-
Iteration:
Based on user feedback, make improvements to your model, interface, or instructions. Test again to verify improvements work as intended.
-
Deployment Planning:
Write a deployment plan: How would users access your application? What support would they need? What maintenance would be required? What would success look like at scale?
Important Considerations
Privacy and Ethics: If your application collects or uses personal data, you must have a plan for protecting user privacy. Consider what data you truly need and how you'll keep it secure. Discuss your plans with your teacher before collecting any personal information.
Project Deliverables
- User research summary with interview notes and problem statement
- Requirements document with user stories and success criteria
- Functional AI application with polished user interface
- User testing report with feedback summary and improvements made
- Deployment plan addressing real-world implementation
- Final presentation demonstrating the application and explaining its impact
Showcase Opportunity
Real-world applications developed through this activity may be eligible for submission to science fairs, technology competitions, or community showcase events. Discuss opportunities with your teacher.
Analyze your AI project through an ethical lens by creating a comprehensive impact statement. This activity develops critical thinking about the societal implications of AI technology and prepares you to be a responsible AI developer.
Learning Objectives
- Evaluate AI systems for potential biases and unfair outcomes
- Identify stakeholders affected by AI deployment and their interests
- Analyze both benefits and potential harms of AI applications
- Develop mitigation strategies for identified risks
- Communicate ethical considerations clearly to diverse audiences
Analysis Framework
-
Bias Analysis:
Examine your training data for potential biases. Are all relevant groups represented? Are some categories over or under-represented? Could your AI perform differently for different demographic groups?
-
Stakeholder Mapping:
Identify everyone who could be affected by your AI: direct users, indirect users, people who might be harmed, organizations, communities. For each stakeholder, note their interests and concerns.
-
Benefit Analysis:
Document the positive impacts of your AI. Who benefits and how? What problems does it solve? What efficiencies does it create? Quantify benefits where possible.
-
Risk Assessment:
Identify potential harms or misuses. What if your AI makes mistakes? What if it's used in ways you didn't intend? Who could be disadvantaged? Consider privacy, fairness, safety, and autonomy.
-
Fairness Evaluation:
Does your AI treat all users equally? Could it discriminate against certain groups? Are there accessibility barriers? How do you define fairness for your application?
-
Privacy Assessment:
What data does your AI collect or use? How is it stored? Who has access? What are the privacy implications? What consent is required?
-
Mitigation Strategies:
For each identified risk, propose specific actions to reduce or eliminate the harm. These might include improving training data, adding safeguards, providing transparency, or limiting use cases.
-
Ongoing Monitoring:
Describe how you would monitor your AI in real-world use to detect problems early. What metrics would you track? How often would you review performance?
Ethical Frameworks to Consider
Research these ethical principles and apply them to your analysis: Beneficence (doing good), Non-maleficence (avoiding harm), Autonomy (respecting user choice), Justice (fairness and equity), and Transparency (explainability and openness).
Impact Statement Components
| Section |
Required Content |
Length |
| Executive Summary |
Brief overview of your AI and its ethical considerations |
1 paragraph |
| Project Description |
What your AI does, how it works, intended use cases |
2-3 paragraphs |
| Stakeholder Analysis |
Who is affected and how, with specific examples |
1-2 pages |
| Bias and Fairness |
Analysis of potential biases with supporting evidence |
1 page |
| Benefits and Risks |
Balanced analysis of positive and negative impacts |
1-2 pages |
| Privacy Assessment |
Data practices and privacy implications |
1/2-1 page |
| Mitigation Plan |
Specific actions to address identified risks |
1 page |
| Monitoring Strategy |
How to track performance and detect problems |
1/2 page |
| Conclusion |
Summary of ethical posture and recommendations |
1 paragraph |
Reflection Questions
- What ethical issues did you discover that you hadn't considered initially?
- How did analyzing ethics change your perspective on your AI project?
- What trade-offs exist between functionality and ethical considerations?
- If you were to rebuild this project, what would you do differently from an ethics perspective?
- What responsibilities do AI developers have to society?
Explore the concept of transfer learning by using pre-trained AI models as a foundation for your own specialized application. This advanced activity introduces you to how professional AI developers leverage existing models to build new applications more efficiently.
Learning Objectives
- Understand the concept of transfer learning and its advantages
- Compare training from scratch versus using pre-trained models
- Implement transfer learning using accessible platforms
- Analyze the efficiency gains and limitations of transfer learning
- Make informed decisions about when to use transfer learning
Exploration Activities
-
Concept Research:
Research transfer learning: What is it? How does it work? Why is it useful? What are its limitations? Write a 1-page summary in your own words with examples.
-
Platform Investigation:
Explore platforms that support transfer learning: Teachable Machine (which uses MobileNet), TensorFlow.js with pre-trained models, or ML5.js. Choose one platform for your experiments.
-
Baseline Project:
Create a simple 3-class image classifier training from scratch. Record: training time, number of samples needed, final accuracy, testing performance.
-
Transfer Learning Project:
Create an equivalent 3-class classifier using transfer learning with a pre-trained model. Record the same metrics. Use the same test images for fair comparison.
-
Comparative Analysis:
Create charts or tables comparing: samples required, training time, accuracy, robustness to new inputs. Calculate percentage improvements where applicable.
-
Limitation Testing:
Test the limitations of your transfer learning model: What happens with completely unfamiliar objects? How well does it generalize? Where does it struggle?
-
Specialized Application:
Build a more complex application (5-7 classes) using transfer learning. Choose a specialized domain (medical imaging, rare animals, industrial parts) where training from scratch would be impractical.
-
Documentation:
Write a technical report explaining transfer learning, your methodology, results, and conclusions. Include recommendations for when to use transfer learning.
Technical Resources
Teachable Machine: Automatically uses MobileNet for transfer learning. Compare "Standard" vs "Fast" training modes.
ML5.js: JavaScript library with pre-trained models. Try the Image Classifier or Feature Extractor examples.
TensorFlow.js: Access pre-trained models like MobileNet, ResNet, or EfficientNet for advanced projects.
Experimental Questions to Answer
- How many training samples were needed with vs. without transfer learning?
- What was the training time difference between approaches?
- Did transfer learning produce higher accuracy? By how much?
- Which approach generalized better to new, unseen examples?
- What types of classification tasks benefit most from transfer learning?
- When would training from scratch be preferable to transfer learning?
- How does the choice of base model affect performance?
Project Deliverables
- Concept research paper explaining transfer learning (1-2 pages)
- Two working AI models: one from scratch, one using transfer learning
- Comparative data analysis with charts and statistics
- Specialized application demonstrating transfer learning advantages
- Technical report with methodology, results, and recommendations (3-5 pages)
- Presentation explaining your findings to classmates
Connect your AI skills to another subject area by creating an AI application that solves a problem or enhances learning in science, mathematics, language arts, social studies, or the arts. This interdisciplinary project demonstrates how AI can be applied across all fields of study.
Project Ideas by Subject
🔬 Science
Plant species classifier, rock and mineral identifier, animal tracking system, chemical compound recognizer, weather pattern analyzer, or microscope image classifier
📐 Mathematics
Handwritten equation solver, geometric shape classifier, graph type identifier, statistical distribution recognizer, or math symbol translator
📚 Language Arts
Story genre classifier, author style identifier, poetry form recognizer, literary device detector, or writing quality analyzer
🌍 Social Studies
Historical artifact classifier, architectural style identifier, cultural symbol recognizer, map feature detector, or historical period determiner
🎨 Arts
Art style classifier, musical genre identifier, color palette analyzer, artistic technique recognizer, or composition evaluator
💪 Physical Education
Exercise form checker, sports move classifier, fitness activity tracker, proper technique evaluator, or athletic skill analyzer
Development Process
-
Subject Area Consultation:
Meet with a teacher from your chosen subject area. Discuss challenges students face, concepts that are difficult to visualize or understand, or tasks that could be automated. Get feedback on your project idea.
-
Educational Research:
Research the specific content area your AI will address. Become knowledgeable enough to ensure your AI provides accurate, educational value. Cite sources in your documentation.
-
Pedagogical Design:
Design your AI not just to classify, but to teach. Include explanations, educational content, fun facts, or interactive quizzes related to classifications.
-
Accuracy Standards:
For educational applications, accuracy is critical. Aim for 85%+ accuracy. Include confidence thresholds and "I'm not sure" responses when confidence is low.
-
User Testing with Target Audience:
Test your educational AI with students from the target grade level. Do they learn from it? Is it engaging? Is it accurate? Collect both quantitative and qualitative feedback.
-
Teacher Review:
Have subject-area teachers review your AI for content accuracy and pedagogical effectiveness. Incorporate their feedback before finalizing.
-
Lesson Plan Creation:
Create a mini-lesson plan showing how a teacher could integrate your AI into their curriculum. Include learning objectives, activity instructions, and assessment ideas.
Excellence Indicators
Outstanding cross-curricular projects demonstrate deep understanding of both AI technology and the subject matter, provide genuine educational value beyond novelty, include rich educational content alongside classification, and receive positive feedback from subject-area teachers.
Project Requirements
- Functional AI application addressing a specific educational need
- 85%+ accuracy with robust testing across diverse inputs
- Educational content integrated into the application interface
- Documentation citing research sources and explaining subject-area concepts
- User testing report with student and teacher feedback
- Mini-lesson plan for classroom integration
- Presentation connecting AI technology to subject-area learning
Online Learning Platforms
- Google AI Education: Free resources including Machine Learning Crash Course and AI principles guides
- Elements of AI: Free online course covering AI basics and ethics (elementsofai.com)
- MIT App Inventor Tutorials: Step-by-step guides for building AI-powered mobile apps
- Kaggle Learn: Free micro-courses on machine learning and AI topics
- Fast.ai: Practical deep learning courses designed for beginners
Video Resources
- Crash Course AI: YouTube series explaining AI concepts with excellent visualizations
- Two Minute Papers: Brief, accessible explanations of cutting-edge AI research
- The Coding Train: Creative coding tutorials including machine learning with ml5.js
- 3Blue1Brown Neural Networks: Deep mathematical explanations with beautiful animations
Competitions and Challenges
- AI4ALL Project Showcase: Annual showcase for K-12 AI projects
- Congressional App Challenge: National competition for student-created apps (including AI)
- Technovation: Global tech entrepreneurship program with AI category
- Science Fair Projects: Many science fairs now welcome AI and machine learning projects