Session 4: Synthesis, Q&A, and Pathway to Continued Learning

From Techniques to Operational EO AI in the Philippines

Stylianos Kotsopoulos

EU-Philippines CoPhil Programme

Session Plan

• Duration: 2 hours (120 minutes)
• Plan:
  • 0–15: 4-day journey recap
  • 15–45: Technique selection matrix
  • 45–75: Best practices for deployment
  • 75–100: CoPhil Campus & community
  • 100–115: Action plans (3-month roadmap)
  • 115–120: Closing & feedback

What You Can Now Do

• Build LSTMs for EO time series (Day 4)
• Train RF, CNNs, U-Net for EO tasks (Days 2–3)
• Use FM/SSL/XAI to scale and explain models (Day 4)
• Design operational pipelines and monitoring (Day 4)

Recap of Days 1–4

Highlights

• Day 1: Python, GEE, data preparation
• Day 2: RF, CNNs (foundations)
• Day 3: U-Net segmentation, object detection
• Day 4: LSTM time series, FM/SSL/XAI

Technique Selection Matrix

• Classification (image/patch) → CNN/Transfer Learning
• Segmentation (pixel-wise) → U-Net / FM fine-tuning
• Detection (bounding boxes) → YOLO/SSD/DETR
• Time series → LSTM/Transformers, temporal validation
• Tabular + EO features → Random Forest/XGBoost

Best Practices

Data-Centric AI

• Fix labels, balance classes, represent regions/seasons

Validation

• Temporal splits, spatial CV, stratified sampling

Monitoring

• Concept/data drift, retraining schedule, dashboards

Deployment

• APIs, performance budgets, documentation, XAI

CoPhil Campus & Community

Stay Connected

• Digital Space Campus (materials, notebooks, recordings)
• SkAI-Pinas: DIMER (models), AIPI (no/low-code)
• PhilSA Space+, NAMRIA Geoportal, PAGASA data
• Monthly meetups, annual EO AI summit, mentorship

3-Month Action Plan

• Week 1: Pick a starter project, run baseline
• Month 1: Proof of concept with validation
• Month 2: Improve data + model, stakeholder review
• Month 3: Pilot deployment + monitoring

Closing & Feedback

• Feedback form and office hours
• Certificates on completion
• Thank you and see you in the community!