Tutorial Schedule and Location

The ECML PKDD 2025 conference will be held in the city of Porto, Portugal from 15 to 19 Setember 2025. This specific tutorial will take about 4 hours, including a 30 min break. The first part of the tutorial will be a presentation of the library and its metrics. The remaining time will be dedicated to hands-on sessions focused on each data modality. A detailed schedule is provided bellow.

pyMDMA presentation (15 mins) – Luís Rosado

• Introduction
• Target modalities
• Metric taxonomy description
• Available metrics, installation, and contribution

Image Tutorial (60 mins) – Ivo Façoco

• Dataset presentation
  • Public RGB dataset
• Input Validation
  • Extraction of image quality metrics
  • Distribution analysis of extracted metrics
• Synthetic Validation
  • Synthetic dataset explanation (model used, number of instances and type of conditioning)
  • Feature extraction with pre-trained models
  • Evaluation of fidelity and diversity concepts
  • Sample selection through quality-based ranking. Comparison of best/worst generated examples via metric outputs.

Short Break (30 mins)

Time-Series Tutorial (60 mins) – Joana Rebelo

• Dataset Presentation
  • Overview of the ECG dataset and its characteristics
• Input Validation
  • Extraction of signal quality metrics
  • Distribution analysis of extracted metrics
• Synthetic Validation
  • Explanation of the synthetic dataset
  • Feature extraction using the Time Series Feature Extraction Library (TSFEL)
  • Evaluation of fidelity and diversity concepts
  • Selection of synthetic samples using metric outputs.

Tabular Tutorial (60 mins) – Pedro Matias

• Dataset Presentation
  • Dataset Loading
    • High-quality public tabular datasets
    • Low-quality public tabular datasets
  • Data Preparation
    • Attribute type detection, encoding, and scaling;
    • Visualization through 2D-embeddings.
• Input Validation
  • Extraction of tabular quality metrics;
  • Dataset selection through quality-based global ranking.
• Synthetic Validation
  • Synthetic Datasets
    • Description of generative models (traditional vs. deep learning);
    • Visualization of real vs. synthetic using 2D-embeddings.
  • Evaluation of fidelity and diversity concepts;