«A Practical Introduction to ML, DL, TL, and RL» seminar
📅 May 06th, 2026
📍 Online (Zoom)
Our recent seminar on AI model selection for doctoral researchers has concluded, marking another successful step in our ongoing series dedicated to digitalization. We extend our sincere gratitude to the attendees who joined us. The session was designed not merely to define terms, but to address a critical challenge in modern research: the tendency to over-engineer solutions by reaching for complex algorithms before fully understanding the problem at hand.
The core of the discussion centered on a strategic roadmap through the four pillars of modern AI: Machine Learning, Deep Learning, Transfer Learning, and Reinforcement Learning. Rather than treating these as isolated buzzwords, we examined the mechanics of how they function and, more importantly, how to deploy them efficiently. Participants explored the taxonomy of intelligence, learning to distinguish clearly between when a classical ML approach is sufficient and when the computational cost of Deep Learning is truly justified. We also delved into the efficiency of Transfer Learning, demonstrating how researchers can leverage pre-trained models to achieve state-of-the-art results even when working with limited datasets.
Beyond the technical architectures, the seminar placed a significant emphasis on the «Researcher’s Toolkit» and the philosophy behind effective model building. We reviewed essential libraries such as PyTorch, TensorFlow, and Scikit-Learn, not just as code repositories, but as bridges between theoretical concepts and practical execution. A recurring theme was the «start small» philosophy, encouraging researchers to validate simple baselines before iterating toward complexity. This approach, combined with rigorous model selection and validation strategies, serves as a safeguard against overfitting and ensures greater reproducibility in experimental results.
Finally, we addressed the broader responsibilities inherent in developing intelligent systems. The conversation touched upon computational ethics, specifically the importance of addressing bias and ensuring interpretability. The consensus was clear: a model is only as valuable as the strategy behind it, and a «smart» model must also be a fair and transparent one. The ultimate goal of the session was to equip every participant with a mental framework that prioritizes building effective solutions over simply building models.
For those who attended, we hope these insights provide a lasting foundation for your future projects. We thank everyone who joined and contributed to the event. Special thanks to Joaquín Ordieres-Meré for the presentation, and to Universidad Politécnica de Madrid and Universidad de Sevilla for their support in organizing this session as part of our ongoing seminar series.
At DIGEST, we are committed to applied training that complements traditional doctoral programs with advanced skills in digital research and industrial optimization.