Three assignments ("Trabajos Prácticos") for the Inteligencia Artificial
course at Tecnológico de Costa Rica, done as a 3-person team (Alejandro
Cerdas, Kener Castillo, Pablo Pérez). Each folder is self-contained, with
its own notebooks, data, and a LaTeX report under documentation/ /
documentacion/.
- TP1 — Gradient Descent, RMSProp, and CMA-ES implemented from scratch in PyTorch, optimizing three 2D benchmark functions with Optuna-tuned hyperparameters.
- TP2 — Comparing TF-IDF, sentence embeddings, and prompted LLMs (Phi-4, SmolLM3) for classifying Spanish financial text as simple vs. complex, with Friedman/Wilcoxon statistical significance testing across the methods.
- TP3 — A multilayer perceptron built from scratch (backprop, momentum, Xavier init) benchmarked on XOR and synthetic data, then compared against transfer-learned AlexNet on a real glaucoma fundus-image dataset.
No unified dependency file exists across the three — each notebook installs or documents its own requirements; see each TP's README for specifics.