Coursework from TC2035 - Redes Neuronales (Tecnológico de Monterrey): a progression from a single perceptron to modern deep architectures (CNNs and a Vision Transformer), plus competitive / unsupervised learning.
| Folder | Contents |
|---|---|
01-fundamentals/ |
Single perceptron, a perceptron cat-classifier, and a simple neural network (MLP) built from the ground up. |
02-supervised-learning/ |
Supervised classification and regression with neural networks - training loops and evaluation. |
03-competitive-learning/ |
Self-Organizing Maps (SOM) and Learning Vector Quantization (LVQ). |
04-convolutional-networks/ |
CNN fundamentals and a LeNet-5 challenge on Fashion-MNIST.* |
05-vision-transformer/ |
A Vision Transformer (ViT) on Fashion-MNIST.* |
runs/ |
Trained-model artifacts for the Fashion-MNIST CNN (see below). |
* Course challenge (reto) built as a team: Diego Arias, Jose Rojas, Josh Haberkern, Steffany Lara.
The runs/ folder ships a trained model and its full evaluation:
- Test accuracy: 87.1% · Top-3 accuracy: 98.6%
best_model.h5(trained weights),confusion_matrix.png,training_curves.png, per-class metrics inevaluation_report.json, andtraining_log.csv.
Python · TensorFlow / Keras · NumPy · pandas · scikit-learn · Matplotlib · seaborn
pip install -r requirements.txtDatasets (e.g. Fashion-MNIST) are loaded via framework utilities and are not stored in this repo.
Steffany Lara - Data Science & Mathematics Engineering, Tecnológico de Monterrey. Team challenges co-authored as noted above.