Cellular Automaton
In this Python file, I add a cellular automaton to a PyTorch kernel, add a residual layer to achieve 99.29% accuracy on test set in MNIST, results better than Robust Training in High Dimensions via Block Coordinate Geometric Median Descent", by Google AI and Amazon Search (2021).
TF-Keras
This folder has the code to customize ResNet Architecture via dictionary config, changing first two layers to receive multispectral images with 9 channels. Final layers are also added, for training and inference.
Santa Fe
This folder has the python code to create Agent-Based Models based in 2 and 5 state cellular automata.
Tensorflow Hub
This folder has the code to generate word embeddings using BERT multilingual model from Tensorflow Hub, in the shape (2,768).
Wav2Vec
Here you can find Python code to finetune Wav2Vec model (300 MB) of Speech Recognition on Common Voice dataset, as well as the code for evaluating the model.
Wav2Vec2-Large-xlsr
These files allow the training of Facebook's Wav2Vec2-Large-xlsr (model 1.5 GB) on Common Voice dataset on a RTX 2060. Some layers are frozen to allow fit in the GPU. Paper available at: https://arxiv.org/abs/2006.11477. A pretrained version of the model is at my Hugging Face repository - Rubens Zimbres:
https://huggingface.co/Rubens/Wav2Vec2-Large-XLSR-53-a-Portuguese