Here you will find codes from projects that I participated. These projects and much more are explained in my podcast. The podcast have weekly episodes all Thursdays and it's avaible in all the podcast platforms and I register it in English and in Portuguese, below you will find the link to spotify:
- Life with AI (In english)
- Vida com IA (Em portugues)
You can follow the podcast on Linkedin and on Instagram to be updated about the episode and also interact with me!
This project was done by (all contributed the same): Brayam Castillo, Filipe Lauar, Gabriel Baker and Vinícius Imaizumi
Project to use StyleGAN, InterfaceGAN, IDInvert and FaceAttributeFAN to manipulate images in StyleGAN's latent space.
You should run the code on google colab.
More details in the file GAN_report.pdf and you will find the explanation of the network achitecture and much more are in my podcast:
- What's behing instagram filters? Understanding the styleGAN. (In english)
- O que está por tras dos filtros do instagram? Explicando a styleGAN. (Em portugues)
You should run this code on google colab. You will just need to choose the files you want to upload and the results will be downloaded to your computer in a zip file.
The explanation of the network achitecture and other applications are in my podcast:
- Transforming normal pictures into famous paintings. Understanding the cycleGAN. (In english)
- Como transformar fotos normais em pinturas famosas. Explicando a cycleGAN. (Em portugues)
In this folder you will find 4 notebooks regarding the implementation and application of some dimensionality redcution and clustering methods.
You can run them on google colab following these link:
- Implementation of PCA, KPCA and ICA.
- Application of PCA, KPCA and ICA.
- Application of Kmeans.
- Implementation and application of forward selection.
The explanation of the problemns, the methods and some applications are in my podcast:
- Dimensionality Reduction and Clustering. Understanding PCA, Kmeans and Autoencoders. (In english)
- Redução de Dimensionalidade e Clustering. Explicando PCA, Kmeans e Autoencoders. (Em portugues)
In this colab notebook you will find my solution to win the image kaggle challange 2020/2021 at Telecom Paris.
To better understand the challange and the solution you can listen this episode in my podcast:
- How did I win a kaggle competition. (In english)
- Como eu ganhei um competição no Kaggle. (Em portugues)
In this folder you will find 10 (4 in portuguese and 6 in english) notebooks regarding the applications of CNNs in classification, detection and segmentation. In these notebooks you will find examples of implementations from scratch, model declaration and fine tuning in pytorch, tensorflow/keras and mxnet. I will keep including notebooks in this folder continuously. You should run these notebooks in google colab.
For more Deep Learning notebooks of CNN applications and much more in Portuguese I recommend this repository from UFMG professors.
The explanation of the classification, detection and segmentation problemns, their applications in the real world and the state of the art for each method are in the following episodes of my podcast:
- Artificial intelligence application is real world using CNNs. (In english)
- Aplicações de inteligência artificial para imagens no mundo real com CNNs. (Em portugues)
In this folder you will find 2 notebooks regarding the applications of GANs. One notebook was writen in portuguese and coded using pytorch to generate images of the MNIST. The other was writen in english and coded using tensorflow and keras to generate images of the MNIST and the CIFAR. I will keep including notebooks in this folder continuously. You should run these notebooks in google colab.
For more Deep Learning notebooks of CNN/GAN applications and much more in Portuguese I recommend this repository from UFMG professors.
Examples of many many applications in the real world using GANs can be founded in the following episodes of my podcast:
- Applications of GANs in the real world. (In english)
- Aplicações de GANs no mundo real. (Em portugues)
In this folder you will find some notebooks regarding the applications of RNNs in NLP and time series forecast. Some notebook were writen in portuguese, others in english. Some notebook were coded in pytorch and others in tensorflow. These informations are in the name of the notebook. This folder will be updated as I have new notebooks about the subject. I advise you to run them in google colab.
In this colab notebook you can see the implementation from scratch of the RNN, LSTM and GRU in pytorch and play with a sentence generator algorithm.
In this colab notebook you will find a code to perform translation from english to french (but you can also perform the opposite by a small change in the code).
For more Deep Learning notebooks of RNN applications and much more in Portuguese I recommend this repository from UFMG professors.
Examples of many many applications in the real world using RNNs can be founded in the following episodes of my podcast:
- Applications of RNNs in the real world. (In english)
- Aplicações de RNNs no mundo real. (Em portugues)
Explanation of RNN, GRU and LSTM algorithms:
- Algorithms for NLP and time series. Understanding RNN, GRU and LSTM. (In english)
- Algoritmos para NLP e time series. Explicando RNN, GRU e LSTM. (Em portugues)
In this colab notebook done by Breno you will find a notebook of a implementation of binary neural networks using tensorflow and larq. You will also find the comparision between the BNN and a normal/full precision neural network.
To better understand the pros and cons of Binary Neural Networks you can listen this episode in my podcast:
- Binary Neural Networks with Breno Skuk. (In english)
- Redes Neurais Binárias com Breno Skuk. (Em portugues)
In this colab notebook you will find some techiniques to perform categorical feature encoding (Label Encoder, One Hot Encoder, Target Encoder and Frequency Encoder) and numerical feature scaling/normalization.
To better understand these techiniques, why and how should you use them, you can listen this episode in my podcast:
- How to fed data into AI algorithms. Understanding categorical feature encoding and scaling/normalization. (In english)
- Como alimentar algorithmos de IA com dados tabulares. Explicando categorical feature encoding e normalização. (Em portugues)
Neural networks, from zero to hero. Tips for training neural networks, self-supervised learning and data augmnetation
In this repository you will find a tutorial about training of neural networks with many notebooks from the beginning untill advanced techniques like self-supervised learning and contrastive learning.
If you want to better understand it all, you can listen to this episode in my podcast:
- Tips for training neural networks, data augmentation and self-supervised learning with Fernando Santos. (In english)
- Dicas para treinar redes neurais, data augmnetation e self-supervised learning com Fernando Santos. (Em portugues)
In this colab notebook you will find a model to perform domain adaptation/semi-supervised learning on the MNIST dataset.
To better understand these techiniques and also the actual state of the art, you can listen this episode in my podcast:
- How to make autonomous vehicles drive in different places? Understanding domain adaptation and semi-supervised learning. (In english)
- Como treinar um carro autônomo na Alemanha e fazer ele dirigir no Brasil? Explicando domain adaptation e semi-supervised learning. (Em portugues)
In this colab notebook you will find some examples of how to fool a neural network using adversarial attacks and also how to understand the prediction of a neural network using saliancy maps.
To better understand how to attack/fool a neural network and also how to defend your network from these attacks, you can listen this episode in my podcast:
- How to fool a neural network. Understanding adversarial attacks. (In english)
- Como enganar um carro autônomo. Explicando adversarial attacks. (Em portugues)
In this folder you will find the implementation of many algorithms for Reinforcement Learning. These implementations were part of the RLD course in Sorbonne University. I would like to thank Luc and Dao that implemented most of these algorithms (the RLD monsters). As it was all implemented in the course, in each folder I puted the project text so you guys can reproduce the experiments that we did. There is in the end a report of everything that was done but it's in French (sorry).
To better understand the idea of Reinforcement Learning, these and other algorithms and some application, you should listen to this episode in my podcast:
- Reinforcement Learning. (In english)
- Aprendizado por Reforço. Entendendo Reinforcement Learning. (Em portugues)
In this colab notebook you will find an implementation of positional encoding and self-attention mechanism of the Transformer network.
To better understand how the whole Transformer architecture work you can listen this episode in my podcast:
- Attention is all you need. Understanding Transformers. (In english)
- Attention is all you need. Entendendo Transformers. (Em portugues)
In this colab notebook you will find an implementation of SRR for both contrastive and rotation losses and using as One Class Classifier OneClassSVM and also Gaussian Mixture + mahalanobis distance. You can apply it in 3 different datasets.