projects
minimum viable experiment series
![](./images/catch.gif)
Like single file experiments to trully grok things? I got you with
Reinforcement Learning and
Evolution Stratagies tutorials.
generative models and unsupervised learning for videos
![](./images/moving_mnist.gif)
I wrote
Recurrent Winner-Take-All and
Perception Updating Networks. Those methods show how to learn features by exploiting temporal context and to predict future frames in a video (think video compressor or robot planning).
![](https://github.com/Tetrachrome/subpixel/blob/master/images/superres_epoch6_detail.png?raw=true)
Enhance! A neural network for image superresolution. This project hit the front page of Hacker News and Trending on Github.
![](https://avatars2.githubusercontent.com/u/13408951?v=3&s=200)
I love bitcoin and blockchain! I wrote
Tierion's
pymerkletools for creating Merkle trees, generating merkle proofs, and verification of merkle proofs.
![](https://github.com/commaai/research/blob/master/images/drive_simulator.gif?raw=true)
I trained generative adversarial and recurrent neural networks to simulate highway videos.
![](./images/selfcoloring.png)
A segmentation network for self driving car data pipeline
![](./images/comma_ai_team.jpg)
CNN+RNN model for deciding where the car should go during Summer at comma.ai
![](./images/eeg.png)
I wrote the CSP pipeline for EEG analysis in Theano and fine-tuned everything end-to-end. I could've won BCI competition 2 T_T
![](./images/swiss_roll.png)
I want to use deep information theoretic learning to compress and make images look pretty.
![](http://rgbd-dataset.cs.washington.edu/imgs/rgbd.png)
I trained convnets to recognize objects in 3D when interning with Paracosm.io
My undergrad and master research was on blind source extraction. We had ECG recordings from a pregnant woman and we had to
separate the ECG signal of the fetus from the signal of the woman. I used kernels methods back then. You kids don't know
that anymore, but kernels used to be cool.
![](./images/STN2.png)
Keras' community is awesome! I contribute with both Tensorflow and Theano code to the source. Some advanced contributions
include Neural Turing Machines and
Spatial Transformer Networks.
![](https://github.com/EderSantana/hubot-recognizer/blob/master/images/pug_it_is.png?raw=true)
I put some deep learning on Slack
![](./images/dlwithpy.png)
video course about deep learning. Mostly using Keras and Theano.
papers, in the end of the day I write machine learning papers for a living