Profile cover photo
Profile photo
Fábio Uechi
204 followers
204 followers
About
Fábio's posts

Post has shared content

Post has shared content
This might be relevant to people who are interested in video prediction. We present a deep convolutional architecture that makes future pixel-level prediction conditioned on the current action. The model can make a good long-term prediction (up to several hundred frames) on Atari game videos. We also show that this prediction capability can be useful for improving reinforcement learning tasks through better informed exploration.

Action-Conditional Video Prediction using Deep Networks in Atari Games
http://arxiv.org/abs/1507.08750

Video:
https://youtu.be/4e-PqfpS8_4

More videos are available at:
https://sites.google.com/a/umich.edu/junhyuk-oh/action-conditional-video-prediction

Post has shared content
Recreating photos in the style of famous artists

In another example of how Deep Neural Networks (DNNs) can be used to create interesting works of art, a group of German researchers have released the paper, A Neural Algorithm of Artistic Style (http://goo.gl/HsFR0C), that introduces a system that renders an input photo in the artistic style of a given piece of art while preserving its overall content.

In their paper, the authors describe how the representations of content and style in their Convolutional Neural Network (CNN) are separable, making it possible to use neural representations to separate and recombine content and style of arbitrary input images. To learn more, read the full paper. For access to their algorithm, visit http://goo.gl/jB7ovU
Photo

Post has shared content
Deep learning has shown remarkable success on some of the world’s most difficult computer science challenges. Today on the Google Research blog we go into how Long Short-term Memory Recurrent Neural Networks (LSTM RNNs) are being used to improve Google Voice transcription. 

Post has shared content
Generate your own Neural Network inspired images with DeepDream

Two weeks ago we blogged about a visualization tool designed to help us understand how neural networks work and what each layer has learned (http://goo.gl/pUfbyH). In addition to gaining some insight on how these networks carry out classification tasks, we found that this process also generated some beautiful art.

Now you can make your own images using an open source IPython notebook, which allows you to choose which layers in the network to enhance, how many iterations to apply and how far to zoom in. Alternatively, different pre-trained networks can be plugged in.

It'll be interesting to see what imagery people are able to generate. If you post images to Google+, Facebook, or Twitter, be sure to tag them with #deepdream so other researchers can check them out too.

Post has attachment
Animated Photo

Post has shared content
These are awesome posters!

For me, the best episode has to be Hardhome. Which was your favorite episode from this season?

Source: http://robotssuicidas.tumblr.com/post/121602429816/game-of-thrones-season-5

#GameOfThrones #ASOIAF #GeorgeRRMartin #GoTSeasonFive #YouWinOrYouDie #WinterIsComing
Animated Photo
Animated Photo
Animated Photo
Animated Photo
Animated Photo
15/06/2015
10 Photos - View album

Post has shared content
A single second can have a big impact on systems that rely on careful sequencing.

On June 30, the world will experience its 26th recorded leap second. If you use Google Compute Engine, it’s important to know how this extra second can affect you. Find out how to prepare: http://goo.gl/LwtzAe
Photo

Post has shared content
Throw the drone in the air like you just don't care.

Post has shared content
Wait while more posts are being loaded