And deep minds breakthrough in audio generation. It is able to generate human-like speech.

https://deepmind.com/blog/wavenet-generative-model-raw-audio/

If you know any fellow data scientists and machine learning guys, please spread the word and forward this message to them. We all work hard on providing you greatest news, so we will appreciate your attention.

How to write your first spam-filter yourself

https://blog.cambridgecoding.com/2016/01/25/implementing-your-own-spam-filter/

Two papers about using video games to train computer vision models.

And who says that video games are useless? 😏

https://arxiv.org/abs/1608.01745
https://arxiv.org/pdf/1608.01745v2.pdf

Machine learning algorithms

Language Model on One Billion Word Benchmark


In this release, we open source a model trained on the One Billion Word Benchmark (http://arxiv.org/abs/1312.3005), a large language corpus in English which was released in 2013. This dataset contains about one billion words, and has a vocabulary size of about 800K words. It contains mostly news data. Since sentences in the training set are shuffled, models can ignore the context and focus on sentence level language modeling.

In the original release and subsequent work, people have used the same test set to train models on this dataset as a standard benchmark for language modeling. Recently, we wrote an article (http://arxiv.org/abs/1602.02410) describing a model hybrid between character CNN, a large and deep LSTM, and a specific Softmax architecture which allowed us to train the best model on this dataset thus far, almost halving the best perplexity previously obtained by others.

Link for the repo: https://github.com/tensorflow/models/tree/master/lm_1b

Keras implementation of deepmind's wavenet paper

https://github.com/usernaamee/keras-wavenet

Generative Visual Manipulation on the Natural Image Manifold

For more details, please visit the project webpage:
https://people.eecs.berkeley.edu/~junyanz/projects/gvm/

"Generative Visual Manipulation on the Natural Image Manifold", Jun-Yan Zhu, Philipp Krähenbühl, Eli Shechtman and Alexei A. Efros. In European Conference on Computer Vision (ECCV). 2016.

(via Deep Learning community on vk.com)

Convolutional Recurrent Neural Networks for Music Classification

https://keunwoochoi.wordpress.com/2016/09/15/paper-is-out-convolutional-recurrent-neural-networks-for-music-classification/

(again, via Deep Learning community on vk.com).

You can also use our @opendatasciencebot to submit an interesting link or get in contact with channel administration.

Fully-Convolutional Siamese Networks for Object Tracking

A new state-of-the-art for real-time tracking at 50-100 fps. It can be used to track objects in videos and stuff.

http://www.gitxiv.com/posts/TvEcWEJabGu7pEHEa/fully-convolutional-siamese-networks-for-object-tracking

Stanford University report on how life will be different with the AI by the 2030.
Spoiler: no skynet just yet.

https://ai100.stanford.edu/2016-report

There is an #opensource repository for automatic image captioning in #tensorflow
As article reports, researches have managed to significally improve quality of recognition.


https://research.googleblog.com/2016/09/show-and-tell-image-captioning-open.html

#deeplearning

European companies involved with machine learning.

https://medium.com/project-juno/european-machine-intelligence-landscape-43a22b44e961

Why top-down approach for machine learning is wrong (or right).


http://machinelearningmastery.com/deep-learning-for-developers/

Google released new ImageNet dataset, but for video.

YouTube8M is a large-scale labeled video dataset that consists of 8 million YouTube video IDs and associated labels from a diverse vocabulary of 4800 visual entities. It also comes with precomputed state-of-the-art vision features from billions of frames, which fit on a single hard disk. This makes it possible to train video models from hundreds of thousands of video hours in less than a day on 1 GPU! 

https://research.google.com/youtube8m/
http://arxiv.org/pdf/1609.08675v1.pdf

Top themes for machine learning application from Forbes

http://www.forbes.com/sites/bernardmarr/2016/09/30/what-are-the-top-10-use-cases-for-machine-learning-and-ai/#58cabbb010cf

New instances for GPU computing @ Amazon.

https://aws.amazon.com/blogs/aws/new-p2-instance-type-for-amazon-ec2-up-to-16-gpus/

Pointer Sentinel Mixture Models: use a pointer but back off to softmax vocab if uncertain
+ WikiText, new LM corpus.

Pointer sentinel-LSTM model achieves state of the art language modeling performance on the Penn Treebank (70.9 perplexity) while using far fewer parameters than a standard softmax LSTM. In order to evaluate how well language models can exploit longer contexts and deal with more realistic vocabularies and larger corpora we also introduce the freely available WikiText corpus.

https://arxiv.org/abs/1609.07843

Google released Open Images, a dataset consisting of ~9 million URLs to images that have been annotated with labels spanning over 6000 categories. They tried to make the dataset as practical as possible: the labels cover more real-life entities than the 1000 ImageNet classes, there are enough images to train a deep neural network from scratch and the images are listed as having a Creative Commons Attribution license.

https://research.googleblog.com/2016/09/introducing-open-images-dataset.html