Jelly Mario

Imaginary Soundscape

Online project from Qosmo generates ambient sounds to Google Streetview panoramas through Deep Learning processes, interpreting the visuals for appropriate sounds:

“Imaginary Soundscape” is a web-based sound installation, in which viewers can freely walk around Google Street View and immerse themselves into imaginary soundscape generated with deep learning models.

… Once trained, the rest was straightforward. For a given image from Google Street View, we can find the best-matched sound file from a pre-collected sound dataset, such that the output of SoundNet with the sound input is the most similar to the output of the CNN model for the image. As the sound dataset, we collected 15000 sound files from internet published under Creative Commons license and filtered with another CNN model on spectrogram trained to distinguish environmental/ambient sound from other types of audio (music, speech, etc.). 

You can try it out for yourself here, and find more background information here

Hackaday Prize Best Product Finalist: Reconfigurable Robots http://ift.tt/2uB4Acd

Introducing SAMOA, an open source platform for mining big data streams.

https://github.com/yahoo/samoa

Machine learning and data mining are well established techniques in the world of IT and especially among web companies and startups. Spam detection, personalization and recommendations are just a few of the applications made possible by mining the huge quantity of data available nowadays. However, “big data” is not only about Volume, but also about Velocity (and Variety, 3V of big data).

The usual pipeline for modeling data (what “data scientists” do) involves taking a sample from production data, cleaning and preprocessing it to make it usable, training a model for the task at hand and finally deploying it to production. The final output of this process is a pipeline that needs to run periodically (and be maintained) in order to keep the model up to date. Hadoop and its ecosystem (e.g., Mahout) have proven to be an extremely successful platform to support this process at web scale.

However, no solution is perfect and big data is “data whose characteristics forces us to look beyond the traditional methods that are prevalent at the time”. The current challenge is to move towards analyzing data as soon as it arrives into the system, nearly in real-time.

For example, models for mail spam detection get outdated with time and need to be retrained with new data. New data (i.e., spam reports) comes in continuously and the model starts being outdated the moment it is deployed: all the new data is sitting without creating any value until the next model update. On the contrary, incorporating new data as soon as it arrives is what the “Velocity” in big data is about. In this case, Hadoop is not the ideal tool to cope with streams of fast changing data.

Distributed stream processing engines are emerging as the platform of choice to handle this use case. Examples of these platforms are Storm, S4, and recently Samza. These platforms join the scalability of distributed processing with the fast response of stream processing. Yahoo has already adopted Storm as a key technology for low-latency big data processing.

Alas, currently there is no common solution for mining big data streams, that is, for doing machine learning on streams on a distributed environment.

Enter SAMOA

SAMOA (Scalable Advanced Massive Online Analysis) is a framework for mining big data streams. As most of the big data ecosystem, it is written in Java. It features a pluggable architecture that allows it to run on several distributed stream processing engines such as Storm and S4. SAMOA includes distributed algorithms for the most common machine learning tasks such as classification and clustering. For a simple analogy, you can think of SAMOA as Mahout for streaming.

SAMOA is both a platform and a library. As a platform, it allows the algorithm developer to abstract from the underlying execution engine, and therefore reuse their code to run on different engines. It also allows to easily write plug-in modules to port SAMOA to different execution engines.

As a library, SAMOA contains state-of-the-art implementations of algorithms for distributed machine learning on streams. The first alpha release allows classification and clustering.

For classification, we implemented a Vertical Hoeffding Tree (VHT), a distributed streaming version of decision trees tailored for sparse data (e.g., text). For clustering, we included a distributed algorithm based on CluStream. The library also includes meta-algorithms such as bagging.

HOW DOES IT WORK?

An algorithm in SAMOA is represented by a series of nodes communicating via messages along streams that connect pairs of nodes (a graph). Borrowing the terminology from Storm, this is called a Topology. Each node in the Topology is a Processor that sends messages to a Stream. The user code that implements the algorithm resides inside a Processor. Figure 3 shows an example of a Processor joining two stream from two source Processors. Here is a code snippet to build such a topology in SAMOA.

TopologyBuilder builder; Processor sourceOne = new SourceProcessor(); builder.addProcessor(sourceOne); Stream streamOne = builder.createStream(sourceOne); Processor sourceTwo = new SourceProcessor(); builder.addProcessor(sourceTwo); Stream streamTwo = builder.createStream(sourceTwo); Processor join = new JoinProcessor(); builder.addProcessor(join).connectInputShuffle(streamOne).connectInputKey(streamTwo);

SWEET! HOW DO I GET STARTED?

1. Download SAMOA

git clone git@github.com:yahoo/samoa.git cd samoa mvn -Pstorm package

2. Download the Forest CoverType dataset.

wget "http://downloads.sourceforge.net/project/moa-datastream/Datasets/Classification/covtypeNorm.arff.zip" unzip covtypeNorm.arff.zip

Forest CoverType contains the forest cover type for 30 x 30 meter cells obtained from US Forest Service (USFS) Region 2 Resource Information System (RIS) data. It contains 581,012 instances and 54 attributes, and it has been used in several papers on data stream classification.

3. Download a simple logging library.

wget "http://repo1.maven.org/maven2/org/slf4j/slf4j-simple/1.7.2/slf4j-simple-1.7.2.jar"

4. Run an Example. Classifying the CoverType dataset with the VerticalHoeffdingTree in local mode.

java -cp slf4j-simple-1.7.2.jar:target/SAMOA-Storm-0.0.1.jar com.yahoo.labs.samoa.DoTask "PrequentialEvaluation -l classifiers.trees.VerticalHoeffdingTree -s (ArffFileStream -f covtypeNorm.arff) -f 100000"

The output will be a sequence of the evaluation metrics for accuracy, taken every 100,000 instances.

To run the example on Storm, please refer to the instructions on the wiki.

I WANT TO KNOW MORE!

For more information about SAMOA, see the README and the wiki on github, or post a question on the mailing list.

SAMOA is licensed under an Apache Software License v2.0. You are welcome to contribute to the project! SAMOA accepts contributions under an Apache style contributor license agreement.

Good luck! We hope you find SAMOA useful. We will continue developing the framework by adding new algorithms and platforms.

Gianmarco De Francisci Morales (gdfm@yahoo-inc.com) and Albert Bifet (abifet@yahoo.com) @ Yahoo Labs Barcelona

Harry Potter and the Neural Network fan fiction

Or, what happens if you train a neural network on the titles and plot summaries of over 100,000 works of Harry Potter fan fiction.

In the decades since the Harry Potter books were published, fans have written literally hundreds of thousands of Harry Potter stories of their own, and shared them online. Can a neural network join in on the fun?

In a way, everything a recurrent neural network writes is fan fiction. A recurrent neural network looks at an example dataset (such as the complete Sherlock Holmes stories) and teaches itself the patterns and conventions that it sees. So, if it’s given Sherlock Holmes stories, it will become obsessed with Holmes and Watson, and if it’s given knock-knock jokes, it will spend all day telling awful knock-knock jokes of its own.

Thanks to an idea by a couple of readers, some heroic work by @b8horpet in scraping (with permission) hundreds of thousands of Harry Potter fan fiction titles and summaries from AO3, and a flexible new recurrent neural network implementation by Chen Liang, the neural network’s latest obsession is Harry Potter.

The Perfect Party by iamisaac Draco has been left alone, and Ginny confused must learn and who has his best friend. They were breathed by a love that didn’t become his grounds and the flowers begin.

This is a typical example of the neural network’s fan fiction - romantic pairings of two or more Harry Potter characters (called “ships” in fan fiction-speak). In this case, it even has chosen a plausible author: iamisaac is a real and fairly prolific fan fiction author whose works do tend to be of the “romantic” variety. 

The Garden by perverse_idyll for lexigilite Ron and Hermione move after a man party. What did her best things go and has to deal with people she loves? How many imperfect love really belonges them and needs to be a person? Or will they learn and more than the war? Mirror Thing by Queen_Elexhan “Are you there for a relationship? I was a sad future for your love.”  Harry and Ginny find out the meaning is.

Shatters by Kis [archived by TheHexFiles_archivist ] Based on the Spot Are It Falls Into A Heir by NextrangeOnTheThree Draco and Hermione share a whole indescribbening.

Again, “perverse_idyll” and “TheHexFiles_archivist” are fairly active authors. (Hi, if you’re reading! The neural network seems to like your writing, and is writing fan fiction of your fan fiction!) Those familiar with Harry Potter fan fiction will not be surprised to learn that the neural network really likes to generate ships; pretty much every combination of characters is represented (some of the more unusual combinations being “The Snow/Voldemort”, “The Ministry/Draco Malfoy”, and “Voldemort/Random Quidditch Child”).

By turning down the neural network’s creativity setting to near-zero, we arrive at its vision of what the quintessential Harry Potter fan fiction would be like - and we also learn its favorite ship:

Persuading by theladyblack Harry and Draco are still a second chance at the end of the war.  Will they be able to do with the fairy tale of the first time they were a strange stranger to the street of the war and the war is over?

It turns out the neural network is obsessed with Harry/Draco, although in a pinch, Sirius/Remus will also do.

The neural network also seems to really like stories about Professor Snape trying to do rather ordinary things:

New Moon Boys by Dungoonke for Loki_Kukaka Severus Snape comes back to a night’s politics.

In the Reason Is Blinders by LittleRoma Severus has been through his lost remote.

In The Alteri Silence by Forest_of_Holly for roscreens41 Snape receives life after plants to do by work over whether they get into. Just Hell.

A Second Chance by DarkCorgi Snape had a second thing, and that is better than anything for for the rest of his life.

Mirror by orphan_account Severus Snape tries to get a lot of dragons and that was to be more than he didn’t expect to continue. He has always been a bit of an old and a baby to stay the way he’d been the brother at Hogwarts and he keeps the chance of meeting… Deception by FlyingEyes Snape is a British Robes of interesting things and worrys like a little fun and sees the pretty battle for a while.

Another thing that happened, which is pretty much my favorite thing ever, is that the neural network apparently encountered some fan fiction stories that were not in English. As a result, it learned to do this from time to time:

The Secretary Of the World  Challenge inspired by GoF and la mating resigns de la mill colors per mereple beruit carteur la pelete el wert rardo completing and herillo intus den una a des rush sentines kelta an transoles… 

Between by Cheyangel13 A series of fivers are unexpectedly depressed and controlled by the bed, with least more from una perfemale erpensa de the maesse akai suidadium dela vida call de la los se terriuus do form en sou dies de fasurard il resisted de for dogs la sementu sein prong colors itu dee adte se sige natard…

The neural network has also learned to employ capital letters:

Les finds love by violet_quill for starstruck1986 Severus Snape wanted him to be more and she likes Draco.  The person he wants an energy to him.  WHALIDE NO GEATIRE SOURR INSPE AHARMANABLISH ALL SOME TO VERY THE RERIDE!!!!!!!

secret Quidditch by snapsleert Collapse and find the second worst and very different. See Gain and Descent motivate surprising death. Unbusing one of the months: should make more bumo.choooshots. HUGULATED

And the neural network occasionally uses content warnings, although it seems to have a rather fuzzy idea about what to warn its readers about:

Better With The Broom Complicate by Margyn_Black Tonks gets more than the best girl of creation. (Rated Maturisle, mark, a violence, contract) (slash] part of themes) ferret.

Art for the Sun a Scary by disillusionist9 A collection of warnings: characters and situations of silence.

Some of the neural network’s stories, though, are just plain weird.

Harry Potter and the Painful Eyes by dark_pook A Birthday drabble about the problems and a woman who shows up a lot less than she checks at Hogwarts in the destiny to the infamous adventure of control of the Art of The Good Boy Kings With Hermione. Harry and the Blue Special Delicious by apolavia_scg An unexpected potions messaged in the world their lives are to find friendship following the day of different pagers. James and Lily come to the summer before the war.

The Perfect Cow by alafaye Severus and Hermione start a horcruxes

Art: Let Draco roll the light of the moon, and means. by Dangelanne What happens after the war. Not drawn to Draco Malfoy jumpers. Originally written in 2008.

Birds of a Saturday by SasuNarufan13 Harry Potter is drunk and discovers he is an alternate universe.

Holly theody by yesIpxdishoftlyGrinli What would be dangerous! Side Voldemort Jones does all lord off the sunshine show.

Lily Evans and the Ravenclaw of a Christmas Surprise by ci Severus angst the truth of a frighten situation for the wink.

Persuasion by Samanthian The Sorting Hat is fighting in one of the houses.

lily’s family by sharkle Harry woke up in searching after a werewolf Sherlock’s picnic. He is furious.

As a bonus, I leave you with some fairly-plausible screennames the neural network invented, which appear not to be taken (yet):

desire_at_the_malfoy SeverelyAshed fishlingthelovely thedarklyblue phantombeers captainingthetrain siriusly_harry DarkVoldember ChildOfAtSperble all_frogs BelladonnaLeek Sneaking_UnicornWitch bluemelooppiesweatled

https://vimeo.com/175247441

Stealth Hounds

Japanese Virtual Reality arcade VR-Zone Shinjuku is hosting a Field VR multiplayer experience based on Ghost In The Shell:

As a rookie of the special force team created by Motoko Kusanagi, join in the fight against the terrorist organization.

Employ iconic, powerful technology from Ghost in the Shell such as optical camouflage, prosthetic body, cyberbrain, etc. Become fully immersed and experience futuristic warfare for yourself. 

More Here

slide the light off you you may find some peace

Photo-editing app FaceApp now includes Black, Asian Indian and Caucasian filters

On Wednesday morning, the photo-editing app FaceApp released new photo filters that change the ethnic appearance of your face.

The app first became popular earlier in 2017 due to its ability to transform people into elderly versions of themselves and different genders.

These new options, however, will likely cause some outrage: The filters are Asian, Black, Caucasian and Indian.

Selfie apps like Snapchat have taken criticism for filters that apply “digital blackface.” In 2016, Snapchat released a Bob Marley filter that darkened the skin and gave users dreadlocks. Snapchat said another one of its 2016 filters was “inspired by anime,” but many people called it “yellowface,” as it seemingly turned the user into an Asian stereotype.

FaceApp’s newest filters, however, don’t pretend they’re anything but racial. Read more (8/9/17 12 PM)

follow @the-future-now