Toolkits :: Remember, remember, the velvet November


node:	Toolkits
template:	2
parent:	Software and Platforms
owner:	ulkas
viewed by:
created:	01.09.2016 - 10:28:15

cwbe coordinatez:
101
63533
8106902
8207635
8207639

ABSOLUT
KYBERIA

permissions
you:	r,
system:	public
net:	yes

software for local (offline) use

000001010006353308106902082076350820763908453622

https://keras.io/

Keras is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano.

000001010006353308106902082076350820763908427321

https://blog.mozilla.org/blog/2017/11/29/announcing-the-initial-release-of-mozillas-open-source-speech-recognition-model-and-voice-dataset/

000001010006353308106902082076350820763908375202

Android and Linux runtimes for neural network model execution
Acceleration support for Qualcomm Hexagon™ DSPs, Qualcomm Adreno™ GPUs and Qualcomm Kryo™, CPUs2
Support for models in Caffe, Caffe2 and TensorFlow formats3
APIs for controlling loading, execution and scheduling on the runtimes
Desktop tools for model conversion
Performance benchmark for bottleneck identification
Sample code and tutorials
HTML Documentation

https://developer.qualcomm.com/software/snapdragon-neural-processing-engine

00000101000635330810690208207635082076390837520208375201

Mobile chip maker Qualcomm wants to enable deep learning-based software development on all kinds of devices, which is why it created the Neural Processing Engine (NPE) for its Snapdragon-series mobile processors. The NPE software development kit is now available to all via the Qualcomm Developer Network, which marks the first public release of the SDK, and opens up a lot of potential for AI computing on a range of devices, including mobile phones, in-car platforms and more. The purpose of the framework is to make possible UX implementations like style transfers and filters (basically what Snapchat and Facebook do with their mobile app cameras) with more accurate applications on user photos, as well as other functions better handled by deep learning algorithms, like scene detection, facial recognition, object tracking and avoidance, as well as natural language processing. Basically anything you'd normally route to powerful cloud servers for advanced process, but done locally on device instead.

000001010006353308106902082076350820763908327292

Sonnet is a new open source library announced by Alphabet’s DeepMind. It is built on top of their existing machine learning library TensorFlow along with extra features that fit DeepMind’s research requirements. Sonnet is designed to make it easier to create complex neural networks using TensorFlow.
https://github.com/deepmind/sonnet

000001010006353308106902082076350820763908312784

https://cloud.google.com/compute/
High-Performance, Scalable VMs
Google Compute Engine delivers virtual machines running in Google's innovative data centers and worldwide fiber network. Compute Engine's tooling and workflow support enable scaling from single instances to global, load-balanced cloud computing.

Compute Engine's VMs boot quickly, come with persistent disk storage, and deliver consistent performance. Our virtual servers are available in many configurations including predefined sizes or the option to create Custom Machine Types optimized for your specific needs. Flexible pricing and automatic sustained use discounts make Compute Engine the leader in price/performance.

00000101000635330810690208207635082076390831278408312785

Google is doubling the maximum number of CPU cores developers can use with a single virtual machine on its Compute Engine service from 32 to 64. These high-power machines are now available in beta across all of Google's standard configurations and as custom machine types, which allow you to select exactly how many cores and memory you want. If you opt to use 64 cores in Google's range of high-memory machine types, you'll also get access to 416GB of RAM. That's also twice as much memory as Compute Engine previously offered for a single machine and enough for running most memory-intensive applications, including high-end in-memory databases. Running your apps on this high-memory machine will set you back $3.7888 per hour (though you do get all of Google's usual sustained-use discounts if you run it for longer, too).

000001010006353308106902082076350820763908302355

https://github.com/google/multibox
This repository contains a reference pre-trained network for the multibox object detector, complementing the Google publications

000001010006353308106902082076350820763908302354

https://github.com/google/inception
This repository contains a reference pre-trained network for the Inception model, complementing the Google publication.

000001010006353308106902082076350820763908296251

Google Brain Team
Make machines intelligent. Improve people’s lives.
https://research.google.com/teams/brain/

00000101000635330810690208207635082076390829625108296254

The project's technology is currently used in the Android Operating System's speech recognition system,[17] photo search for Google+[18] and video recommendations in YouTube.[19]

00000101000635330810690208207635082076390829625108296253

The so-called "Google Brain" project began in 2011 as a part-time research collaboration between Google Fellow Jeff Dean, Google Researcher Greg Corrado, and Stanford University professor Andrew Ng.[1][2][3] Ng had been interested in using deep learning techniques to crack the problem of artificial intelligence since 2006, and in 2011 began collaborating with Dean and Corrado to build a large-scale deep learning software system, DistBelief,[4] on top of Google's cloud computing infrastructure. Google Brain started as a Google X project and became so successful that it was graduated back to Google: Astro Teller has said that Google Brain paid for the entire cost of Google X.[5]

In June 2012, the New York Times reported that a cluster of 16,000 computers dedicated to mimicking some aspects of human brain activity had successfully trained itself to recognize a cat based on 10 million digital images taken from YouTube videos.[3] The story was also covered by National Public Radio[6] and SmartPlanet.[7]

In March 2013, Google hired Geoffrey Hinton, a leading researcher in the deep learning field, and acquired the company DNNResearch Inc. headed by Hinton. Hinton said that he would be dividing his future time between his university research and his work at Google.[8]

On 26 January 2014, multiple news outlets stated that Google had purchased DeepMind Technologies for an undisclosed amount. Analysts later announced that the company was purchased for £400 million ($650M USD / €486M), although later reports estimated the acquisition was valued at over £500 million.[9][10][11][12][13][14][15] The acquisition reportedly took place after Facebook ended negotiations with DeepMind Technologies in 2013, which resulted in no agreement or purchase of the company.[16]

000001010006353308106902082076350820763908295081

https://kafka.apache.org/
Kafka™ is used for building real-time data pipelines and streaming apps. It is horizontally scalable, fault-tolerant, wicked fast, and runs in production in thousands of companies.

00000101000635330810690208207635082076390829508108295082

Open sourced in 2010, the Apache Kafka distributed streaming platform is now used at more than a third of Fortune 500 companies (as well as seven of the world's top 10 banks). An anonymous reader writes:
Co-creator Neha Narkhede says "We saw the need for a distributed architecture with microservices that we could scale quickly and robustly. The legacy systems couldn't help us anymore." In a new interview with TechRepublic, Narkhede explains that while working at LinkedIn, "We had the vision of building the entire company's business logic as stream processors that express transformations on streams of data... [T]hough Kafka started off as a very scalable messaging system, it grew to complete our vision of being a distributed streaming platform."

Narkhede became the CTO and co-founder of Confluent, which supports enterprise installations of Kafka, and now says that being open source "helps you build a pipeline for your product and reduce the cost of sales... [T]he developer is the new decision maker. If the product experience is tailored to ensure that the developers are successful and the technology plays a critical role in your business, you have the foundational pieces of building a growing and profitable business around an open-source technology... Kafka is used as the source-of-truth pipeline carrying critical data that businesses rely on for real-time decision-making."

000001010006353308106902082076350820763908280054

http://www.movidius.com/news/movidius-announces-deep-learning-accelerator-and-fathom-software-framework
1461855269-Screen-Shot-2016-04-27-at-3.13.16-PM_1425_830_s_c1.png

1461855269-Screen-Shot-2016-04-27-at-3.13.16-PM_1425_830_s_c1.png

"Pretty much any device with a USB port will be able to use advanced neural networks," reports PC Magazine, announcing the new Fathom Neural Compute Stick from chip-maker (and Google supplier) Movidius. "Once it's plugged into a Linux-powered device, it will enable that device to perform neural network functions like language comprehension, image recognition, and pattern detection," and without even using an external power supply.

Device manufacturers could now move AI-level processing from the cloud down to end users, PC Magazine reports, with one New York computer science professor saying the technology means that now "every robot, big and small, can now have state-of-the-art vision capabilities."

The article argues that this standalone, ultra-low power neural network could start the creation of a whole new category of next-generation consumer technologies.

000001010006353308106902082076350820763908249471

Theano is a Python library that allows you to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. It can use GPUs and perform efficient symbolic differentiation. http://www.deeplearning.net/software/theano
https://en.wikipedia.org/wiki/Theano_(software)
https://github.com/Theano/Theano

00000101000635330810690208207635082076390824947108302344

At Google's inaugural TensorFlow Dev Summit in Mountain View, California, today, Google announced the release of version 1.0 of its TensorFlow open source framework for deep learning, a trendy type of artificial intelligence. Google says the release is now production-ready by way of its application programing interface (API). But there are also new tools that will be part of the framework, which includes artificial neural networks that can be trained on data and can then make inferences about new data. Now there are more traditional machine learning tools, including K-means and support vector machines (SVMs), TensorFlow's engineering director, Rajat Monga, said at the conference. And there's an integration with the Python-based Keras library, which was originally meant to ease the use of the Theano deep learning framework. And there are now "canned estimators," or models, Monga said, including simple neural networks to start using quickly.
http://venturebeat.com/2017/02/15/google-releases-tensorflow-1-0-with-new-machine-learning-tools/

00000101000635330810690208207635082076390824947108249476

o kolko lepsie je to ako chainer?

mne totizto zatial najjednoduchsie na win islo rozbehat prave chainer?

ako sa vie pohrat s GPUvsCPU

napriklad chainer ma tu vyhodu ze vysledny model sa da pouzit aj na GPU aj na CPU

0000010100063533081069020820763508207639082494710824947608302346

pozri posledne prispevky, vysla nova verzia tensorflowu od ktorej sa odvijaju zevraj aj theano veci

0000010100063533081069020820763508207639082494710824947608249483

Neviem, ale theano je jeden z backendov pre keras (druhy je tensorflow) a keras je velmi popularny kvoli jednoduchosti pouzitia.

00000101000635330810690208207635082076390824947108249472

momentalne akcia na slashdot deals:
https://deals.slashdot.org/sales/introduction-to-deep-learning-and-artificial-intelligence

000001010006353308106902082076350820763908247032

http://www.ibm.com/watson/developercloud/project-intu.html
https://rg-gateway.mybluemix.net/
https://github.com/watson-intu/

Project Intu is an experimental service that allows developers to quickly and seamlessly integrate various cognitive services, such as Conversation and Speech-to-Text, with the capabilities of various devices, spaces and physical objects. Project Intu allows the device to act out the interaction with the user. Instead of needing to program each individual movement of a device or avatar, Project Intu makes it easy to combine movements that are appropriate for performing specific tasks like assisting a customer in a retail setting or greeting a visitor in a hotel in a way that is natural for the end user.

Project Intu lets you build cognitive experiences running on a wide variety of operating systems, including Raspberry PI, MacOS, Windows and Linux.

00000101000635330810690208207635082076390824703208247036

Intu is an architecture that enables Watson services in devices that perceive by vision, audition, olfaction, sonar, infrared energy, temperature, and vibration. Intu-enabled devices express themselves and interact with their environments, other devices, and people through speakers, actuators, gestures, scent emitters, lights, navigation, and more.

000001010006353308106902082076350820763908241458

Tuktu is a big data analytics platform that focuses on ease of use. The idea of the platform is that its users can focus on (business) logic rather than dealing with technical implementation details. This means that programming tedious pieces of code is a thing of the past. Tuktu comes with a couple of key characteristics:

Support for real-time and batch processing
Synchronous and asynchronous processing
Simple REST API
A visual modeller that allows to create jobs using drag-and-drop modelling
Its own distributed file system, called TDFS, that is very easy to use - alternatively, Tuktu integrates seamlessly with HDFS (and local files of course)
Its own in-memory distributed key-value store, called TuktuDB, for quickly storing and retrieving data
Built-in real-time visualization capabilities for a number of pre-defined visuals
Native support for web-analytics
Periodic (cron) jobs can be scheduled natively from Tuktu
By default there is no master/slave architecture, so no single point of failure
Switch seamlessly between distributed and local (even transactional) computation paradigms
Modular setup, with a number of modules like machine learning or social
Automatic generation of documentation using the meta-files in the modeller
Easy usage and setup: download the zip file and run the startup script (Windows, Mac and Linux supported)

000001010006353308106902082076350820763908241455

https://www.microsoft.com/en-us/research/product/cognitive-toolkit/

The Microsoft Cognitive Toolkit—previously known as CNTK—empowers you to harness the intelligence within massive datasets through deep learning by providing uncompromised scaling, speed and accuracy with commercial-grade quality and compatibility with the programming languages and algorithms you already use. Hear about the team that developed the Cognitive Toolkit, or read more below.

000001010006353308106902082076350820763908223361

https://cloud.google.com/bigquery/

000001010006353308106902082076350820763908220830

https://research.googleblog.com/2016/09/show-and-tell-image-captioning-open.html
https://github.com/tensorflow/models/tree/master/im2txt

A TensorFlow implementation of the image-to-text model described in the paper:

"Show and Tell: Lessons learned from the 2015 MSCOCO Image Captioning Challenge."

Oriol Vinyals, Alexander Toshev, Samy Bengio, Dumitru Erhan.

IEEE transactions on pattern analysis and machine intelligence (2016).

Full text available at: http://arxiv.org/abs/1609.06647

000001010006353308106902082076350820763908215395

ulkas 15.09.2016 - 14:06:07 [1K] , level: 1, UP NEW

Torch

http://torch.ch/
A SCIENTIFIC COMPUTING FRAMEWORK FOR LUAJIT

Torch is a scientific computing framework with wide support for machine learning algorithms that puts GPUs first. It is easy to use and efficient, thanks to an easy and fast scripting language, LuaJIT, and an underlying C/CUDA implementation.

A summary of core features:



a powerful N-dimensional array

lots of routines for indexing, slicing, transposing, ...

amazing interface to C, via LuaJIT

linear algebra routines

neural network, and energy-based models

numeric optimization routines

Fast and efficient GPU support

Embeddable, with ports to iOS, Android and FPGA backends

00000101000635330810690208207635082076390821539508215398

Torch7 provides a Matlab-like environment for state-of-the-art machine learning algorithms. It is easy to use and provides a very efficient implementation, thanks to an easy and fast scripting language (Lua) and a underlying C implementation.

Modified to be compiled and used with Android

For CUDA-enabled version: NVIDIA CodeWorks for Android: https://developer.nvidia.com/codeworks-android. For CPU-only version : Android NDK and Android SDK
https://github.com/soumith/torch-android

00000101000635330810690208207635082076390821539508215396

Torch Packages

Tensor Library
Tensor defines the all powerful tensor object that provides multi-dimensional numerical arrays with type templating.
Mathematical operations that are defined for the tensor object types.
Storage defines a simple storage interface that controls the underlying storage for any tensor object.
File I/O Interface Library
File is an abstract interface for common file operations.
Disk File defines operations on files stored on disk.
Memory File defines operations on stored in RAM.
Pipe File defines operations for using piped commands.
High-Level File operations defines higher-level serialization functions.
Useful Utilities
Timer provides functionality for measuring time.
Tester is a generic tester framework.
CmdLine is a command line argument parsing utility.
Random defines a random number generator package with various distributions.
Finally useful utility functions are provided for easy handling of torch tensor types and class inheritance.

000001010006353308106902082076350820763908207646

https://github.com/zer0n/deepframeworks
https://github.com/soumith/convnet-benchmarks
https://twitter.com/fchollet/status/765212287531495424

000001010006353308106902082076350820763908207643

https://cntk.ai/

Computational Network Toolkit
Production-quality, Open Source, Multi-machine, Multi-GPU,
Highly efficient RNN training,
Speech, Image, Text

CNTK (http://www.cntk.ai/), the Computational Network Toolkit by Microsoft Research, is a unified deep-learning toolkit that describes neural networks as a series of computational steps via a directed graph. In this directed graph, leaf nodes represent input values or network parameters, while other nodes represent matrix operations upon their inputs. CNTK allows to easily realize and combine popular model types such as feed-forward DNNs, convolutional nets (CNNs), and recurrent networks (RNNs/LSTMs). It implements stochastic gradient descent (SGD, error backpropagation) learning with automatic differentiation and parallelization across multiple GPUs and servers. CNTK has been available under an open-source license since April 2015. It is our hope that the community will take advantage of CNTK to share ideas more quickly through the exchange of open source working code.