Top10 most popular open source machine learning projects in June

• Average number of Github collections: 764⭐️

• Machine learning making need open source, https://github.com/Mybridge/machine-learning-open-source.

• Topics: NLP Architecture, Video classification, Mlflow, Classic Games, Dragonfire, Opencv, Computer Vision, Star GAN, Glow, Generative Compression

These open source projects are useful to programmers, so hopefully you’ll find an interesting one that piques your interest.

Intel AI Lab’S NLP Architecture: A Python library that explores state-of-the-art NLP deep learning topologies and techniques [Github 1194 stars]

The current version of NLP Architect contains features that are interesting both from a research perspective and for practical use:

• The NLP core model can provide powerful language feature extraction capabilities for NLP workflows: for example, profilers (BIST) and NP chunker

• NLU modules that provide first-class performance: e.g., Intent extraction (IE), Name entity Recognition (NER)

• Modules that address semantic understanding: e.g., connotation extraction, most common word meanings, NP embedded representations (e.g., NP2V)

• Conversational AI components: Such as the ChatBot application, including the conversational system, sequence chunking, and IE

• End-to-end DL applications using new topologies: e.g. Q&A, machine reading comprehension

Thanks to Intel Nervana[Intel AI LABS for introducing NLP architecture:

https://ai.intel.com/introducing-nlp-architect-by-intel-ai-lab/

Making links:

https://github.com/NervanaSystems/nlp-architect

Video-nonlocal-net: a non-local neural network for Video classification, developed with Caffe2

About Facebook Research: https://medium.com/ @fbResearch

The paper address: https://arxiv.org/pdf/1711.07971.pdf

Making links:

https://github.com/facebookresearch/video-nonlocal-net

Open source platform for the entire machine learning lifecycle

MLflow is currently an alpha version, meaning API and data formats are subject to change and will not run on Windows.

Making links:

https://github.com/databricks/mlflow

[Github 905 stars]

Gym Retro is used by OpenAI to study reinforcement learning (RL) algorithms and to study generalization. Previous research on RL has focused on optimizing agents to solve a single task. With Gym Retro, we can study the ability to generalize between games that are similar in concept but different in appearance. In addition, OpenAI also introduced a new integration tool for Wechat game Sister Tian.

Video address: https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/retro-heavy/output2.mp4

This version includes games for the Sega Genesis and Sega main systems, as well as nintendo NES, SNES and Game Boy consoles. It also includes initial support for The Sega Game Gear, Nintendo Game Boy Color, Nintendo Game Boy Advance and NEC TurboGrafx. Several released game integrations (including those in Gym Retro’s data/experiments folder) are in beta status. Due to the scale of the changes involved, the code will only be available on some games for the time being.

OpenAI introduced: https://blog.openai.com/gym-retro/

Making links:

https://github.com/openai/retro/tree/develop

Open Source AI Assistant for Ubuntu Linux Distribution [Github 688 stars]

Supported Environment:

Dragonfire Execute command steps:

• Search for built-in commands and evaluate algebraic expressions

• Try learning advanced NLP and database management techniques

• Ask the all-knowing Q&A engine (thanks to everyone who contributed to Wikipedia)

• Deep Conversation system was used to respond, which is a SEQ2SEQ neural network trained using Cornell Movie-Dialogs Corpus

Dragonfire uses Mozilla DeepSpeech to understand your Speech commands and the Festival Speech Synthesis System to handle text-to-speech tasks.

You can go to the Gitter chat rooms (https://gitter.im/DragonComputer/Lobby), or Twitter account with Dragonfire experience the pleasure of talking to her in person.

DRAGON.COM PUTER is introduced:

https://github.com/DragonComputer/Dragonfire

Making links:

https://github.com/DragonComputer/Dragonfire

Face, video, text detection and recognition project (using automatic translator: Chinese – > English) [Github 1482 stars].

function

1. Face detection and recognition (pictures and videos)

2. Outline of the logo

3. Head composition (putting hats on people)

4. Digital makeup (lipstick, eyebrows, eyes, etc.)

5. Gender identification

6. Facial expression recognition (anger, disgust, fear, happiness, sadness, surprise, calm and other seven emotions)

7. Video object extraction

8. Image repair (can be used for watermark removal)

9. The picture is automatically colored

10. Eye Tracking (to be perfected)

11. Face change (to be perfected)

The development environment

• Windows 10 (X64)

• Python 3.6.4 radar echoes captured

• OpenCV 3.4.1 track

• Dlib 19.8.1

• Face_recognition 1.2.2

• Keras 2.1.6

• Tensorflow 1.8.0 comes with

• Tesseract OCR 4.0.0 – beta. 1

Making links:

https://github.com/vipstone/faceai

Embedded Computer Vision and Machine Learning Library (CPU optimization and IoT features) [Github 557 stars]

SOD is the modern cross-platform embedded computer vision and machine learning library and has released a set of used to deal with the in-depth study of senior media analysis and API, including real-time, multidisciplinary training the model of object detection and embedded system resources and the Internet of things device, designed to provide a common infrastructure for computer vision applications, And accelerate the use of machine awareness in open source and commercial products.

SOD’s current computer vision algorithms support but are not limited to mobile robots, AR/VR, genetics, human-computer interaction, machine automation, etc.

Notable SOD functions:

• Designed for real-world and real-time applications.

• State of the art CPU optimized deep neural network, including brand new exclusive RealNets architecture.

• No patent required, advanced computer vision algorithms.

• Support for major image formats.

• Simple, clean, and easy to use API.

• In-depth understanding of limited computing resources, embedded systems and iot devices.

• Easy interpolation using OpenCV or any other proprietary API.

• The pre-training model is available for most architectures.

• Support CPU RealNets model training.

• Complete, cross-platform, high quality source code.

• SOD is free, written in C, and can compile and run on almost any platform and architecture. Merge – Merge all SOD source files into a single C file (SOD.c) for easy deployment.

• Open source, actively develop and maintain products.

• Developer friendly support channel (https://sod.pixlab.io/support.html)

SOD programming guide: https://sod.pixlab.io/intro.html

Symisc Systems home page: https://sod.pixlab.io/

Making the link: https://github.com/symisc/sod

StarGAN’s simple Tensorflow implementation (CVPR 2018 Oral) [Github 382 stars], thanks to Junho Kim.

StarGAN is a framework for solving multi-domain image-to-image conversion problems using a single dataset, combining multiple datasets containing different tag sets and flexibly using these tags for image translation.

The thesis links: https://arxiv.org/pdf/1711.09020.pdf

Making links:

https://github.com/taki0112/StarGAN-Tensorflow

Neural Network Hardware Accelerator Compiler [Github 603 stars], thanks PyTorch.

Glow is a machine learning compiler and execution engine designed for a variety of hardware goals and used as a back end for advanced machine learning frameworks. The compiler can perform state of the art compiler optimization and code generation for neural network diagrams. The library is still in the experimental and development stage.

The working principle of

Glow reduces the traditional neural network data flow graph to a two-stage strongly typed intermediate representation (IR). Advanced IR allows the optimizer to perform domain-specific optimizations. The lower levels of instruction-based ONLY IR allow the compiler to perform memory-related optimizations, such as instruction scheduling, static memory allocation, and copy elimination. At the lowest level, the optimizer performs machine-specific code generation to take advantage of its hardware features. Glow’s downscaling phase features a large number of input operators as well as a large number of hardware targets, by eliminating the need for the compiler to implement all operators on all targets. The reduction phase is designed to reduce the input space and allow the new hardware back end to focus on a few linear algebra primitives. The concept is described in the arXiv paper.

The thesis links: https://arxiv.org/abs/1805.00907

Making links:

https://github.com/pytorch/glow

Generative-compression: TensorFlow implementation of GAN extreme learning image compression. [Github 225 stars] thanks to Justin Tan.

The method of Generative-compression was proposed by Agustsson et al in Generative Adversarial Networks for Extreme Learned Image compression.

The thesis links: https://arxiv.org/pdf/1804.02958.pdf

Making address:

https://github.com/Justin-Tan/generative-compression

Original link:

https://medium.mybridge.co/machine-learning-open-source-projects-of-the-month-v-june-2018-d87e2ca3e13f