Some information about machine learning at Google Developer Conference 2016

Time:2020-3-25

Some information about machine learning at Google Developer Conference 2016

Today, Google held a developer conference (GDD) at the National Convention Center in Beijing. Google hasn’t heard from China for several years. This time it’s very meaningful. I love Google, and I just developed an android app a few days ago. I signed up, and I was lucky to get admission.

Main contents of the conference

  • What new features can Android 7. X enable developers to develop

  • Promote firebase

  • Promote angular

  • Promote progressive web apps (PWA)

  • Promote Google play

  • Promote AdWords and AdMob

  • Material design

  • Tensorflow, machine learning, etc

Because the conference was divided into seven sessions, only the new features of Android 7. X and machine learning were listened to. Next, I will mainly record some dry goods that I got today.

TensorFlow

The speaker introduced the use of tensorflow with two examples of linear regression and image recognition.

Characteristic

  • Fast, flexible, and scalable open-source machine learning library

  • One system for research and production

  • Runs on CPU, GPU, TPU and Mobile

learning resource

  • Tutorials: https://tensorflow.org

  • Image recognition: https://tensorflow.org/tutori…

  • Word embeddings: https://tensorflow.org/tutori…

  • Language Modeling: https://tensorflow.org/tutori…

  • Translation: https://tensorflow.org/tutori…

  • TensorFlow playground: http://goo.gl/mXhncM

  • TensorFlow Workshop: https://github.com/random-for…

  • Udacity Course: https://udacity.com/course/de…

Deep learning

Current technical level

  • Images: classifying; adding text descriptions

  • Language: translation; summarization

  • Speech: recognition; production

  • Games: alphago; Atari
    And more.

Repeatability study

  • Inception: https://research.googleblog.c…

  • Show and tell: https://research.googleblog.c…

  • Translation: https://research.googleblog.c…

  • Summarization https://research.googleblog.c…

  • WaveNet: https://deepmind.com/blog/wav…

  • Music: https://magenta.tensorflow.or…

  • Parsey McParseface: https://research.googleblog.c…

Among them, perception is image classification, show and tell is the research of input image output text description; translation and summary are literal translation, not to mention much; WaveNet is the TTS (text to speech) technology that Google has made a breakthrough recently, which will naturally simulate human voice; music refers to the technology based on reinforcement learning Parsey mcparseface is a kind of parsing English parser, which can distinguish the subject predicate object complement and other components in sentences, and is known as the most accurate prediction parser in the world.

Open source gadgets

  • Style Transfer: https://github.com/cysmith/ne…

  • Fast Style Transfer: https://github.com/lengstrom/…

  • ColorNet: https://github.com/pavelgonch…

  • Super Resolution: https://github.com/david-gpu/…

  • TTS: https://github.com/ibab/tenso…

  • Speech Recognition: https://github.com/buriburisu…

  • Many more: https://github.com/jtoy/aweso…

The first style transfer, which has been changing the painting style, has been broken since the famous neural network paper a neural algorithm of artistic style came out. Recently, the popular prism is based on this principle. The others are well understood, that is, literally.

learning resource

In addition to the above learning resources about tensorflow, there are:

  • Totally new to ML?: Recipes (YouTube) http://goo.gl/KewA03

  • Stanford’s CS231n: http://cs231n.github.io

  • Udacity’s Machine Learning Nanodegree: http://goo.gl/ODpXj4

  • Chris Olah’s blog: http://colah.github.io

  • Michael Nielsen’s book: http://neuralnetworksanddeepl…

Among them, Stanford’s cs231n is recommended by Josh Gordon, the keynote speaker of tensorflow and art.

Above.