Google announces the official version of TensorFlow: Open Source Machine Learning

On February 16, Google suddenly posted an announcement on its WeChat official account "Google Blackboard": the first TensorFlow Developer Summit was held in Mountain View, California, USA, and TensorFlow 1.0 was officially released.

As for TensorFlow, we reported on it as early as 2015. On November 9, 2015, local time in the United States (November 10, Beijing time), Google Research announced the launch of the second-generation machine learning system TensorFlow, which is open source.

TensorFlow is Google's internal machine learning system for many years. Its predecessor was the DistBelief internal machine learning project established in 2011.

As for why TensorFlow was launched and open sourced, Google said at the time: "There should be a real tool that researchers can use to try out their crazy ideas. If those ideas work, they will be able to turn them directly into products without the researchers having to rewrite the code."

TensorFlow is flexible, portable, easy to use, and fully open source, so anyone can use it for free.

What improvements are there in the new TensorFlow 1.0?

Regarding the above features, Google said that compared with the existing versions, the new TensorFlow 1.0 launched this time has the following major improvements:

Faster: The TensorFlow team will release several model updates to show how to make full use of TensorFlow 1.0. It uses the term "faster than you can imagine" to describe it;

More flexible: TensorFlow 1.0 introduces a high-level application programming interface with more modules for TensorFlow. At the same time, the TensorFlow team has announced the introduction of the tf.keras module that is fully compatible with Keras (another high-level neural network library);

Always-on: TensorFlow 1.0 ensures the stability of the Python API, making it easier to get new features without breaking existing code.

In addition, TensorFlow 1.0 has the following highlights:

1) The Python API was adjusted to be more similar to NumPy;

2) Experimental APIs for Java and Go;

3) After merging skflow and TF Slim, high-level API modules transferred from tf.contrib.learn: tf.layers, tf.metrics and tf.losses;

4) An experimental version of XLA (domain-specific compiler for TensorFlow graphs) was released for CPUs. XLA is developing rapidly, so expect more progress in future releases;

5) Introduced the TensorFlow debugger, which is a command-line interface and application program interface for running TensorFlow programs;

6) Demonstration of a new system for object detection and localization, and stylization of camera images;

7) Installation improvements: Added Python 3 docker image and made pip package compatible with PyPI. Now you can install TensorFlow by simply calling "pip install tensorflow".

Google said the TensorFlow ecosystem will continue to grow by using dynamic batching technologies such as Fold, web tools such as Embedding Projector, and updating existing tools such as TensorFlow Serving.

What are some practical uses of TensorFlow?

TensorFlow celebrated its first anniversary on November 10 last year (Beijing time). In such a short period of time, it became the framework with the most forked repositories on GitHub.

On this day, Google also played a trick:

Tracking Australia's endangered manatees

Using the latest image recognition technology from TensorFlow, researchers have enabled computers to "learn" to identify manatees in giant aerial photos at a speed far exceeding that of humans, and with an accuracy 1.4 times that of the naked eye.

Sorting and storing cucumbers

A vegetable farmer in Japan used TensorFlow to build an automatic sorting and storage system for the large amount of cucumbers he harvested. After automatically taking pictures, TensorFlow automatically sorted them into up to nine different quality levels.

Google Neural Machine Translation.

TensorFlow and Tensor Processing Units (TPUs) are hardware accelerators specifically built for the Google Neural Machine Translation (GNMT) model, which reduces translation errors by 55% to 85% by processing complete sentences as a whole instead of translating words and phrases in the sentence individually. This is a very Google approach.

In addition, it is reported that TensorFlow is now used in more fields, including assisting in the diagnosis of Parkinson's syndrome, train positioning, composing music, etc.

In a 100-page report, Goldman Sachs highly praised the value of TensorFlow: "In promoting AI integration, the open source application TensorFlow has set a precedent for other cloud platforms and research communities to use company resources. At the same time, Google is taking advantage of its own advantages, such as TPU, to fully utilize the open source world to provide the company with a competitive advantage, even though its machine learning library is open source."

Just on January 31, Google released TensorFlow 1.0.0-rc0, which can run machine learning applications on smartphone-level hardware.

Now, it has finally reached its official version. In short, Google has tasted the sweetness of open source.

As a winner of Toutiao's Qingyun Plan and Baijiahao's Bai+ Plan, the 2019 Baidu Digital Author of the Year, the Baijiahao's Most Popular Author in the Technology Field, the 2019 Sogou Technology and Culture Author, and the 2021 Baijiahao Quarterly Influential Creator, he has won many awards, including the 2013 Sohu Best Industry Media Person, the 2015 China New Media Entrepreneurship Competition Beijing Third Place, the 2015 Guangmang Experience Award, the 2015 China New Media Entrepreneurship Competition Finals Third Place, and the 2018 Baidu Dynamic Annual Powerful Celebrity.