The TensorFlow Developer Summit 2018 was held at the Computer History Museum in California on March 31, 2018. The conference will bring together machine learning developers from all over the world for a day of technology sharing and demonstrations.

There weren’t many surprise announcements at the conference.

Of course, there are some noteworthy changes.

One of the most discussed topics was TensorFlow’s support for more programming languages. Mostly JavaScript and Swift.

For one thing,TensorFlow has released tensorflow.js, a machine learning framework for JavaScript developers

This is a machine learning framework for JavaScript developers that can define and train models entirely in the browser, import offline training TensorFlow and Keras models for prediction, and seamlessly support WebGL.

Using tensorflow.js in a browser allows you to extend many more application scenarios, including interactive machine learning, where all data is stored on the client side, and so on.

In fact, the new tensorflow. js release is based on the previous deeplearn.js, but integrated into TensorFlow.

Google also offers several examples of tensorflow.js:

Game: Emoji treasure hunt

Address: emojiscavengerhunt.withgoogle.com/

More can be found at: js.tensorflow.org/

Second, TensorFlow for Swift will be open source in April

Although the project is still in its early stages, there is a lot of anticipation. For example, Jeremy Howard, founder of Fast. Ai and former president of Kaggle, listed it as one of the summit’s most important announcements and said, “Can we finally put Python behind us?”

Less information about TensorFlow for Swift, interested can visit the following address: www.tensorflow.org/community/s…

In addition, TensorFlow has some new features.

Includes the TensorFlow Hub. “Designed to facilitate the publication, discovery, and use of reusable parts of the model… They contain variables that have been pre-trained on large data sets and can be retrained with a smaller data set to improve generalization or speed up training “.

This part of the explanation is quoted from Google’s official wechat account TensorFlow.

Cloud TPU will also be faster and stronger.

If you’re interested in learning more about the summit, please visit the newly launched TensorFlow blog at blog.tensorflow.org (jump medium.com).

TensorFlow 1.7.0 was released early

The 2017 TensorFlow Developer conference was also held in Mountain View on February 16 last year. At the conference, Google introduced Version 1.0 of TensorFlow.

However, there will definitely be no version 2.0 at this developer conference.

Shortly before this developer conference, Google released Version 1.7.0 of TensorFlow. Major improvements include moving the Eager mode out of contrib and so on.

Most notably, since this release, TensorFlow has been fully integrated with Nvidia’s TensorRT.

As a library, TensorRT can optimize TensorFlow’s FP16 floating-point and INT8 integer calculations, but also maximize throughput, reduce GPU reasoning delays, and more.

According to Google, TensorFlow with TensorRT runs ResNET-50 eight times faster than the unintegrated version.

For more information on TensorFlow version 1.7.0, visit GitHub. Address: github.com/tensorflow/…

In mid-month, Stack Overflow released a survey of 100,000 programmers. According to the survey, TensorFlow is the most popular framework for programmers.

TensorFlow is the most popular developer framework, with 73.5% of current programmers saying they want to continue using it, followed by Torch/PyTorch in third place and 68% of users planning to continue using it.

TensorFlow is at number 3 and Torch/PyTorch is at number 10 among the most wanted frameworks for programmers to learn. 15.5% of programmers who have not yet used TensorFlow plan to learn it, while 4.5% of those who have not used Torch/PyTorch plan to embrace it.

At the same time, there was a lot of dissatisfaction with the two machine learning frameworks, with 26.5 percent of TensorFlow users wanting to leave it.

The above content is from the wechat public account QbitAI, Xia Yi pretended to be from the Computer History Museum, the copyright belongs to the author. Commercial reprint please contact the author for authorization, non-commercial reprint please indicate the source. Some contents are different from the original text.

Read more at ➡️What killers were announced at TensorFlow developer Summit 2018?

TensorFlow Chinese documentation 📋

To help developers and researchers understand how to learn and use TensorFlow, we have prepared the following. TensorFlow in Chinese is also provided.