6 Things to Know About TensorFlow

Dmytro works as a Back-End Developer at Swan Software Solutions. In his spare time, he enjoys Heroes 3 and Running. His other hobby is Crossfit. Dmytro trained people in the fitness regimen for fifteen years. In five years, Dmytro’s goal is to be a Machine Learning Engineer. His favorite inspirational quote–a journey of a thousand miles begins with the first step–will probably help him reach that goal!

Today, Dmytro shares six things one should know about TensorFlow.

1. It is a powerful machine learning framework

TensorFlow is a great Machine Learning framework for those cases where you need to process huge amounts of data and/or when there is a need to train large and deep neural networks. Some of its uses include the prevention of vision loss by diagnosing diabetic retinopathy, the preservation of nature by alerting authorities to illegal deforestation, and the search for distant planets.

2. Neural networks can be built line by line

Building neural networks on TensorFlow becomes easier if you know about the Keras add-on. It is convenient and easy for prototyping, which was previously very lacking in TensorFlow. The tf.keras API will appeal to anyone who prefers object-oriented programming and layer-by-layer modeling of neural networks. In a few lines of code, one can create a feed-forward neural network using standard techniques such as regularization.

3. Not just Python supports TensorFlow

Support for just one language – Python – has long been a problem for those developers who do not speak it. You can now work with TensorFlow in many other languages from R to Swift and JavaScript.

4. Ability to work in the browser

As for JavaScript, it is possible to train and run models in the browser using TensorFlow.js.

5. Lite version for low-power devices

The TensorFlow Lite platform allows you to run neural networks on a variety of devices and its speed is three times faster than the original version of TensorFlow. So, machine learning can be implemented even on a Raspberry Pi or a smartphone.

6. More powerful and specialized hardware resources

Those who are tired of waiting for their regular processor to finish processing data to train a neural network can use Google’s cloud tensor processors, which are designed specifically to work with the TensorFlow library. At the moment, beta testing of the third version of TPU is open.

We appreciate Dmytro taking the time to share. If you’d like to discover more about how Swan Software Solutions can help your company with a custom solution, contact us today!