Google Accelerates Machine Learning With TensorFlow 1.0

Machine learning is a very popular topic right now. Previously neural networks were something that only top computer scientists and grad students had access to, but now there are various open-source machine learning frameworks available for anyone to use including Spark ML, Theano, Microsoft’s CNTK, and of course Google’s TensorFlow. And now, Google are pushing the machine learning boundaries even further with the release of TensorFlow 1.0.


The new version of TensorFlow is more user-friendly and now includes a built-in Estimator function, and also supports more traditional machine learning algorithms such as Support Vector Machines (SVM), Random Forest, and K-Means. Combine this with neural networks, and you have one pretty neat version of TensorFlow that’s ideal for hybrid issues. It also offers vast performance improvements and scaling compared to the original version.

With Keras you can build a model that will analyze videos and is just as powerful as TensorFlow. This interface is designed in a way that is much more user-friendly, making it really simple to create high-end networks. But, one of the most impressive features of TensorFlow is its ability to run on most smartphones. The new version uses the Hexagon DSP that’s integrated into Qualcomm’s Snapdragon 820 CPU to power many of its applications, including Word Lens and Translate, even when offline.

If you want to get involved and see why everyone’s talking about it, TensorFlow 1.0 is available for download now, but you will have to install Keras as a separate package for now until the release of 1.2 comes along where Google are looking to integrate it. A helpful script will also be provided to help users update their existing code if need be. To get the best experience, users are advised to run any machine learning tool on a supported GPU. However, there are some options that involve using the cloud instead that are worth exploring, such as Floyd Hub where you can simply pay for the amount of time you need to build and run your model.



More News To Read

Comments

comments