Release 2017b Highlights

Release 2017b makes it easier than ever to do deep learning and data analytics with MATLAB®. You can capture more of your software design in Simulink®, and use new tools for comprehensive verification and validation. 

MATLAB makes deep learning more accessible to everyone, even if you're not an expert. Creating a classifier always starts with labeled data, and when you need to start from scratch with your own data, you can save time and effort by using the image labeling app. Then, you can use DAG networks and custom layers to produce more complex and accurate network structures, like GoogLeNet.  

When you’re ready to deploy a trained deep learning network, the new GPU Coder automatically generates CUDA code, which can run directly on NVIDIA GPUs. This gives you amazing performance compared to running the network on other frameworks. And with the introduction of LSTM networks, deep learning with MATLAB extends beyond images to any kind of time series and text data.

So, let’s talk data analytics. It’s often difficult to extract meaningful insights from massive repositories of text data. Using our new text analytics tools, you can do things like use maintenance logs to inform on predictive maintenance. And those of you in finance can quickly perform sentiment analysis on news data, which can be incorporated into your trading algorithms.  

Of course, all this starts with getting your data into MATLAB. You can now interface directly with MongoDB and Azure blob storage, and use the updated Database Explorer App to visually explore any relational database without needing to know SQL.   

Once your data is in MATLAB, you can visualize your results in new ways, like geobubble and wordcloud. If your data is too big to fit into memory, remember that tall arrays enable you to interact with it using the same familiar syntax. You can use tall arrays directly with even more statistics and machine learning algorithms as well as familiar graphing functions.

Simulink has new products that help you verify and validate your models. You can model requirements and trace them to your design in Simulink, and as you build your models, you can quickly measure how well they comply to your standards and guidelines. Finally, you can determine how effectively you're testing your models and generated code.  

You now have more ways to model and combine software components, along with the ability to simulate them with a scheduler to see how they behave before embedding them into a software environment.  

If you’re excited about these new features, 17b makes it easier than ever to upgrade your Simulink models, libraries, and projects, and introduces the Code Compatibility Report to help update your MATLAB code. 

There’s plenty more in this release, so be sure to download 17b and check out more content while it installs. Thanks for watching, and don’t hesitate to leave us your feedback.  

Product Focus