And how MATLAB helps you take advantage of them.

Last post*, Steve Eddins wrote about some of the new features in the latest release. Today, I’d like to talk about how these new features fit into some larger trends we’re seeing in deep learning.

You may have noticed we continue to add more features for the intermediate stage of deep learning today: between building your first model and having a finished product.

This void between starting and finishing is where we see a lot of engineers spend a huge portion of their time with various tasks such as:

• Increasing the accuracy of a model with parameter tuning.
• Converting models to C or CUDA to take advantage of speed and hardware.
• Experimenting with new network architectures for transfer learning.

We are starting to see new trends emerge in response to these tedious and time-consuming tasks.

While the following trends aren’t MATLAB specific, our latest release has the capabilities to fully embrace this intermediate stage of deep learning. If you’re past “What is Deep Learning?”, read on to explore 3 trends that are emerging after getting into the weeds with deep learning.

---

Trend #1: Cloud Computing

We all know training complicated networks takes time. Adding to that, techniques like Bayesian optimization - which will train your network multiple times with different training parameters – can provide powerful results at a cost: more time. An option to alleviate some of this pain is to move from local resources to clusters (HPC) or the cloud. The cloud is emerging as a great resource: providing the latest hardware, multiple GPUs at one time, and only paying for resources when they are needed.

How MATLAB helps you with this trend:

• Check out the MATLAB cloud computing page.
• And MATLAB specific NVIDIA GPU Cloud (NGC) support in our documentation.
• There's also a walk-through video on how to set up MATLAB and NGC here.

Trend #2: Interoperability

Let’s face it: There isn’t a single framework that can provide ‘best-in-class’ for everything about deep learning from start to finish. The trend of interoperability between deep learning frameworks, primarily through ONNX.ai, is allowing users switch in and out of deep learning frameworks at their convenience. MathWorks is part of a collective pushing this trend forward, which is why it’s a great time to check out a variety of deep learning frameworks.

How MATLAB helps you with this trend:

• MATLAB has ONNX import and export capabilities through this support package

Trend #3: Multi-deployment options

Let’s say you’ve made it to the finish line: You have a deep learning model to perform the task you envisioned. Now you need to get the model to its final destination. Multi-deployment can have various definitions, so let me define this as “deploy your model to the right location depending on your specific need.” This could be the web, your phone, embedded processors, or GPUs.

If your goal is GPUs, CUDA is providing the best and most efficient processing through code optimization. Yes, CUDA has been around for a while, but optimization libraries like TensorRT and Thrust are worth a look. It’s not unheard of for TensorRT to speed up ordinary CUDA code 30%, and that’s beyond the 200% speedup you can get converting framework-specific code to CUDA. (We'll talk more specific numbers in future posts about performance.)

How MATLAB helps you with this trend:  

• GPU Coder is the product to watch. It was named Embedded Vision’s 2018 product of the year.
• MATLAB has coder tools and support packages to various devices: including iOS, Android, and FPGA to name a few
• Though not deep learning specific, I've heard App Designer now supports Web App deployment.

With our most recent release, MATLAB has the capabilities allowing you to fully embrace these trends, and we'll continue to respond as the trends change and evolve. This release is a particularly good one for new deep learning features, and I encourage you to take a deeper look.    

---

*Introductions: As Steve mentioned in his last post, I’ll be taking over the blog, and I’m very excited for this new challenge! For those interested, I want to introduce myself and talk about my vision for this blog.

As you may have seen, I’ve been warming up for this role as a guest blogger writing about "deep learning in action" (part 1, part 2 and part 3) And prior to that, I took over the ‘pick of the week’ blog and wrote about our deep learning tutorial series. I maintain the MATLAB for Deep Learning content, and I appear in a few videos on our site from time to time.

Background: I've been at MathWorks 5 years. I started as an Application Engineer, which meant I got to travel to customer sites and present to customers, specializing in image processing and computer vision. My theory is most of you reading this have never experienced a MathWorks seminar, and I’d like that to change. I now work in marketing**: My current job is making sure everyone knows about the capabilities of the tools and how to solve their problems, and this blog fits well within this job description.

**Some people shudder when they hear the word marketing. It’s still a technical role, I just happen to also like spending time on better wording, formatting and visualizations. It’s a win-win for you. You’ll see!

Blog Vision: My vision for this blog in one word is access. I have access to insider information because I work here. My goal for this blog is to be the source of that information. I want to talk about the behind-the-scenes deep learning things that you may not see by reading the documentation. While I can’t share future plans, I can give insight, demos, developer Q&A time, and code you won’t find in the product. That’s the vision. I hope you’ll join me on this journey!