Includes updates and bug fixes for deep learning.
Providing a framework for designing deep neural networks
GPU Coder, NVIDIA library support
MathWorks announced Release 2018b (R2018b) on the 19th, adding new features to MATLAB and Simulink. This release includes updates and bug fixes for deep learning across the product line. The new Deep Learning Toolbox, which replaces Neural Network Toolbox, provides a framework for designing and implementing deep neural networks. Engineers in image processing, computer vision, signal processing, and systems engineering can use MATLAB to more easily design complex network architectures and improve the performance of deep learning models.

MathWorks has improved user productivity and usability in R2018b through:

The Deep Network Designer app allows users to create complex network architectures or modify pre-trained complex networks for transfer learning.

Network learning performance has been improved beyond desktop capabilities. Supports cloud providers with MATLAB Deep Learning Containers on NVIDIA GPU Cloud and MATLAB Reference Architectures for Amazon Web Services and Microsoft Azure.

Broad support for domain-specific workflows: Make large-scale data collection tasks easier and faster, including ground truth labeling apps for audio, video, and application-specific data stores.

In R2018b, GPU Coder improved inference performance by supporting NVIDIA libraries and adding optimizations such as auto-tuning, layer fusion, and buffer minimization. Additionally, deployment support for Intel and ARM platforms was added using Intel MKL-DNN and the ARM Compute Library.

MathWorks recently demonstrated its commitment to interoperability with the ONNX community, enabling collaboration between MATLAB users and other deep learning frameworks. Using the new ONNX converter in R2018b, engineers can import and export models from supported frameworks such as PyTorch, MxNet, and TensorFlow. This interoperability allows models trained in MATLAB to be used in other frameworks, and similarly, models trained in other frameworks can be brought into MATLAB for tasks such as debugging, validation, and embedded deployment. Additionally, R2018b provides a curated set of reference models accessible with a single line of code, and additional model importers enable models from Caffe and Keras-Tensorflow.

“As deep learning becomes more widespread across industries, engineers and scientists across diverse disciplines need to be able to use, access, and apply it broadly,” said David Rich, MATLAB marketing director at MathWorks. “Now, deep learning novices and experts alike can learn, apply, and conduct advanced research in MATLAB using an integrated deep learning workflow that spans from research to prototyping to production.”