R2022a at a Glance

MATLAB Product Family

MATLAB

• 3 New Apps: Data Cleaner, Hardware Manager, Code Compatibility Analyzer
• pcode: Use new pcode file format with stronger obfuscation
• Python: Use simpler keyword passing with name=value syntax, and editor support for Python code
• Graphics: Create animated GIF files; pass tables directly to plot, plot3, polarplot, and other line plotting functions
• App Building: Create custom UI components with App Designer
• Live Editor: Develop your own Live Editor tasks
• N-D Array Math: tensorprod, pagemldivide, pagemrdivide, and pageinv
• Parquet Files: Use conditional filtering (Predicate Pushdown) with rowfilter; support for reading and writing of row groups and nested data structures

AI, Data Science, and Statistics

Deep Learning Toolbox

• Model Pruning: Create compressed versions of models that consume less computational resources
• MATLAB Deep Learning Model Hub: Discover and use pretrained deep learning models
• Deployment of Imported Models: Generate code for additional Keras and ONNX built-in layers
• Experiment Manager Batch Experiments: Run multiple experiments simultaneously on a remote cluster (requires MATLAB Parallel Server)
• 1D Convolutional Networks: Create and train networks with 1-D transposed convolution for sequence and time series data

Statistics and Machine Learning Toolbox

• Machine Learning Apps: Save and resume sessions, rank and select features by importance, and reserve data for testing
• Incremental Machine Learning: Perform drift detection on streaming data; train kernel and multi-class linear models
• Deployment: Generate C/C++ code for inference with shallow neural nets
•  Simulink: Simulink block for Gaussian process regression

Curve Fitting Toolbox

• Curve Fitter app: Use new toolstrip design for improved usability; select fitting and validation data from MATLAB table data types

Control Systems

Model Predictive Control Toolbox

• MISRA C 2012: Implement MISRA C:2012-compliant controllers with linear MPC and ADAS blocks in Simulink

System Identification Toolbox

• Nonlinear System Identification: Create Hammerstein-Wiener models that use regression functions based on machine learning algorithms

Simulink Design Optimization

• Surrogate Optimization Solver in Response Optimizer and Parameter Estimator Apps: Speed up time-consuming optimization problems

Reinforcement Learning Toolbox

• Model-Based Policy Optimization Agent: Use a model of the environment to improve sample efficiency and exploration
• Multi-Agent Reinforcement Learning: Train multiple agents in a centralized manner for more efficient exploration and learning

Predictive Maintenance Toolbox

• Deployment: Generate C/C++ code for RUL similarity models, rotating machinery metrics, and nonlinear signal features
• Diagnostic Feature Designer: Extract stationary time series features from signal data

Math and Optimization

Optimization Toolbox

• Problem-Based Optimize Live Editor Task: Solve optimization problems and systems of equations using a visual interface

Global Optimization Toolbox

• Problem-Based Optimize Live Editor Task: Solve global and multiobjective optimization problems using a visual interface

Symbolic Math Toolbox

• Symbolic Matrix Functions: Perform parameter-dependent linear algebra calculations in compact matrix notation

Partial Differential Equation Toolbox

• Electromagnetic Analysis: Solve time-harmonic wave scattering and transmission problems using a finite element method
• Thermal Reduced Order Models: Approximate dynamic characteristics of a thermal model for faster execution (e.g., for batteries and CPUs)

Application Deployment

MATLAB Compiler SDK

• Microservice Integration: Create a microservice Docker image using the compiler.package.microserviceDockerImage function

MATLAB Production Server

• Web Request Handlers: Support for custom URL routes and request payloads

Code Generation

AUTOSAR Blockset

• Adaptive AUTOSAR: Use ara::com methods and ara::com with service-oriented communication support events
• Classic AUTOSAR: Simulate basic software component event failure and recovery, including Release 19-11

DDS Blockset

• Support for RTI Connext Micro communication middleware

Fixed-Point Designer

• Estimate design costs of data memory consumption and operator counts in generated code

MATLAB Coder and GPU Coder

• Generate generic C/C++ code for more deep learning layers
• Improve performance for deep learning network layers, including SIMD
• Support for additional TensorFlow-Keras and ONNX built-in layers
• Incorporate pretrained TensorFlow Lite models for simulation and code generation

Simulink Coder

• Specify tunable parameters for protected models

Embedded Coder

• Use deployment types to simplify configuration of top and reference model interfaces
• Improve compliance for MISRA C:2012, MISRA C++:2008, and AUTOSAR C++14
• Profile stack usage to assess memory consumption