Latest Features

Learn about the latest MATLAB features for machine learning

R2020b Highlights for Machine Learning

AutoML

Automatically select the best model and associated hyperparameters for regression

Simulink

Simulate and generate code and speed up training of SVM models in Simulink

Model Interpretability

Obtain locally interpretable model-agnostic explanations by finding important predictors (LIME)

Explore all of the Latest Machine Learning Features

Interactive Apps

  • Use the Classification Learner and Regression Learner apps to interactively explore data, select features, and train and evaluate supervised classification and regression models
  • Perform automated tuning of hyperparameters and apply cost matrices from within the learner apps
  • Fit data to a wide range of probability distributions and explore the effects of changing parameter values using the Distribution Fitter app

Related Products: Statistics and Machine Learning Toolbox

Automated Model Optimization

  • New Automatically select the best model and associated hyperparameters for regression and classification (fitrauto and fitcauto)
  • Automatically tune hyperparameters using Bayesian optimization
  • Automatically select a subset of relevant features using techniques like neighborhood component analysis (NCA) and feature ranking
  • Parallelize the execution of automated optimization methods on multiple cores using Parallel Computing Toolbox, and scale to clouds and clusters using MATLAB Parallel Server

Related Products: MATLAB Parallel ServerParallel Computing ToolboxStatistics and Machine Learning Toolbox

Machine Learning and Statistical Algorithms 

  • Leverage commonly used algorithms for classification and regression, such as linear and generalized linear models, support vector machines, decision trees, ensemble methods, and more
  • Use popular clustering algorithms including k-means, k-mediods, hierarchical clustering, Gaussian mixture, and Hidden Markov models
  • New Obtain locally interpretable model-agnostic explanations by finding important predictors (LIME)
  • New Train linear regression and binary classification models incrementally
  • New Extrapolate partial class labels to the entire data set using graphs and self-trained models (fitsemigraph, fitsemiself)
  • Use density-based spatial clustering of applications with noise (DBSCAN) and spectral clustering of arbitrary shapes
  • Run statistical and machine learning computations faster than with open-source tools

Related Products: Statistics and Machine Learning Toolbox

Data Visualization

  • Explore the structure of your data and relationships between features through scatter plots, box plots, dendrograms, and other standard statistical visualizations
  • Use advanced dimensionality reduction algorithms like Stochastic Neighbor Embedding (t-SNE)
  • Visualize high-density data with improved scatter plots in the Classification Learner app
  • Create confusion matrices from tall arrays

Related Products: Statistics and Machine Learning Toolbox

Deployment and Simulink Integration

  • Automatically generate C/C++ code for many popular classification, regression, and clustering algorithms
  • Deploy to devices with limited memory and/or power using fixed-point and single precision arithmetic
  • Update parameters of deployed models such as SVM, linear models, and decision trees, without regenerating C/C++ prediction code
  • New Simulate and generate code for SVM models in Simulink

Related Products: MATLAB Coder, MATLAB Compiler, Statistics and Machine Learning Toolbox

Big Data 

  • Use tall arrays with many classification, regression, and clustering algorithms to train models on data sets that do not fit in memory
  • Fit multiclass classification models, perform hyperparameter optimization, and specify cost with tall arrays
  • Use fast approximate means, quantiles, and non-stratified partitions on out-of-memory data

Related Products: Parallel Computing Toolbox, Statistics and Machine Learning Toolbox