Getting started with AI in MATLAB

Dr Julia Hoerner Amith Kamath

AI Academic Liaison Manager/ Europe AI Academic Liaison Manager/ Asia-Pacific

[email protected] [email protected]

1
 In this workshop we will …
Learn the Fundamentals
of AI

Learn the Fundamentals Learn the Fundamentals

of Deep Learning of Machine Learning

Use real-world ECG Signals

Discuss
deployment
options
2
Set-Up Instructions
▪ Set up a MathWorks account if you don’t have one​
– please use Google Chrome browser
– go to https://www.mathworks.com/mwaccount/

▪ Copy the materials via the MATLAB Drive

– go to https://drive.matlab.com/sharing/bac1e750-8f81-411e-bc94-9db1d5195e77
– Click on Add to my Files/Copy Folder
– You should see a separate folder saying “Owned By: Me” (on the right-hand side)

▪ Activate the workshop license and launch MATLAB Online​

– go to https://www.mathworks.com/licensecenter/classroom/MATLAB_EXPO_8726324/
– click Access MATLAB Online
– log in using your MathWorks account

3
AI in Research and Industry

University of Twente: Augmented Reality of DKFZ Heidelberg: Deep Learning for Tumor
blood flow Detection

Automatic Defect Oversteering Seismic Event

Detection Detection Detection
4
MathWorks Focus on Deep Learning and AI for Engineering and
Science

Predictive Maintenance Computational Finance

• Bearing Prognosis Predictive Maintenance • Machine Learning for Financial
• Pump Fault Diagnosis Toolbox™ Statistical Arbitrage Toolbox™

Land-Use Classification Robotics

• Semantic Segmentation for Image Processing • Avoid Obstacles using Robotics System
Multispectral Images Toolbox™ Reinforcement Learning Toolbox™

Lidar Automated Driving

• Lidar Point Cloud Semantic Segmentation Lidar • Deep Learning Vehicle Detector Automated
• 3-D Object Detection Using PointPillars Toolbox™ • Occupancy Grid with Semantic Segmentation Driving Toolbox™

Radar Visual Inspection

• Radar Waveform Classification Phased Array • Manufacturing Defect Detection Image Processing
• Pedestrian and Bicyclist Classification System Toolbox™ • Anomaly Detection for Cloth Manufacturing Toolbox™

Wireless Communications Audio

• Modulation Classification Communications • Speech Command Recognition Audio
• Detect WLAN Router Impersonation Toolbox™ • Cocktail Party Source Separation Toolbox™

Reinforcement Learning Medical Imaging

• Train Biped Robot to Walk Reinforcement • 3-D Brain Tumor Segmentation Image Processing
• PMSM Motor Control Learning Toolbox™ • Breast Cancer Tumor Classification Toolbox™

5
What is AI?

Artificial Intelligence: The ability of a computer to perform tasks

commonly associated with intelligent beings like learning or problem-solving.

Machine Learning: Learning a task from data without relying on a

predetermined equation. (User may need to provide data features.)

Deep Learning: Learning from raw data without

predetermined features using neural networks with many layers

1950s 2015
6
 Types of Machine Learning
Type of Learning Categories of Algorithms

Objective:
Regression
Easy and accurate computation of day-
Supervised
ahead system load forecast
Learning

Develop predictive Classification

model based on both
Machine input and output data
Learning

Unsupervised
Clustering
Learning

Discover an internal
representation from
input data only
7
 Types of Machine Learning
Type of Learning Categories of Algorithms

Objective:
Regression Train a classifier to classify human
Supervised activity from sensor data
Learning

Classification
Data:
Develop predictive
model based on both
Machine input and output data
Learning Inputs 3-axial Accelerometer
3-axial Gyroscope
Outputs

Unsupervised
Clustering
Learning

Discover an internal
representation from
input data only
8
 Types of Machine Learning
Type of Learning Categories of Algorithms

Objective:
Regression
Given data for engine speed and
Supervised
Learning vehicle speed, identify clusters

Develop predictive Classification

model based on both
Machine input and output data
Learning

Unsupervised
Clustering
Learning

Discover an internal
representation from
input data only
9
 Types of Machine Learning
Type of Learning Categories of Algorithms Examples of Models

Support
Discriminant Naïve Nearest Neural
Regression Vector
Analysis Bayes Neighbor Networks
Machines
Supervised
Learning
Linear
Ensemble Decision Neural
Classification Regression, SVR, GPR
Methods Trees Networks
GLM
Machine
Learning

Hidden
Unsupervised K-Means, Gaussian Neural
Clustering K-Medoids
Hierarchical
Mixture
Markov
Networks
Learning Model

Deep Learning

Reinforcement
Learning
10
Machine Learning and Deep Learning Datatypes

Image Signal

Numeric Text

11
 AI Workflow

Data Preparation AI Modeling Simulation & Test Deployment

Data cleansing and Model design and Integration with

Embedded devices
preparation tuning complex systems

Hardware
Human insight System simulation Enterprise systems
accelerated training

Simulation- System verification Edge, cloud,

Interoperability
generated data and validation desktop

I Iteration and Refinement

12 12
 Practical Example: Classify Heart Condition

Normal
ECG Data* Model
Abnormal

ECG characteristics (for ML): ECG transformation (for DL):

*Dataset was curated for 2017 PhysioNet challenge: "normal" ECG data was obtained from the
MIT-BIH Normal Sinus Rhythm database available at https://physionet.org/content/nsrdb/1.0.0/, and
“abnormal” from MIT-BIH Arrythmia database at https://www.physionet.org/content/mitdb/1.0.0/ 13
 ML Exercise: Classify Heart Condition

Goal: classify heartbeat signal into normal and

abnormal using machine learning

To Do:
– Go to Exercises folder
– Open Ex1_ECG_ML_FeaturesStats.mlx and follow along with
the instructor

14
 Beyond traditional Machine Learning

Machine Learning
Machine Learning

Deep Learning
Neural Networks
with many Hidden
Layers

Deep Learning

▪ Learns directly from data

▪ More Data = better model
▪ Computationally intensive
▪ Not interpretable

15
 What is Deep Learning?

▪ DL uses neural networks and was inspired by the

human brain
▪ DL neural networks consist of
– Neurons arranged in layers
– Layer combinations
– Learnable parameters (weights and biases)
– Hyperparameters (e.g. learning rate, number of epochs,
mini batch size, etc.)
▪ Most commonly, DL is used for:
– Classification: Output is categorical (or discrete)
– Regression: Output is numerical (or continuous)
– (Can also be used to generate things, e.g. GANs)

16
 What is Deep Learning?

▪ Layers are like blocks

– Stack on top of each other
– Replace one block with a different one
▪ Information is usually passed in a forward pass (but
can also be passed backwards)
▪ Weights and biases are adjusted in a backward pass
(backpropagation) using a gradient descent

▪ There are different networks for different applications

(e.g. CNNs for images, RNNs for sequential data)

17
Signal Processing Architectures

Feature Engineering

Recurrent Neural Network (RNN)

(e.g. Long Short-Term Memory (LSTM) Networks)

Spectrogram

OR

Convolutional Neural Networks (CNN)

Time-Frequency Transformation 18
 Quick overview of Convolution Neural Networks

▪ CNNs are typically used to classify images

▪ CNNs extract features of different granularities
▪ A lot of pretrained CNN models exist in MATLAB
▪ A very good starting point to use with transfer learning

19
 DL Exercise: Classify Heart Condition

Goal: classify heartbeat signal into normal and

abnormal using deep learning

To Do:
– Go to Exercises folder
– Open Ex2_ECG_DL_CNN.mlx and follow along with the
instructor

The example using LSTM can be found here.

20
 AI Workflow

Data Preparation AI Modeling Simulation & Test Deployment

Data cleansing and Model design and Integration with

Embedded devices
preparation tuning complex systems

Hardware
Human insight System simulation Enterprise systems
accelerated training

Simulation- System verification Edge, cloud,

Interoperability
generated data and validation desktop

I Iteration and Refinement

21 21
AI Workflow

Signal Labeler
Experiment
+
Manager
Audio Labeler

EXPO talks:
Deploying AI on PLC (Plenary)

Image Labeler Cloud Data Workflows for Scientists

+ ONNX and Engineers (DataBricks & AWS)
Video Labeler

22
Biggest Challenge: Use Machine or Deep Learning?

23
 Resources for Learning

▪ Get Free Online (hands-on) Training

24
Resources for Teaching

▪ Teach with MATLAB and Simulink

▪ Teaching Science with MATLAB

▪ Virtual Labs & Projects

▪ Online teaching

▪ Support with developing individual AI courses,

please contact us:
[email protected] or
[email protected]
25
Thank you

© 2021 The MathWorks, Inc. MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See mathworks.com/trademarks
for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.

26