Developing Deep

Learning Algorithms
using MATLAB
David Willingham

© 2015 The MathWorks, Inc.

1
New MATLAB framework makes deep learning
easy and accessible

2
Object Recognition using Deep Learning
ACCESS

LEARN

INTEGRATE

Training Millions of images from 1000 different categories

(using GPU)
Prediction Real-time object recognition using a webcam connected to a
laptop

3
 What is Deep Learning ?
Cat
Deep learning is a type of machine learning that learns tasks
directly from data

Dog
Learned Features Car
Dog

Bird
… Cat
Bird

Car

5
 What is Deep Learning ?
Cat

Dog
Learned Features Car
Dog

Bird
… Cat
Bird

Data Task
Car

6
Why is Deep Learning So Popular Now?

Human
Accuracy

Source: ILSVRC Top-5 Error on ImageNet 7

Deep Learning Enablers

Acceleration with GPUs

Massive sets of labeled data

Availability of state of the art models from experts

8
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

 Handle and label large sets of images

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks

 Access and use models from experts

9
Agenda

Image classification using pre-trained network

Transfer learning to classify new objects

Locate & classify objects in images and video

10
Agenda

Image classification using pre-trained network

Transfer learning to classify new objects

Locate & classify objects in images and video

11
Image Classification Using Pre-trained Network (Video)

12
Convolutional Neural Networks

13
Visualize Deep Learning Features

15
Agenda

Image classification using pre-trained network

Transfer learning to classify new objects

Locate & classify objects in images and video

16
Why Train a New Model ?

o Models from research do not work on your data

o Pre-trained model not available for your data type

o Improve results by creating a model specific to your problem

17
 Two Approaches for Deep Learning
1. Train a Deep Neural Network from Scratch

2. Fine-tune a pre-trained model (transfer learning)

18
Why Perform Transfer Learning

 Requires less data and training time

 Reference models (like AlexNet, VGG-16, VGG-19) are great feature

extractors

 Leverage best network types from top researchers

19
Example: Classify Vehicles With Transfer Learning

Convolution

Convolution

Convolution

Convolution

Connected
Activation

Activation

Activation

Activation
Pooling

Pooling

Pooling

Pooling

Layers
Fully
… 1000 Category
Classifier

AlexNet

car
suv
5 Category
AlexNet pickup Classifier
van
truck

New Data 20
Transfer Learning to Classify New Objects

21
New MATLAB framework makes deep learning
easy and accessible

22
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

imageDS = imageDatastore(dir)
Easily manage large sets of images
 Handle and label large sets of images

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks

 Access and use models from experts

23
MATLAB makes Deep Learning Easy and Accessible

Training modes supported:

Learn about new MATLAB capabilities to
Auto Select
GPU
 Handle and label large sets of images Multi GPU (local)
Multi GPU (cluster)

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks

 Access and use models from experts

Acceleration with Multiple GPUs

24
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

Training Accuracy Plot
 Handle and label large sets of images

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks Deep Dream Network Activations

 Access and use models from experts

25
Feature Visualization
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

 Handle and label large sets of images

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks

Curated Set of Pretrained Models

 Access and use models from experts
Access Models with 1-line of MATLAB Code
Net1 = alexnet
Net2 = vgg16
Net3 = vgg19

26
Regression Support for Deep Learning

Classification vs. Regression

 Classification – outputs categories/labels
 Regression – outputs numbers

Supported by new regression layer:

routputlayer = regressionLayer('Name','routput')

Example predict facial key-points:

27
Agenda

Image classification using pre-trained network

Transfer learning to classify new objects

Locate & classify objects in images and video

28
Is Object Recognition/Classification Enough ?

Car Label for entire image

Car ? SUV? Truck?

29
Object Detection – Locate and Classify Object

TRUCK CAR
SUV

30
Goal: Create Object Detector to Locate Vehicles

Step 1: Label / Crop data

Step 2: Train detector

Step 3: Use detector

31
Video: Object Detection using Faster R-CNN

32
Label Images with MATLAB

33
Labeling Videos with MATLAB

34
New MATLAB framework makes deep learning
easy and accessible

35
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

 Handle and label large sets of images

 Accelerate deep learning with GPUs

Image Labeler
 Visualize and debug deep neural networks

 Access and use models from experts

Video Labeler 36
Object Detection Frameworks in MATLAB

Machine Learning
1. Cascade Object Detector
2. Aggregate Channel Features (ACF)

Same labels , train any detector.

Deep Learning
1. R-CNN
2. Fast R-CNN
3. Faster R-CNN

39
MATLAB makes Deep Learning Easy and Accessible

Learn about new MATLAB capabilities to

 Handle and label large sets of images

 Accelerate deep learning with GPUs

 Visualize and debug deep neural networks

 Access and use models from experts

40
Thank You

41