ECCV2010: feature learning for image classification, part 0

Andrew Ng

Feature learning for image

classificationKai Yu and Andrew Ng

Andrew Ng

Computer vision is hard

Andrew Ng

Machine learning and feature representations

Input

Input spaceMotorbikes“Non”-Motorbikes

Learningalgorithm

pixel 1

pixel 1

pixel 2

Andrew Ng

Machine learning and feature representations

Input

Input space Feature spaceMotorbikes“Non”-Motorbikes

Feature representation

Learningalgorithm

pixel 1 “wheel”

handle

wheel

Andrew Ng

How is computer perception done?

Image Low-levelvision features

Recognition

Low-level statefeatures Action

Helicopter

Audio Low-levelaudio features

Speakeridentification

Object detection

Audio classification

Helicopter control

Andrew Ng

Learning representations

Sensor Learningalgorithm

Feature Representation

Andrew Ng

Computer vision features

SIFT Spin image

HoG RIFT

Textons GLOH

Andrew Ng

Audio features

ZCR

Spectrogram MFCC

RolloffFlux

Problems of hand-tuned features1. Needs expert knowledge

2. Time-consuming and expensive3. Does not generalize to other domains

Andrew Ng

Computer vision is more than pictures

Camera array

3d range scans (flash lidar) Audio

Can we automatically learn good feature representations?

Images

Thermal Infrared

Video

Andrew Ng

Learning representations

Sensor Learningalgorithm

Feature Representation

Andrew Ng

Sensor representation in the brain

[BrainPort; Martinez et al; Roe et al.]

Seeing with your tongueHuman echolocation (sonar)

Auditory cortex learns to see.

Auditory Cortex

Andrew Ng

Unsupervised feature learning

Find a better way to represent images than pixels.

Andrew Ng

The goal of Unsupervised Feature Learning

Unlabeled images

Learningalgorithm

Feature representation

Andrew Ng

Tutorial outline

1. Current methods.

2. Sparse coding for feature learning.

— Break —

3. Advanced classification.

4. Advanced concepts & deep learning.