SlideShare uma empresa Scribd logo
1 de 30
Baixar para ler offline
Compressed Sensing Techniques for Sensor Data
using Unsupervised Learning
SONG CUI

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Outline
• Concept of compressed sensing
• Basic theories with interpretations
• Case studies:
Medical imaging (Considered as a sensor network problem)
Wearable electroencephalography (EEG)
Recommendation system

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Concept of compressed sensing
Data compression

Data acquisition

Sensor (limited computational capability, transmission power et al.)

• Feature extraction
• Prediction
• Decision making

Data analytics

Learning algorithms

CPU or GPU (excellent computational capability)
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Compression techniques
• Traditional lossy compression techniques: JPEG, wavelet et al.
• Traditional compression techniques are not efficient for sparse
data.

Non-sparse data

Sparse data
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Why JPEG is ineffective for sparse data
• Discrete cosine transform converts the image (2-D matrix) from

spatial domain to frequency domain in JPEG.
• Insignificant coefficients in frequency domain are discarded in JPEG.
• Sparse data have comparable coefficients in all frequencies.

Keep the first 5 coefficients in
frequency domain and
recover the signal

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data compression

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data compression
• Data compression is through underdetermined linear system.

• It is a dimension reduction process in machine learning.
y = Cx , Where n<m

• Compression matrix C is pre-determined where the compression ratio
is: m/n.
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Comparison with PCA
• The compression matrix C in compressed sensing is pre-determined
with some restrictions.
• PCA is more computational demanding:
Normalizing means and variances in training samples x1 , x2 , x3 , x4 ..., xm .
m
Finding the eigenvalues and eigenvectors for (∑ ( xi xi T )) / m.
i =1
Picking up the first n principle eigenvectors to form C.

• Data compression is efficient and requires limited or no
computational power which favors applications (e. g. mobile sensors)
that have limited power, data storage, transmission, and
computational capabilities.
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data reconstruction

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data reconstruction
Can we fully recover the original signal x from its compressed version
y?
y = Cx, Where n<m

The answer is no in general:

y = x1 + x2

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data reconstruction
• However, the signal can be fully recovered if:
The sparsity of the original signal x is s (the maximum number of
non-zero entry is s).
C must satisfy restricted isometry property (RIP) which means any
s columns in matrix C are independent.

Sparsity is the key!

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Data reconstruction method
• Ill imposed inverse Problem: find the x from y:
y = Cx

ɶ
ɶ
min || x ||0 s.t . y = Cx

If RIP holds and x is sparse
• L1 norm minimization is used in practice:

ɶ
ɶ
min || x ||1 s.t . || y − Cx ||2 ≤ δ

Minimizing

ɶ
ɶ
|| y − Cx ||2 +λ || x ||1

E. Candès and T. Tao, IEEE Trans. Inform. Theory, vol. 51, no. 12, pp. 4203-4215 (2005).

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Why L1 regularization?
Analogy to help
understand L1 and L2
regularizations:
L1 norm minimization

Personal income tax
rate

Original sparse signal

ɶ
ɶ
|| y − Cx ||2 + λ || x || p (p = 1or 2)

Numerical examples are from Prof. W.
K. Ma’s lecture notes
(http://www.ee.cuhk.edu.hk/~wkma).

L2 norm minimization
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Parallel computing for large-scale datasets
Data compression:
Data recovery:
1 CPU

y = Cx

( x1 , y1 ), ( x2 , y2 ), ( x3 , y3 ),...

Multi-thread computing in GPU or CPU clusters
ɶ
ɶ
ɶ
|| y1 − Cx ||2 +λ || x ||1 ⇒ x1

ɶ
ɶ
ɶ
|| y1 − Cx ||2 +λ || x ||1 ⇒ x1
ɶ
ɶ
ɶ
|| y2 − Cx ||2 +λ || x ||1 ⇒ x2

ɶ
ɶ
ɶ
|| y2 − Cx ||2 +λ || x ||1 ⇒ x2

ɶ
ɶ
ɶ
|| y3 − Cx ||2 +λ || x ||1 ⇒ x3
......

ɶ
ɶ
ɶ
|| y3 − Cx ||2 +λ || x ||1 ⇒ x3
….

….

Distributed methods are available for single L1 regularization problem.
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Case study 1: Medical imaging
• Positron

emission tomography
(PET) is capable of measuring
positron-emitting radionuclides.

• It is a medical diagnostic
instrument for oncology,
neuroimaging, and cardiology
applications.

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Background

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Compressed sensing multiplexing circuits
• Data is sparse in spatial domain.
• Data compression is
implemented on PCB circuit
boards with pre-amplifiers,
resistors and capacitors.

P. D. Olcott et al., IEEE NSS-MIC Conference Record
p. 3224 (2011).

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
General applications to wireless sensor network

The figure is from J. Haupt et al., IEEE Signal Processing Mag.
pp. 92 Mar. 2008.

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Case study 2: Wearable electroencephalography (EEG)

Figures are from: AJ. Casson et al., IEEE Eng. Med. Biol. Mag 29:44–56 (2010)

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Trend of wearable EEG
• Long-term monitoring capability is needed and large-scale data can
be generated.
• Data are processed with machine learning algorithms in the
remote end with strong computational capabilities and large
database.
• Wearable data transmission enables device miniaturization and
body area network applications.
• Applications include sleep disorders and augmented cognition.

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Challenges in wearable EEG
• Electrode Technology
• Battery power consumption
Data acquisition
Data transmission

One possible solution: Compressed sensing!

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Compressed sensing for scalp EEG
The data is not sparse in time
domain.
The data has sparse
representations in terms of
other basis functions.

A. M. Abdulghani et al., Med. Biol. Eng. Comput. 50:1137–1145 (2012).

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Case study 3: Recommendation system
Think about how you build up
a recommendation system!
• Matrix factorization methods
rui = u + qiT pu

• Baseline methods
rui = u + bi + bu

User – item rating matrix: M
Item 1
User 1

Item 2

User 2

Item 4

?

5

Item 3

3

2

User 3

4

3

2

User 4

5

4

?

…………

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Interpretation of matrix completion
• We assume that there are
correlations among some users
rating on the same items
(dependency in rows).
• We assume there are
correlations among ratings on
some items from the same user.

User – item rating matrix: M
Item 1
User 1

Item 2

User 2

Item 4

?

5

Item 3

3

2

User 3

4

3

2

User 4

5

4

?

We hypothesis that rank (M) is small!
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Problem formulation
Find a matrix X which fills
up the unknown useritem ratings and satisfy:

User – item rating matrix: M
NP hard
problem

Minimize rank(X), s.t .X iu = M iu (M iu ∈ Ω)
where rank(X) =|| Σ ||0 , X = U ΣV *

Item 1
User 1

Item 2

User 2

Item 4

?

5

Item 3

3

2

User 3

where

r

|| X ||* = ∑ σ i , σ i

is the singular value

3

2

User 4
Minimize || X ||* ,s.t .X iu = M iu (M iu ∈ Ω)

4
5

4

?

E. Candès, et al., Foundations of Computational Mathematics,
vol. 9, pp. 717 (2009).

i =1

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Connection with compressed sensing
• A more robust and computational efficient approach:
Minimize || X − M ||F +λ || X ||*

∑∑ (X
i

u

2

iu

r

− M iu ) + λ ∑ σ i

Previous results:
Minimize || y − Cx ||2 + λ || x ||1

i =1

Interpretation: M is a compressed version
of X!

• The method has been tested on Netflix dataset.
N. Srebro and R. Salakhutdinov., Advances in Neural Information Processing Systems, vol. 23, pp. 2056
(2010).

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Challenges
• Does the hypothesis min rank (X) always match with the truth?
• How to deal with non-uniform sampling (e. g. Some user have much
more ratings than others)?
Weighted regularization

• How to deal with cold start problems in the model?
• How to incorporate additional information such as sex, time
drifting, and geography location in the model?

Personal opinions!
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Summary
• Compressed sensing is a new data compression and recovery method.
• It is effective for sparse data.
Sparse in time domain.
Sparse in frequency domain.
Sparse in other representations.

•It is useful for mobile sensors which has large-scale data transmission,
limited battery power, computational capabilities and requires device
miniaturization.
• It has seen applications in machine learning such as recommendation
systems.

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
Extra materials and links
• Stanford course website: http://statweb.stanford.edu/~candes/stats330/index.shtml
• A very resourceful website: Nult Blanche’s blog: http://nuit-blanche.blogspot.com/p/teachingcompressed-sensing.html covers updates from theories and applications such as MRI and
machine learning.
• You can also add me on Linkedin: http://www.linkedin.com/profile/view?id=71703589 or
contact andycui1@hotmail.com if you want to have further discussions.

COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED

Mais conteúdo relacionado

Mais procurados

Lecture 2 Introduction to digital image
Lecture 2 Introduction to digital imageLecture 2 Introduction to digital image
Lecture 2 Introduction to digital imageVARUN KUMAR
 
Lecture 4 Relationship between pixels
Lecture 4 Relationship between pixelsLecture 4 Relationship between pixels
Lecture 4 Relationship between pixelsVARUN KUMAR
 
Image Compression and Reconstruction Using Artificial Neural Network
Image Compression and Reconstruction Using Artificial Neural NetworkImage Compression and Reconstruction Using Artificial Neural Network
Image Compression and Reconstruction Using Artificial Neural NetworkIRJET Journal
 
Camp IT: Making the World More Efficient Using AI & Machine Learning
Camp IT: Making the World More Efficient Using AI & Machine LearningCamp IT: Making the World More Efficient Using AI & Machine Learning
Camp IT: Making the World More Efficient Using AI & Machine LearningKrzysztof Kowalczyk
 
Neural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmNeural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmMostafa G. M. Mostafa
 
IRJET- Handwritten Decimal Image Compression using Deep Stacked Autoencoder
IRJET- Handwritten Decimal Image Compression using Deep Stacked AutoencoderIRJET- Handwritten Decimal Image Compression using Deep Stacked Autoencoder
IRJET- Handwritten Decimal Image Compression using Deep Stacked AutoencoderIRJET Journal
 
IRJET- Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...
IRJET-  	  Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...IRJET-  	  Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...
IRJET- Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...IRJET Journal
 
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentationKevinTsai67
 
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGE
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGEAPPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGE
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGEcscpconf
 
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...Dr. Amarjeet Singh
 
Developing Document Image Retrieval System
Developing Document Image Retrieval SystemDeveloping Document Image Retrieval System
Developing Document Image Retrieval SystemKonstantinos Zagoris
 
FPGA Implementation of Decision Based Algorithm for Removal of Impulse Noise
FPGA Implementation of Decision Based Algorithm for Removal of Impulse NoiseFPGA Implementation of Decision Based Algorithm for Removal of Impulse Noise
FPGA Implementation of Decision Based Algorithm for Removal of Impulse NoiseIRJET Journal
 
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...ijcseit
 
Dynamic Two-Stage Image Retrieval from Large Multimodal Databases
Dynamic Two-Stage Image Retrieval from Large Multimodal DatabasesDynamic Two-Stage Image Retrieval from Large Multimodal Databases
Dynamic Two-Stage Image Retrieval from Large Multimodal DatabasesKonstantinos Zagoris
 
Matlab Based Image Compression Using Various Algorithm
Matlab Based Image Compression Using Various AlgorithmMatlab Based Image Compression Using Various Algorithm
Matlab Based Image Compression Using Various Algorithmijtsrd
 

Mais procurados (19)

J017426467
J017426467J017426467
J017426467
 
Lecture 2 Introduction to digital image
Lecture 2 Introduction to digital imageLecture 2 Introduction to digital image
Lecture 2 Introduction to digital image
 
Lecture 4 Relationship between pixels
Lecture 4 Relationship between pixelsLecture 4 Relationship between pixels
Lecture 4 Relationship between pixels
 
Image Compression and Reconstruction Using Artificial Neural Network
Image Compression and Reconstruction Using Artificial Neural NetworkImage Compression and Reconstruction Using Artificial Neural Network
Image Compression and Reconstruction Using Artificial Neural Network
 
Camp IT: Making the World More Efficient Using AI & Machine Learning
Camp IT: Making the World More Efficient Using AI & Machine LearningCamp IT: Making the World More Efficient Using AI & Machine Learning
Camp IT: Making the World More Efficient Using AI & Machine Learning
 
Kk3517971799
Kk3517971799Kk3517971799
Kk3517971799
 
Neural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) AlgorithmNeural Networks: Least Mean Square (LSM) Algorithm
Neural Networks: Least Mean Square (LSM) Algorithm
 
IRJET- Handwritten Decimal Image Compression using Deep Stacked Autoencoder
IRJET- Handwritten Decimal Image Compression using Deep Stacked AutoencoderIRJET- Handwritten Decimal Image Compression using Deep Stacked Autoencoder
IRJET- Handwritten Decimal Image Compression using Deep Stacked Autoencoder
 
IRJET- Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...
IRJET-  	  Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...IRJET-  	  Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...
IRJET- Jeevn-Net: Brain Tumor Segmentation using Cascaded U-Net & Overall...
 
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation
2022 03 22_蔡煒俊_u-net_convolutional_networks_for_biomedical_image_segmentation
 
Automatic Image Annotation
Automatic Image AnnotationAutomatic Image Annotation
Automatic Image Annotation
 
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGE
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGEAPPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGE
APPLICATION OF IMAGE FUSION FOR ENHANCING THE QUALITY OF AN IMAGE
 
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...
High Speed Data Exchange Algorithm in Telemedicine with Wavelet based on 4D M...
 
Developing Document Image Retrieval System
Developing Document Image Retrieval SystemDeveloping Document Image Retrieval System
Developing Document Image Retrieval System
 
FPGA Implementation of Decision Based Algorithm for Removal of Impulse Noise
FPGA Implementation of Decision Based Algorithm for Removal of Impulse NoiseFPGA Implementation of Decision Based Algorithm for Removal of Impulse Noise
FPGA Implementation of Decision Based Algorithm for Removal of Impulse Noise
 
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...
MULTI-OBJECTIVE ENERGY EFFICIENT OPTIMIZATION ALGORITHM FOR COVERAGE CONTROL ...
 
Dynamic Two-Stage Image Retrieval from Large Multimodal Databases
Dynamic Two-Stage Image Retrieval from Large Multimodal DatabasesDynamic Two-Stage Image Retrieval from Large Multimodal Databases
Dynamic Two-Stage Image Retrieval from Large Multimodal Databases
 
MultiModal Retrieval Image
MultiModal Retrieval ImageMultiModal Retrieval Image
MultiModal Retrieval Image
 
Matlab Based Image Compression Using Various Algorithm
Matlab Based Image Compression Using Various AlgorithmMatlab Based Image Compression Using Various Algorithm
Matlab Based Image Compression Using Various Algorithm
 

Semelhante a Compressed sensing techniques for sensor data using unsupervised learning

Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...
Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...
Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...Yan-Wei Lee
 
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETSFAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETScsandit
 
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSING
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSINGDATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSING
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSINGcscpconf
 
ai in electrical.pptx
ai in electrical.pptxai in electrical.pptx
ai in electrical.pptxGamerM04
 
An optimized discrete wavelet transform compression technique for image trans...
An optimized discrete wavelet transform compression technique for image trans...An optimized discrete wavelet transform compression technique for image trans...
An optimized discrete wavelet transform compression technique for image trans...IJECEIAES
 
Low Cost Visual Support System for Challenged People
Low Cost Visual Support System for Challenged PeopleLow Cost Visual Support System for Challenged People
Low Cost Visual Support System for Challenged PeopleChristo Ananth
 
Neural Networks and Deep Learning for Physicists
Neural Networks and Deep Learning for PhysicistsNeural Networks and Deep Learning for Physicists
Neural Networks and Deep Learning for PhysicistsHéloïse Nonne
 
Segmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding CellSegmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding CellSeunghyun Hwang
 
Mx net image segmentation to predict and diagnose the cardiac diseases karp...
Mx net image segmentation to predict and diagnose the cardiac diseases   karp...Mx net image segmentation to predict and diagnose the cardiac diseases   karp...
Mx net image segmentation to predict and diagnose the cardiac diseases karp...KannanRamasamy25
 
Sparse Sampling in Digital Image Processing
Sparse Sampling in Digital Image ProcessingSparse Sampling in Digital Image Processing
Sparse Sampling in Digital Image ProcessingEswar Publications
 
Alternative Computing
Alternative ComputingAlternative Computing
Alternative ComputingShayshab Azad
 
FACE RECOGNITION USING ELM-LRF
FACE RECOGNITION USING ELM-LRFFACE RECOGNITION USING ELM-LRF
FACE RECOGNITION USING ELM-LRFAras Masood
 
A benchmark dataset to evaluate sensor displacement in activity recognition
A benchmark dataset to evaluate sensor displacement in activity recognitionA benchmark dataset to evaluate sensor displacement in activity recognition
A benchmark dataset to evaluate sensor displacement in activity recognitionOresti Banos
 
Various Applications of Compressive Sensing in Digital Image Processing: A Su...
Various Applications of Compressive Sensing in Digital Image Processing: A Su...Various Applications of Compressive Sensing in Digital Image Processing: A Su...
Various Applications of Compressive Sensing in Digital Image Processing: A Su...IRJET Journal
 
Computing Just What You Need: Online Data Analysis and Reduction at Extreme ...
Computing Just What You Need: Online Data Analysis and Reduction  at Extreme ...Computing Just What You Need: Online Data Analysis and Reduction  at Extreme ...
Computing Just What You Need: Online Data Analysis and Reduction at Extreme ...Ian Foster
 

Semelhante a Compressed sensing techniques for sensor data using unsupervised learning (20)

Seminar nov2017
Seminar nov2017Seminar nov2017
Seminar nov2017
 
Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...
Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...
Endobronchial Ultrasound Image Diagnosis Using Convolutional Neural Network (...
 
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETSFAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
FAST ALGORITHMS FOR UNSUPERVISED LEARNING IN LARGE DATA SETS
 
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSING
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSINGDATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSING
DATA COMPRESSION USING NEURAL NETWORKS IN BIO-MEDICAL SIGNAL PROCESSING
 
Computer Design Concepts for Machine Learning
Computer Design Concepts for Machine LearningComputer Design Concepts for Machine Learning
Computer Design Concepts for Machine Learning
 
ai in electrical.pptx
ai in electrical.pptxai in electrical.pptx
ai in electrical.pptx
 
An optimized discrete wavelet transform compression technique for image trans...
An optimized discrete wavelet transform compression technique for image trans...An optimized discrete wavelet transform compression technique for image trans...
An optimized discrete wavelet transform compression technique for image trans...
 
Low Cost Visual Support System for Challenged People
Low Cost Visual Support System for Challenged PeopleLow Cost Visual Support System for Challenged People
Low Cost Visual Support System for Challenged People
 
40120140507007
4012014050700740120140507007
40120140507007
 
40120140507007
4012014050700740120140507007
40120140507007
 
Ml ppt at
Ml ppt atMl ppt at
Ml ppt at
 
Neural Networks and Deep Learning for Physicists
Neural Networks and Deep Learning for PhysicistsNeural Networks and Deep Learning for Physicists
Neural Networks and Deep Learning for Physicists
 
Segmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding CellSegmenting Medical MRI via Recurrent Decoding Cell
Segmenting Medical MRI via Recurrent Decoding Cell
 
Mx net image segmentation to predict and diagnose the cardiac diseases karp...
Mx net image segmentation to predict and diagnose the cardiac diseases   karp...Mx net image segmentation to predict and diagnose the cardiac diseases   karp...
Mx net image segmentation to predict and diagnose the cardiac diseases karp...
 
Sparse Sampling in Digital Image Processing
Sparse Sampling in Digital Image ProcessingSparse Sampling in Digital Image Processing
Sparse Sampling in Digital Image Processing
 
Alternative Computing
Alternative ComputingAlternative Computing
Alternative Computing
 
FACE RECOGNITION USING ELM-LRF
FACE RECOGNITION USING ELM-LRFFACE RECOGNITION USING ELM-LRF
FACE RECOGNITION USING ELM-LRF
 
A benchmark dataset to evaluate sensor displacement in activity recognition
A benchmark dataset to evaluate sensor displacement in activity recognitionA benchmark dataset to evaluate sensor displacement in activity recognition
A benchmark dataset to evaluate sensor displacement in activity recognition
 
Various Applications of Compressive Sensing in Digital Image Processing: A Su...
Various Applications of Compressive Sensing in Digital Image Processing: A Su...Various Applications of Compressive Sensing in Digital Image Processing: A Su...
Various Applications of Compressive Sensing in Digital Image Processing: A Su...
 
Computing Just What You Need: Online Data Analysis and Reduction at Extreme ...
Computing Just What You Need: Online Data Analysis and Reduction  at Extreme ...Computing Just What You Need: Online Data Analysis and Reduction  at Extreme ...
Computing Just What You Need: Online Data Analysis and Reduction at Extreme ...
 

Último

Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)wesley chun
 
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdfUnderstanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdfUK Journal
 
[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdf[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdfhans926745
 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...Neo4j
 
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024The Digital Insurer
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerThousandEyes
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationSafe Software
 
A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?Igalia
 
Slack Application Development 101 Slides
Slack Application Development 101 SlidesSlack Application Development 101 Slides
Slack Application Development 101 Slidespraypatel2
 
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptxHampshireHUG
 
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfThe Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfEnterprise Knowledge
 
Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024The Digital Insurer
 
Advantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your BusinessAdvantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your BusinessPixlogix Infotech
 
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptx
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptxEIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptx
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptxEarley Information Science
 
How to convert PDF to text with Nanonets
How to convert PDF to text with NanonetsHow to convert PDF to text with Nanonets
How to convert PDF to text with Nanonetsnaman860154
 
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...Igalia
 
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking MenDelhi Call girls
 
GenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationGenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationMichael W. Hawkins
 
CNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of ServiceCNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of Servicegiselly40
 
Boost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityBoost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityPrincipled Technologies
 

Último (20)

Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
 
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdfUnderstanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
 
[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdf[2024]Digital Global Overview Report 2024 Meltwater.pdf
[2024]Digital Global Overview Report 2024 Meltwater.pdf
 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
 
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?
 
Slack Application Development 101 Slides
Slack Application Development 101 SlidesSlack Application Development 101 Slides
Slack Application Development 101 Slides
 
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
 
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdfThe Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
 
Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024
 
Advantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your BusinessAdvantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your Business
 
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptx
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptxEIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptx
EIS-Webinar-Prompt-Knowledge-Eng-2024-04-08.pptx
 
How to convert PDF to text with Nanonets
How to convert PDF to text with NanonetsHow to convert PDF to text with Nanonets
How to convert PDF to text with Nanonets
 
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
 
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men08448380779 Call Girls In Greater Kailash - I Women Seeking Men
08448380779 Call Girls In Greater Kailash - I Women Seeking Men
 
GenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationGenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day Presentation
 
CNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of ServiceCNv6 Instructor Chapter 6 Quality of Service
CNv6 Instructor Chapter 6 Quality of Service
 
Boost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityBoost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivity
 

Compressed sensing techniques for sensor data using unsupervised learning

  • 1. Compressed Sensing Techniques for Sensor Data using Unsupervised Learning SONG CUI COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 2. Outline • Concept of compressed sensing • Basic theories with interpretations • Case studies: Medical imaging (Considered as a sensor network problem) Wearable electroencephalography (EEG) Recommendation system COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 3. Concept of compressed sensing Data compression Data acquisition Sensor (limited computational capability, transmission power et al.) • Feature extraction • Prediction • Decision making Data analytics Learning algorithms CPU or GPU (excellent computational capability) COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 4. Compression techniques • Traditional lossy compression techniques: JPEG, wavelet et al. • Traditional compression techniques are not efficient for sparse data. Non-sparse data Sparse data COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 5. Why JPEG is ineffective for sparse data • Discrete cosine transform converts the image (2-D matrix) from spatial domain to frequency domain in JPEG. • Insignificant coefficients in frequency domain are discarded in JPEG. • Sparse data have comparable coefficients in all frequencies. Keep the first 5 coefficients in frequency domain and recover the signal COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 6. Data compression COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 7. Data compression • Data compression is through underdetermined linear system. • It is a dimension reduction process in machine learning. y = Cx , Where n<m • Compression matrix C is pre-determined where the compression ratio is: m/n. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 8. Comparison with PCA • The compression matrix C in compressed sensing is pre-determined with some restrictions. • PCA is more computational demanding: Normalizing means and variances in training samples x1 , x2 , x3 , x4 ..., xm . m Finding the eigenvalues and eigenvectors for (∑ ( xi xi T )) / m. i =1 Picking up the first n principle eigenvectors to form C. • Data compression is efficient and requires limited or no computational power which favors applications (e. g. mobile sensors) that have limited power, data storage, transmission, and computational capabilities. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 9. Data reconstruction COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 10. Data reconstruction Can we fully recover the original signal x from its compressed version y? y = Cx, Where n<m The answer is no in general: y = x1 + x2 COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 11. Data reconstruction • However, the signal can be fully recovered if: The sparsity of the original signal x is s (the maximum number of non-zero entry is s). C must satisfy restricted isometry property (RIP) which means any s columns in matrix C are independent. Sparsity is the key! COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 12. Data reconstruction method • Ill imposed inverse Problem: find the x from y: y = Cx ɶ ɶ min || x ||0 s.t . y = Cx If RIP holds and x is sparse • L1 norm minimization is used in practice: ɶ ɶ min || x ||1 s.t . || y − Cx ||2 ≤ δ Minimizing ɶ ɶ || y − Cx ||2 +λ || x ||1 E. Candès and T. Tao, IEEE Trans. Inform. Theory, vol. 51, no. 12, pp. 4203-4215 (2005). COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 13. Why L1 regularization? Analogy to help understand L1 and L2 regularizations: L1 norm minimization Personal income tax rate Original sparse signal ɶ ɶ || y − Cx ||2 + λ || x || p (p = 1or 2) Numerical examples are from Prof. W. K. Ma’s lecture notes (http://www.ee.cuhk.edu.hk/~wkma). L2 norm minimization COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 14. Parallel computing for large-scale datasets Data compression: Data recovery: 1 CPU y = Cx ( x1 , y1 ), ( x2 , y2 ), ( x3 , y3 ),... Multi-thread computing in GPU or CPU clusters ɶ ɶ ɶ || y1 − Cx ||2 +λ || x ||1 ⇒ x1 ɶ ɶ ɶ || y1 − Cx ||2 +λ || x ||1 ⇒ x1 ɶ ɶ ɶ || y2 − Cx ||2 +λ || x ||1 ⇒ x2 ɶ ɶ ɶ || y2 − Cx ||2 +λ || x ||1 ⇒ x2 ɶ ɶ ɶ || y3 − Cx ||2 +λ || x ||1 ⇒ x3 ...... ɶ ɶ ɶ || y3 − Cx ||2 +λ || x ||1 ⇒ x3 …. …. Distributed methods are available for single L1 regularization problem. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 15. Case study 1: Medical imaging • Positron emission tomography (PET) is capable of measuring positron-emitting radionuclides. • It is a medical diagnostic instrument for oncology, neuroimaging, and cardiology applications. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 17. Compressed sensing multiplexing circuits • Data is sparse in spatial domain. • Data compression is implemented on PCB circuit boards with pre-amplifiers, resistors and capacitors. P. D. Olcott et al., IEEE NSS-MIC Conference Record p. 3224 (2011). COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 18. General applications to wireless sensor network The figure is from J. Haupt et al., IEEE Signal Processing Mag. pp. 92 Mar. 2008. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 19. Case study 2: Wearable electroencephalography (EEG) Figures are from: AJ. Casson et al., IEEE Eng. Med. Biol. Mag 29:44–56 (2010) COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 20. Trend of wearable EEG • Long-term monitoring capability is needed and large-scale data can be generated. • Data are processed with machine learning algorithms in the remote end with strong computational capabilities and large database. • Wearable data transmission enables device miniaturization and body area network applications. • Applications include sleep disorders and augmented cognition. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 21. Challenges in wearable EEG • Electrode Technology • Battery power consumption Data acquisition Data transmission One possible solution: Compressed sensing! COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 22. Compressed sensing for scalp EEG The data is not sparse in time domain. The data has sparse representations in terms of other basis functions. A. M. Abdulghani et al., Med. Biol. Eng. Comput. 50:1137–1145 (2012). COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 23. Case study 3: Recommendation system Think about how you build up a recommendation system! • Matrix factorization methods rui = u + qiT pu • Baseline methods rui = u + bi + bu User – item rating matrix: M Item 1 User 1 Item 2 User 2 Item 4 ? 5 Item 3 3 2 User 3 4 3 2 User 4 5 4 ? ………… COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 24. Interpretation of matrix completion • We assume that there are correlations among some users rating on the same items (dependency in rows). • We assume there are correlations among ratings on some items from the same user. User – item rating matrix: M Item 1 User 1 Item 2 User 2 Item 4 ? 5 Item 3 3 2 User 3 4 3 2 User 4 5 4 ? We hypothesis that rank (M) is small! COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 25. Problem formulation Find a matrix X which fills up the unknown useritem ratings and satisfy: User – item rating matrix: M NP hard problem Minimize rank(X), s.t .X iu = M iu (M iu ∈ Ω) where rank(X) =|| Σ ||0 , X = U ΣV * Item 1 User 1 Item 2 User 2 Item 4 ? 5 Item 3 3 2 User 3 where r || X ||* = ∑ σ i , σ i is the singular value 3 2 User 4 Minimize || X ||* ,s.t .X iu = M iu (M iu ∈ Ω) 4 5 4 ? E. Candès, et al., Foundations of Computational Mathematics, vol. 9, pp. 717 (2009). i =1 COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 26. Connection with compressed sensing • A more robust and computational efficient approach: Minimize || X − M ||F +λ || X ||* ∑∑ (X i u 2 iu r − M iu ) + λ ∑ σ i Previous results: Minimize || y − Cx ||2 + λ || x ||1 i =1 Interpretation: M is a compressed version of X! • The method has been tested on Netflix dataset. N. Srebro and R. Salakhutdinov., Advances in Neural Information Processing Systems, vol. 23, pp. 2056 (2010). COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 27. Challenges • Does the hypothesis min rank (X) always match with the truth? • How to deal with non-uniform sampling (e. g. Some user have much more ratings than others)? Weighted regularization • How to deal with cold start problems in the model? • How to incorporate additional information such as sex, time drifting, and geography location in the model? Personal opinions! COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 28. Summary • Compressed sensing is a new data compression and recovery method. • It is effective for sparse data. Sparse in time domain. Sparse in frequency domain. Sparse in other representations. •It is useful for mobile sensors which has large-scale data transmission, limited battery power, computational capabilities and requires device miniaturization. • It has seen applications in machine learning such as recommendation systems. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 29. Extra materials and links • Stanford course website: http://statweb.stanford.edu/~candes/stats330/index.shtml • A very resourceful website: Nult Blanche’s blog: http://nuit-blanche.blogspot.com/p/teachingcompressed-sensing.html covers updates from theories and applications such as MRI and machine learning. • You can also add me on Linkedin: http://www.linkedin.com/profile/view?id=71703589 or contact andycui1@hotmail.com if you want to have further discussions. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED
  • 30. COPYRIGHT©2013 SONG CUI. ALL RIGHTS RESERVED