Opensense @ LCAV · 2014. 1. 23. · Efficientdatacollec-onin par-cipatorysensing...

Preview:

Citation preview

Efficient  data  collec-on  in  par-cipatory  sensing

Runwei  Zhang,  

joint  work  with  Zichong  Chen,LCAV,EPFL

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Contents

‣ Introduc<on

‣ Our  2-­‐step  adap<ve  algorithm

‣ Simula<on  results

‣ Problem  formula<on

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

‣Sensors  equipped

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

‣Sensors  equipped

Cameras

GPS

Microphones

Compass

Gyroscope

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

Thermometer

Hygrometer

Barometer

CO,CO2,NO,O3

‣Sensors  equipped

Cameras

GPS

Microphones

Compass

Gyroscope

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

Thermometer

Hygrometer

Barometer

CO,CO2,NO,O3

‣ How  it  works‣Sensors  equipped

Cameras

GPS

Microphones

Compass

Gyroscope

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

Thermometer

Hygrometer

Barometer

CO,CO2,NO,O3

‣ How  it  works‣Sensors  equipped

Cameras

GPS

Microphones

Compass

Gyroscope

Mobile  sensor

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

‣ Achieve  certain  overall  objec<ve‣ Reconstruct  noise/temperature  map

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

‣ Cost  per  sensor  reading‣ Energy  consump<on  

‣ Network  bandwidth‣ Privacy  leakage

‣ Achieve  certain  overall  objec<ve‣ Reconstruct  noise/temperature  map

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Background  in  par-cipatory  sensing

‣ Cost  per  sensor  reading‣ Energy  consump<on  

‣ Network  bandwidth‣ Privacy  leakage

‣ Achieve  certain  overall  objec<ve‣ Reconstruct  noise/temperature  map

Reduce  unnecessary  sensor  readings?

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

All  sensors  always  ac<ve

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

All  sensors  always  ac<ve

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

All  sensors  always  ac<ve

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

All  sensors  always  ac<ve ANer  turning  off

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Why  can  we  turn  off  some  sensors?

All  sensors  always  ac<ve ANer  turning  off

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings‣ Maintain  coverage  (small  reconstruc6on  error)

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings‣ Maintain  coverage  (small  reconstruc6on  error)

‣ Achieve  fairness

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings‣ Maintain  coverage  (small  reconstruc6on  error)

‣ Achieve  fairness

‣ Challenge

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings‣ Maintain  coverage  (small  reconstruc6on  error)

‣ Achieve  fairness

‣ Challenge‣ Sensors’  moving  trajectories  unknown

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Objec-ve  &  Challenge

‣ Objec<ve‣ Exploit  the  spa6al  correla6ons‣ Reduce  data  readings‣ Maintain  coverage  (small  reconstruc6on  error)

‣ Achieve  fairness

‣ Challenge‣ Sensors’  moving  trajectories  unknown‣ Decisions  should  be  made  online

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Contents

‣ Introduc<on

‣ Our  2-­‐step  adap<ve  algorithm

‣ Simula<on  results

‣ Problem  formula<on

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

‣ Maximum  coverage  problem

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

‣ Maximum  coverage  problem

Sensor Coverage  range

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

‣ Maximum  coverage  problem

Sensor Coverage  range

Reduce

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

‣ Maximum  coverage  problem

Sensor Coverage  range

Objec<ve:  Maximize  number  of  covered  cells

Constraint:  Number  of  ac<ve  sensor

Reduce

Friday, May 24, 13

s6

s7

y7

y8

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

s1

s2

s3

s4

s5

y1y2

y3

y4

y5

y6

Sensors Cells

Friday, May 24, 13

s6

s7

y7

y8

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

s1

s2

s3

s4

s5

y1y2

y3

y4

y5

y6

Sensors Cells

A B:  A  can  sense  B

Friday, May 24, 13

Max

X

ej2E

yj

s.t.

X

i

si k

X

ej2Ai

si � yj

0 yi 1, si 2 {0, 1}.s6

s7

y7

y8

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

s1

s2

s3

s4

s5

y1y2

y3

y4

y5

y6

Sensors Cells

A B:  A  can  sense  B

Friday, May 24, 13

Max

X

ej2E

yj

s.t.

X

i

si k

X

ej2Ai

si � yj

0 yi 1, si 2 {0, 1}.s6

s7

y7

y8

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

s1

s2

s3

s4

s5

y1y2

y3

y4

y5

y6

Sensors Cells

A B:  A  can  sense  B

k=5?

Friday, May 24, 13

Max

X

ej2E

yj

s.t.

X

i

si k

X

ej2Ai

si � yj

0 yi 1, si 2 {0, 1}.s6

s7

y7

y8

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  sta-c  sensor  networks

s1

s2

s3

s4

s5

y1y2

y3

y4

y5

y6

Sensors Cells

A B:  A  can  sense  B

k=5?

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenarioAt  6me  slot  n=1:

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

Reduce

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

Reduce

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

At  6me  slot  n=2:

Reduce

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

1

2

3

4 56

At  6me  slot  n=2:

Reduce

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

1

2

3

4 56

At  6me  slot  n=2:

Reduce

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

1

2

3

4 56

At  6me  slot  n=2:

Reduce

Reduce

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

12

3

45

6

At  6me  slot  n=1:

1

2

3

4 56

At  6me  slot  n=2:

Reduce

Reduce

12

3

45

6

1

2

3

4 56

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

At  6me  slot  n=1

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

12

3

45

6

At  6me  slot  n=1

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

Sensor

12

3

45

6

At  6me  slot  n=1

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

Sensor

12

3

45

6

At  6me  slot  n=1

b(n) = A(n) · s(n)

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

A B:  A  resides  in  B

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

Sensor

12

3

45

6

At  6me  slot  n=1 y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells...

...

B C:  B  can  sense  C

b(n) = A(n) · s(n)

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

A B:  A  resides  in  B

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

Sensor

12

3

45

6

At  6me  slot  n=1 y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells...

...

B C:  B  can  sense  C

b(n) = A(n) · s(n)

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

A B:  A  resides  in  B

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

At  6me  slot  n=2

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

1

2

3

4 56

At  6me  slot  n=2

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

1

2

3

4 56

At  6me  slot  n=2s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

s(n)2

Sensors

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

1

2

3

4 56

At  6me  slot  n=2s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

s(n)2

Sensors

...

A B:  A  resides  in  B

...b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

b(n) = A(n) · s(n)

Sensor  loca6ons

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

1

2

3

4 56

At  6me  slot  n=2s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

s(n)2

Sensors

...

B C:  B  can  sense  C

...y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells

...

A B:  A  resides  in  B

...b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

b(n) = A(n) · s(n)

Sensor  loca6ons

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

The  par-cipatory  sensing  scenario

Sensor

1

2

3

4 56

At  6me  slot  n=2s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

s(n)2

Sensors

...

B C:  B  can  sense  C

...y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells

...

A B:  A  resides  in  B

...b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

b(n) = A(n) · s(n)

Sensor  loca6ons

Friday, May 24, 13

Max

0

@lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j ,� lim sup

n!1V (✓(n)

)

1

A

s.t. eT · s(n) k, 8n 2 N+

A(n) · s(n) = b(n), 8n 2 N+

X

ej2Ai

s(n)i � y(n)j , 8n 2 N+

✓(n)=

1

n

nX

t=1

s(t), 8n 2 N+

0 y(n)i 1, s(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Op-mal  schedule  with  known  trajectories

13

Friday, May 24, 13

Max

0

@lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j ,� lim sup

n!1V (✓(n)

)

1

A

s.t. eT · s(n) k, 8n 2 N+

A(n) · s(n) = b(n), 8n 2 N+

X

ej2Ai

s(n)i � y(n)j , 8n 2 N+

✓(n)=

1

n

nX

t=1

s(t), 8n 2 N+

0 y(n)i 1, s(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Op-mal  schedule  with  known  trajectories

13

coverage

Friday, May 24, 13

Max

0

@lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j ,� lim sup

n!1V (✓(n)

)

1

A

s.t. eT · s(n) k, 8n 2 N+

A(n) · s(n) = b(n), 8n 2 N+

X

ej2Ai

s(n)i � y(n)j , 8n 2 N+

✓(n)=

1

n

nX

t=1

s(t), 8n 2 N+

0 y(n)i 1, s(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Op-mal  schedule  with  known  trajectories

13

coverage fairness

Friday, May 24, 13

Max

0

@lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j ,� lim sup

n!1V (✓(n)

)

1

A

s.t. eT · s(n) k, 8n 2 N+

A(n) · s(n) = b(n), 8n 2 N+

X

ej2Ai

s(n)i � y(n)j , 8n 2 N+

✓(n)=

1

n

nX

t=1

s(t), 8n 2 N+

0 y(n)i 1, s(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Op-mal  schedule  with  known  trajectories

13

coverage fairness

Number  of  ac6ve  sensors

Friday, May 24, 13

Max

0

@lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j ,� lim sup

n!1V (✓(n)

)

1

A

s.t. eT · s(n) k, 8n 2 N+

A(n) · s(n) = b(n), 8n 2 N+

X

ej2Ai

s(n)i � y(n)j , 8n 2 N+

✓(n)=

1

n

nX

t=1

s(t), 8n 2 N+

0 y(n)i 1, s(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Op-mal  schedule  with  known  trajectories

13

coverage fairness

Number  of  ac6ve  sensors

Adap6ve  algorithms?

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Contents

‣ Introduc<on

‣ Our  2-­‐step  adap<ve  algorithm

‣ Simula<on  results

‣ Problem  formula<on

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

Friday, May 24, 13

Sensor

12

3

45

6

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

Friday, May 24, 13

Sensor

12

3

45

6

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

Friday, May 24, 13

Sensor

12

3

45

6

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells...

...

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

Friday, May 24, 13

Sensor

12

3

45

6

Max lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j

s.t. eT · b(n) k, 8n 2 N+

X

ej2Ai

b(n)i � y(n)j , 8n 2 N+

A(n) · e � b(n), 8n 2 N+

0 y(n)i 1, 8n 2 N+

b(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells...

...

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

Friday, May 24, 13

Sensor

12

3

45

6

Max lim inf

n!1

1

n

nX

t=1

X

ej2E

y(n)j

s.t. eT · b(n) k, 8n 2 N+

X

ej2Ai

b(n)i � y(n)j , 8n 2 N+

A(n) · e � b(n), 8n 2 N+

0 y(n)i 1, 8n 2 N+

b(n)i 2 {0, 1}, 8n 2 N+.

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  1:  choose  subsets  of  cells

y(n)1

y(n)2

y(n)3

y(n)4

y(n)5

y(n)6

y(n)7

Cells...

...

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...{A(n)}n2N+{b(n)}n2N+ is  decided  by  

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Sensor

12

3

45

6

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Sensor

12

3

45

6

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...A(n)

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Sensor

12

3

45

6

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...A(n)

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Sensor

12

3

45

6

Min lim supn!1

V (✓(n))

s.t. A(n) · s(n) = b(n), 8n 2 N+

✓(n) =1

n

nX

t=1

s(t), 8n 2 N+

s(n) 2 {0, 1}M , 8n 2 N+.

ORIGINAL-­‐OPT:s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...A(n)

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Step  2:  choose  ac-ve  sensors  within  cells

Sensor

12

3

45

6

Min lim supn!1

V (✓(n))

s.t. A(n) · s(n) = b(n), 8n 2 N+

✓(n) =1

n

nX

t=1

s(t), 8n 2 N+

s(n) 2 {0, 1}M , 8n 2 N+.

ORIGINAL-­‐OPT:s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...A(n)

Adap6ve  algorithms?

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

At  6me  slot  n

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

At  6me  slot  n

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

(2)  Within  each  chosen  cell,choose  the  ac6ve  sensor

with  the  smallest

m

✓(n�1)m

At  6me  slot  n

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

(2)  Within  each  chosen  cell,choose  the  ac6ve  sensor

with  the  smallest

m

✓(n�1)m

At  6me  slot  n

Friday, May 24, 13

✓(n�1)3 < ✓(n�1)

2

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

(2)  Within  each  chosen  cell,choose  the  ac6ve  sensor

with  the  smallest

m

✓(n�1)m

At  6me  slot  n

Friday, May 24, 13

✓(n�1)3 < ✓(n�1)

2

✓(n�1)6 < ✓(n�1)

5

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

(2)  Within  each  chosen  cell,choose  the  ac6ve  sensor

with  the  smallest

m

✓(n�1)m

At  6me  slot  n

Friday, May 24, 13

✓(n�1)3 < ✓(n�1)

2

✓(n�1)6 < ✓(n�1)

5

LCAV  –  AudioVisual  Communica1ons  Laboratory

“Let  the  laziest  serve”  (LLS)

s(n)1

s(n)3

s(n)4

s(n)5

s(n)6

Sensors

s(n)2

b(n)1

b(n)2

b(n)3

b(n)4

b(n)5

b(n)6

b(n)7

Sensor  loca6ons...

...

✓(n�1)m =

1

n� 1

n�1X

t=1

s(t)m , 8m

(1)  Calculate  the  frac6on  of  ac6ve  sensing  6mes  for  each  sensor

(2)  Within  each  chosen  cell,choose  the  ac6ve  sensor

with  the  smallest

m

✓(n�1)m

At  6me  slot  n

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

@V (✓)

@✓i>

@V (✓)

@✓j✓i > ✓j

When  the  objec6ve  func6on  fulfills    

if

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

@V (✓)

@✓i>

@V (✓)

@✓j✓i > ✓j

When  the  objec6ve  func6on  fulfills    

if nega6ve  entropy  Lp-­‐norm

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

@V (✓)

@✓i>

@V (✓)

@✓j✓i > ✓j

When  the  objec6ve  func6on  fulfills    

if

s(n)LLS  let                be  op6mal  w.r.t  the  linearized  problem  

nega6ve  entropy  Lp-­‐norm

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

@V (✓)

@✓i>

@V (✓)

@✓j✓i > ✓j

When  the  objec6ve  func6on  fulfills    

if

s(n)LLS  let                be  op6mal  w.r.t  the  linearized  problem  

nega6ve  entropy  Lp-­‐norm

Min r>V (✓(n�1)) · xs.t. A

(n) · x = b

(n),

x 2 [0, 1]M .

ORIGINAL-­‐LINEARIZED:

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Gradient  upda-ng  direc-on

@V (✓)

@✓i>

@V (✓)

@✓j✓i > ✓j

When  the  objec6ve  func6on  fulfills    

if

s(n)LLS  let                be  op6mal  w.r.t  the  linearized  problem  

nega6ve  entropy  Lp-­‐norm

Min r>V (✓(n�1)) · xs.t. A

(n) · x = b

(n),

x 2 [0, 1]M .

ORIGINAL-­‐LINEARIZED:

integer  constraints  are  changed  into  non-­‐integer  ones

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1 Sensor  2

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1 Sensor  2

...

...

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1 Sensor  2

A(n) 2 {A[1],A[2],A[3] · · · }

Trajectories  of  all  sensorsare  drawn  from  a  large  markov  chain  

...

...

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1 Sensor  2

A(n) 2 {A[1],A[2],A[3] · · · }

Trajectories  of  all  sensorsare  drawn  from  a  large  markov  chain  

Min V (⇥)

s.t. A[l] · q[l] = b[l], l = 1, · · ·L,

⇥ =LX

l=1

Pl · q[l], l = 1, · · ·L

q[l] 2 {0, 1}M , l = 1, · · ·L.

MARKOV-­‐OPT:

...

...

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chainsSensor  1 Sensor  2

A(n) 2 {A[1],A[2],A[3] · · · }

Trajectories  of  all  sensorsare  drawn  from  a  large  markov  chain  

Min V (⇥)

s.t. A[l] · q[l] = b[l], l = 1, · · ·L,

⇥ =LX

l=1

Pl · q[l], l = 1, · · ·L

q[l] 2 {0, 1}M , l = 1, · · ·L.

MARKOV-­‐OPT:

...

...

Pl        is  the  sta6onary  distribu6on A(n) = A[l]

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

n1 n2 n3

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

n1 n2 n3

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Ev(j)l =Pl

P3, The  expected  6mes  that

during  the  recurrent  round          obeys  sta6onaryj

A(n) = A[l]

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

n1 n2 n3

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

n 2 [nj�1 + 1, nj ] A(n) = A[l]At  each  6me                                                        with                                  ,                  is  op6mal  w.r.ts(n)

Ev(j)l =Pl

P3, The  expected  6mes  that

during  the  recurrent  round          obeys  sta6onaryj

A(n) = A[l]

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

n1 n2 n3

Min r>V (✓(n�1)) · xs.t. A

[l] · x(n) = b

[l],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

n 2 [nj�1 + 1, nj ] A(n) = A[l]At  each  6me                                                        with                                  ,                  is  op6mal  w.r.ts(n)

Ev(j)l =Pl

P3, The  expected  6mes  that

during  the  recurrent  round          obeys  sta6onaryj

A(n) = A[l]

Friday, May 24, 13

A(n) : A[3],A[1],A[2],A[4],A[2],A[3],A[1],A[4],A[3], · · ·

n : 1, 2, 3, 4, 5, 6, 7, 8,

n1 n2 n3

Min r>V (✓(n�1)) · xs.t. A

[l] · x(n) = b

[l],

x 2 [0, 1]M .

⇡Min r>V (✓(nj�1)) · xs.t. A

[l] · x(n) = b

[l],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

n 2 [nj�1 + 1, nj ] A(n) = A[l]At  each  6me                                                        with                                  ,                  is  op6mal  w.r.ts(n)

Ev(j)l =Pl

P3, The  expected  6mes  that

during  the  recurrent  round          obeys  sta6onaryj

A(n) = A[l]

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

P3/P3

Min r>V (✓(nj�1)) · xs.t. A

[3] · x = b

[3],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

P3/P3

Min r>V (✓(nj�1)) · xs.t. A

[3] · x = b

[3],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

...

...

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

P3/P3

Min r>V (✓(nj�1)) · xs.t. A

[3] · x = b

[3],

x 2 [0, 1]M .

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

...

...

In  all,  is  op6mal  w.r.t  the  linearized  version  of  MARKOV-­‐OPT

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

P3/P3

Min r>V (✓(nj�1)) · xs.t. A

[3] · x = b

[3],

x 2 [0, 1]M .

Min rTV (✓(nj�1)) ·LX

l=1

Plq[l]

s.t. A[l] · q[l] = b[l], l = 1, · · ·L,q[l] 2 {0, 1}M , l = 1, · · ·L.

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

...

...

In  all,  is  op6mal  w.r.t  the  linearized  version  of  MARKOV-­‐OPT

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

P1/P3Ev(j)l

Min r>V (✓(nj�1)) · xs.t. A

[1] · x = b

[1],

x 2 [0, 1]M .

P2/P3

Min r>V (✓(nj�1)) · xs.t. A

[2] · x = b

[2],

x 2 [0, 1]M .

P3/P3

Min r>V (✓(nj�1)) · xs.t. A

[3] · x = b

[3],

x 2 [0, 1]M .

Min rTV (✓(nj�1)) ·LX

l=1

Plq[l]

s.t. A[l] · q[l] = b[l], l = 1, · · ·L,q[l] 2 {0, 1}M , l = 1, · · ·L.

LCAV  –  AudioVisual  Communica1ons  Laboratory

When  trajectories  are  from  markov  chains

LLS  is  asympto6cally  op6mal  

...

...

In  all,  is  op6mal  w.r.t  the  linearized  version  of  MARKOV-­‐OPT

During  each  recurrent  round                                                          ,  n 2 [nj�1 + 1, nj ]

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

‣ Totally  memoryless  and  adap<ve

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

‣ Totally  memoryless  and  adap<ve

‣ Does  not  know  trajectories  in  advance

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

‣ Totally  memoryless  and  adap<ve

‣ Does  not  know  trajectories  in  advance

‣ Achieves  op<mal  coverage  in  step  1

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

‣ Totally  memoryless  and  adap<ve

‣ Does  not  know  trajectories  in  advance

‣ Achieves  op<mal  coverage  in  step  1

‣ Achieves  asympto<cally  op<mal  fairness  in  step  2  

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Summary  of  the  2-­‐step  adap-ve  algorithm

‣ Totally  memoryless  and  adap<ve

‣ Does  not  know  trajectories  in  advance

‣ Achieves  op<mal  coverage  in  step  1

‣ Achieves  asympto<cally  op<mal  fairness  in  step  2  

‣ Not  necessarily  op<mal  when  consider  step  1  and  step  2  jointly.

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Contents

‣ Introduc<on

‣ Our  2-­‐step  adap<ve  algorithm

‣ Simula<on  results

‣ Problem  formula<on

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Simula-ons

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Simula-ons

‣ Bus  trajectories‣ Synthesized  from  Lausanne  bus  <metables

‣ 327  buses,  400  bus  stops,  18km*6km

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Simula-ons

‣ Bus  trajectories‣ Synthesized  from  Lausanne  bus  <metables

‣ 327  buses,  400  bus  stops,  18km*6km‣ Taxis  trajectories‣ Real  data  from  San  Francisco  

‣ 500  taxis,  30  days

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Simula-ons

‣ Bus  trajectories‣ Synthesized  from  Lausanne  bus  <metables

‣ 327  buses,  400  bus  stops,  18km*6km‣ Taxis  trajectories‣ Real  data  from  San  Francisco  

‣ 500  taxis,  30  days‣ We  have  not  applied  the  reduc<on  in  step  1.

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Synthesizing  Lausanne  bus  trajectories

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Synthesizing  Lausanne  bus  trajectories

1 km

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Synthesizing  Lausanne  bus  trajectories

1 km

‣  Acquire  all  the  GPS  loca<ons  of  400  bus  stops

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Synthesizing  Lausanne  bus  trajectories

1 km

‣  Interpolate  with  <metables  of  37  bus  lines

‣  Acquire  all  the  GPS  loca<ons  of  400  bus  stops

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Synthesizing  Lausanne  bus  trajectories

1 km

‣  Interpolate  with  <metables  of  37  bus  lines

‣  Acquire  all  the  GPS  loca<ons  of  400  bus  stops

‣  Add  GPS  noise,  Poisson  delay  noise

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

‣ The  obtained  sensing  schedule

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

‣ The  obtained  sensing  schedule

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

Buses traveling to remote areas

‣ The  obtained  sensing  schedule

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

Buses traveling to remote areas

Weekends

‣ The  obtained  sensing  schedule

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

Buses traveling to remote areas

Weekends

‣ The  obtained  sensing  schedule

‣ Energy  savings

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

0.5 1 1.5 2 2.5 30

10

20

30

40

AI side length/km

Ener

gy sa

ving

s

Buses traveling to remote areas

Weekends

‣ The  obtained  sensing  schedule

‣ Energy  savings

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency

time / 10 minutes

bus

ID

100 200 300 400 500 600 700

50

100

150

200

250

300

0.5 1 1.5 2 2.5 30

10

20

30

40

AI side length/km

Ener

gy sa

ving

s

Buses traveling to remote areas

Weekends

‣ The  obtained  sensing  schedule

‣ Energy  savings

Friday, May 24, 13

−50 0 50 100 150 200 250 300 3500

0.05

0.1

0.15

0.2

Node idThe

ener

gy c

ost t

o th

e ba

selin

e

RandomOur algorithmOptimal

LCAV  –  AudioVisual  Communica1ons  Laboratory

Fairness

Sorted bus id

Active time ratio

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Using  San  Francisco  taxis  data

‣ Ac<ve  sensing  loca<ons  accumulated  with  <me

Without our scheme With our scheme

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Energy  efficiency  &  fairness

Friday, May 24, 13

LCAV  –  AudioVisual  Communica1ons  Laboratory

Future  works

‣ Validate  the  reconstruc<on  error‣ Consider  temporal  correla<ons

x

y

t

Friday, May 24, 13

Recommended