msc projects at bt innovate - keith briggs: :...

24
BT Research Wireless networks Projects MSc projects at BT Innovate Keith Briggs [email protected] Mobility Research Centre, BT Innovate http://keithbriggs.info University of York MSc talk 2009-02-11 1515

Upload: others

Post on 27-Jul-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

MSc projects at BT Innovate

Keith Briggs [email protected]

Mobility Research Centre, BT Innovatehttp://keithbriggs.info

University of York MSc talk 2009-02-11 1515

Page 2: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

BT Research - Adastral Park

Page 3: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

BT Research centres

• Broadband Centre

• Foresight Centre

• IT futures research Centre

• Intelligent systems Centre

• Mobility Centre

• Networks Centre

• Pervasive ICT Centre

• Security Centre

• Asian Research Centre

• Disruptive lab at MIT

Page 4: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Wireless networks

0

64

0

65

0

83

0

93

0

102

0

104

0

141

0

150

1

18

1

28

130 1

60

1

116

1

144

3 623

64

365 3

73

3

101

3

114

3

121

3

150

4

23

4

29

4

77

4

81

4 87499

5

10

5

21

5

24

5

27

531

5 325

33

5

38

5535

112

5

118

5

120

6

7

6

8

6 156

40

6

44

6

52

6

54

6

56

6

74

6 826

86

6

88

6

121

6

1327

8

7

17

7

40

7 527

54

7

55

7567

57

7

58

7

69

7

70

7

71

7

74

7

86

7

88

7132

7

146

7

148

7

153

7

154

7

155

7

156

815

8

40

8

44

8

52

8

54

8

56

8

74

8

82

8868

88

8

121

8

132

8

155

8

156

9979

136

9

137

10

24

102710

31

10

32

10

38

10

53

10

112

10120

11 122

122612

51

12

63

12

68

12

76

12

80

12

84

12124 12

129

12149

13

16

13

41

13

42

13

43

13

50

13

92

13

133

14

41

14

42

14

50

14

92

14

122

1412814

133

15

40

15

44

15

54

1582

15

86

15

132

1643

1722 17

52

17 5517

56

1757175817

69

17 7017

71

17

72

17

102

17

138

17

142

17146

17

148

17

153

17

154

17 15517 156

18

30

18

62

19

35

19

37

19

38

19

131

20

44

20

74

20

88

20

91

20

95

20 9820

103

20

132

20134

20

140

20

145

21

31

21

32

213321

36

21

53

21

112

21118

22 5722

69

22

72

22100 22

125

22127 22

130

22

138

22

142

22146

22

153

23

29

23

51

2368

23

76

23

77

23

87

23

94

2399

23

124

23

149 24

27

24

31

24

32

24

53

24

112

24

120

2526 25

59

25

63

25

64

25

65

25 7325

78

25

84

25

114

25

123

26

63

26

68

26

76

26

78

26

80

26

84

26124 26

129

26

149

27

31

27

32

27

38

27

53

27

112

2712029

39

2981 29

89

29

99

3059 30

60

30

73

30123

31 3231

37

31

38

3153

31112

31

118

31

120

32

33

32

38

3253 32

112

32

118

32

120

33 3433

36

33

53

33

112

33118 34

36

35

37

35

131

35

152

36

118

36

135

37

38

37

144 38

53

38

112

38 11838

120

39

51

39

81

40

44

40

52

4054 40

56

40

74

40

82

4086

40

88

40

103

40

132

40

155

40

156

4142 41

46

41

50

41

92

41

128

4113342

46

42

50

42

92

42

128

42 133

44

52

44

54

4474

44

82

44

86

4488

44

103

44132 44

134

44

155

44

156

45

61

45

113

46

67

46

106

46

133

47

49

47

66

48

50

48

92

48

119

49

66

49

108

49

109

49

110

49

119

509250

119

50

128

50

133

51

80

51

124

51 12951

149

52

54

52

55

5256

52

57

52

58

52

70

52

74

52

86

52

88

52

132

52

148

52

154

52

155

52

156

53

112

53

118

53

120

54

55

54

56

54

70

54

74

5486

54

88

54

132

54

155

54

15655

56

5557

5558

5569

557055

71

55

74

55

88

55

132

55

142

55

146

55

148

55

153

55

154

55 15555156

56

57

56

58

56

69

56

70

56

71

56

74

56

86

56

88

56 13256

146

56

148

56

153

56

154

56

155

56

156575857

69

5770

57

71

57

72

57

102

57

138

57

142

57146 57

148

57

153

57

154

57 15557 15658

69

5870

58

71

58

72

5874

5888

58

102

58

138

58

142

58146 58

148

58

153

58

154

58 15558 156

59

60

59

63

59

73

59

78

59

114

59 123

60

73

60

123

61

113

62

101

62114

62

121

63

64

63

65

63

73

637863

84

63

114

6465

64

78

64

84

64

93

64101

64

104

64

114

64

150

65

78

65

84

65

93

65

101

65

104

65

114

65

150

66

108

66

109

66

110

67

98

67

106

67

134

67

140

67

145

68

76

6884

68

87

68

94

68

124

68

129

68

149

6970

69

71

69

72

69

138

69142 69

146

69

148

69

153

69154

69

155

69

15670

71

70

74

70

88

70

132

70

142

70

146

70

148

70

153

70

154

70 15570156

71

74

71

88

71

142

71

146

7114871

154

71

155

71

156

72

125

72

138

72

142

72

146

72

151

73

78

73

114

73

123

74

86

748874

103

74132

74

148

74

154

74

155

74156

75

90

75

91

75

95

7596 75

147

76

84

7687 76

94

76

124

76

149

77

85

77

87

77

99

77

126

78

84

78

114

78

123

79

115

80

115

80

129

80

149

81

89

81

99

82

86

83

93

83

102

83

104

83

138

83

141

83

150

83

153

84

93

84

94

84

129

84

149

85

126

86

88

86

121

86

132

86

155

86

156

87

94

87

99

87

124

87

149

88

103

88132

88

148

88

154

88

155

88156

9091 90

95

90

96

90

98

90

139

90

140

90

145

91

95

91

96

91

140

91

145

92

119

92

128

92

133

93

94

93

104

93

141

93

150

95

98

95

106

95

134

95

139

95

140

95

145

97

136

97

137

98

103

98106

98134

98

139

98

140

98

145

99

124

100

125

100127100

143

101

114

101

121

101

150

102

104

102130 102

138

102

141

102

146

102

153

103

106

103

132

103

134

103

140

103

145

104

141

104

150

105

139

105

147

106134 106

139

106

140

106

145

108109108110109110

111117

112

118

112

120

114

123

114

150

115

136

115

137

116

144

122

128

124149

125

127

125

143

127

143

128

133

129

149

130

138

130

141

131

152

132

134

132

154

132

155

132

156

134

140

134

145

136137

138

142

138

146

138

153

139

140

139

145140145

141

150

142

146

142

148

142

153

142 154

146

148

146

153

146

154

146155

146 156

148

154

148

155

148

156

154

155

154

156155156

Page 5: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Page 6: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Page 7: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Page 8: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Page 9: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Graph concepts

1

0

2

3

4

5

• clique - a complete subgraph

• clique number ω - the number of nodes ina largest clique

• coloring - an assignment of colors tonodes in which no neighbours have thesame color

• chromatic number χ - the number ofcolors in a coloring with a minimalnumber of colors

• edge coloring - an assignment of colors toedges in which no edges with a commonendpoint have the same color

Page 10: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Graph concepts

1

0

2

3

4

5

• clique - a complete subgraph• clique number ω - the number of nodes in

a largest clique

• coloring - an assignment of colors tonodes in which no neighbours have thesame color

• chromatic number χ - the number ofcolors in a coloring with a minimalnumber of colors

• edge coloring - an assignment of colors toedges in which no edges with a commonendpoint have the same color

Page 11: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Graph concepts

1

0

2

3

4

5

• clique - a complete subgraph• clique number ω - the number of nodes in

a largest clique• coloring - an assignment of colors to

nodes in which no neighbours have thesame color

• chromatic number χ - the number ofcolors in a coloring with a minimalnumber of colors

• edge coloring - an assignment of colors toedges in which no edges with a commonendpoint have the same color

Page 12: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Graph concepts

1

0

2

3

4

5

• clique - a complete subgraph• clique number ω - the number of nodes in

a largest clique• coloring - an assignment of colors to

nodes in which no neighbours have thesame color

• chromatic number χ - the number ofcolors in a coloring with a minimalnumber of colors

• edge coloring - an assignment of colors toedges in which no edges with a commonendpoint have the same color

Page 13: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Graph concepts

1

0

2

3

4

5

• clique - a complete subgraph• clique number ω - the number of nodes in

a largest clique• coloring - an assignment of colors to

nodes in which no neighbours have thesame color

• chromatic number χ - the number ofcolors in a coloring with a minimalnumber of colors

• edge coloring - an assignment of colors toedges in which no edges with a commonendpoint have the same color

Page 14: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Bridge design by Costas Skarakis from 2008

14.6182

6.81

38

9.7555

14.6182

31.6678

6.3336

11.3

563

4.5425

14

.61

82

14

.61

82

1.9

51

1

6.3336

8.4447

13

.86

17

44.940

2

31.6

678

10.6

149

31.3212

13

.86

17

1.1

94

6

6.3336

6.0556

5.9729

44.940

2

2.7498

9.047912

.077

8

31.6678

6.0

38

9

3.6088

6.0389

12

.66

71

3.6088

6.0389

31.6678

6.0

38

9

9.0479

31.3212

12.0778

44.9402

2.74

98

5.9729

44.9402

31.6678

6.3336

10.6

149

6.0556

13

.86

17

13

.86

17

1.1

94

6

6.3336

8.4447

14

.61

82

14.618

2

31.6

678

11.3

563

14

.61

82

1.9

51

1

6.3336

4.5425

9.7555

14.618

2

6.8138

Page 15: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Quantifying TV whitespace for cognitive radio operation

Cognitive radio is being intensively studied for opportunistic accessto the so-called TV whitespace: large portions of the VHF/UHFTV bands will become available after the digital switchover. Usingaccurate digital TV (DTV) coverage maps together with adatabase of DTV transmitters, we would like to develop anaccurate methodology for identifying TVWS frequencies at anygiven location in the United Kingdom.A preliminary study has already proved the worth of this approach;we wish to take it further and make the results much more precise.The project largely involves computational geometry: such ideas asconvex hulls and point-in-polygon tests. These ideas are well worthknowing about for other applications. This is a project with directapplicability to a currently hot topic in telecoms, as well as usingfundamental geometric algorithms that need to be well-designed tomake the whole computation feasible.

Page 16: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

The maths of MIMO

MIMO refers to multiple antenna systems currently beinginvestigated for use in wireless data networks. There is a lot of nicemaths used in this topic, especially for the decoding - maximumlikelihood estimation, singular-value decomposition, etc.I think a good project would be to study the basics and write anexposition emphasizing the mathematics that the method dependsupon. There is plenty of engineering expertise in my research labwhich we could draw on. An open-ended research question wouldbe to what extent exact maximum likelihood estimation can beused, or is its computational complexity too high?

Page 17: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

SAT

The SAT (satisfiability) problem is a very basic problem in logic: todetermine whether a given Boolean function has an input whichmakes the output true. There are many good algorithms nowavailable to solve SAT problems, and many practical problems areequivalent to, and can be translated into SAT.This project would make a comparison of available solvers, andstudy their application to some practical problems in optimization,scheduling, and graph coloring. The fun part would be writingprograms to translate these into SAT!

Page 18: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Maximum-likelihood period estimation

This project would be to study the paper (only 4 pages!):Maximum-likelihood period estimation from sparse, noisy timingdata, by Robby McKilliam and I. Vaughan L. Clarksonand write a program to implement the method. This is animportant problem area with applications in telecoms.

Page 19: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Railway route planning allowing for delays

A previous project looked at statistical models in the delay todeparture time of trains. An interesting result was obtained - thatthe delays follow a q-exponential distribution.This project would develop algorithms and software for planningrail journeys, which takes account of the delay distribution. Theaim would be to be able to specify a specific set of trains to takefrom A to B, where the final arrival time has a probabilisticguarantee, such as “the probability of being more than 10 minuteslate is less than 3%”.

Page 20: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Chromatic number of geometric random graphs

Geometric random graphs are models of wireless communicationnetworks. The chromatic number tells us how many frequencychannels are needed to operate the network without interference.I am currently doing some work on how the chromatic number ofpurely random graphs (without the geometric structure) dependson the density (number of links per node). I find some veryinteresting results - namely, if the graph is big enough, thechromatic number is almost surely determined. This project wouldbe to examine the analogue of this behaviour for geometric randomgraphs. It will not be exactly true, but might be in someappropriate limiting cases. There is already quite a lot of work inthis area which wold have to be taken into account.

Page 21: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Semidefinite programming and graph theory

Semidefinite programming (SDP) is a kind of generalized linearprogramming. Many practical optimization problems can beformulated as SDPs. There has been rapid progress in the last 20years on understanding the theory of SDP, but the development ofSDP software that is convenient to use has lagged behind. Goodsoftware is now available. This project would use sdpsol for smalltest problems, and then move to DSDP for large problems. Theaim of the project is to see what performance one can achieve onreal graph theory problems of the types that come up in networkmodelling. Such problems include the Lovasz number (a lowerbound for the chromatic number), the maximal stable set problem,and the maximum cut problem. An important outcome of theproject would be a determination of how the computation timescales with problem size. Geometric optimization problems couldalso be studied.

Page 22: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Fast random selection

Consider the problem: a large number N of objects are presentedto us one by one, and we wish to either: Select exactly somespecified number n6N of them Select each with some specifiedprobability p.The problem becomes hard to solve efficiently when p is small,because to generate a random number which usually results in arejection is inefficient. It would be better to skip over a block ofitems. Ways to do these were proposed by Vitter in ACM TransMath Soft 13, 58 (1987).This project would investigate these methods and check theirefficiency in practice. There are applications to the generation oflarge random graphs.

Page 23: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Random sampling of set partitions

A partition of [n]=0, 1, ..., n−1 is an assignment each element to asubset called a block, without regard to the labelling of the block, ororder of elements within a block. The block structure of a partition maybe indicated by listing the elements and separating blocks with a symbolsuch as —. Thus for [5], 01—2—34 and 2—10—43 are the samepartition. If there are k blocks, we speak of a k-partition. The number ofk-partitions of [n] is given by the Stirling number of the second kindS(n, k), which can be computed from the recursionsS(n, k)=S(n−1, k−1)+kS(n−1, k), S(n, 1)=S(n, n)=1. The totalnumber of partitions of [n] is given by the sum over k of S(n, k), andthis is called a Bell number B(k).There are algorithms for generating all partitions in a Gray code order,due to Ehrlich and Ruskey. In such a sequence, only one element movesto a new block on each step. Beyond about n=15, there are too manypartitions to allow us to construct all of them in reasonable time.

Thus, for large sets, simulations must rely on a uniform random sampling

procedure. This project would investigate such methods and produce a

well-designed C library for use in other projects.

Page 24: MSc projects at BT Innovate - Keith Briggs: : homekeithbriggs.info/documents/York_MSc_2009-02-11.pdf2009/02/11  · This project is inspired by the paper HyperLoglog: the analysis

BT Research Wireless networks Projects

Fast counting

This project is inspired by the paper HyperLoglog: the analysis of anear-optimal cardinality estimation algorithm of Flajolet et al. Howdo we estimate (rapidly, and with minimal storage) the number ofdistinct items in a very large set (or data stream)? In other words,the problem concerns estimating the cardinality of a multiset.Flajolet et al. previously developed their loglog algorithm, and Iworked on an efficient C implementation of this. They have nowimproved this with the hyperloglog algorithm, and this projectwould be to implement this and compare its performance inpractice with alternatives.The algorithm should have practical applications in informatics; forexample, counting the number of different packet types in anetwork.