Tweet
Tweet

Slide 1

Slide 1 text

Marco De Nadai Into the city A Multi-Disciplinary Investigation of Urban Life

Slide 2

Slide 2 text

By 2010 urban population began to exceed the rural one UN population division, 2010 Urban Rural Population:

Slide 3

Slide 3 text

By 2010 urban population began to exceed the rural one UN population division, 2010 90% of global gross added value comes from urban areas (2006) Urban Rural Population:

Slide 4

Slide 4 text

Cities, very difficult to explain 4 COMPLEX Not only agglomeration of residents, factories, shops • Millions of individuals • Continuously evolving A small change generates a cascading throughout

Slide 5

Slide 5 text

The role of data mining and machine learning 5 DATA MINING • Inexpensive way to understand mechanisms; • New stimulus to social research; MACHINE LEARNING • New tools to expand the notion of what is predictable; Shmueli, Galit. "To explain or to predict?." Statistical science 25, no. 3 (2010): 289-310.

Slide 6

Slide 6 text

Predict deprivation and per-capita income solely relying on mobility diversity and social diversity 6 Pappalardo, L., Vanhoof, M., Gabrielli, L., Smoreda, Z., Pedreschi, D., & Giannotti, F. (2016). An analytical framework to nowcast well-being using mobile phone data. International Journal of Data Science and Analytics

Slide 7

Slide 7 text

Predict poverty from satellite imagery (75% variation of economic outcomes) 7 Jean, N., Burke, M., Xie, M., Davis, W. M., Lobell, D. B., & Ermon, S. (2016). Combining satellite imagery and machine learning to predict poverty. Science

Slide 8

Slide 8 text

Predict crime rates from POIs, mobility, demographics 8 Wang, H., Kifer, D., Graif, C., & Li, Z. (2016). Crime rate inference with big data. In Proceedings of the 22nd ACM SIGKDD.

Slide 9

Slide 9 text

Understand underlying mechanisms of a city 9 Arbesman, Samuel. Overcomplicated: Technology at the Limits of Comprehension. Penguin, 2016. Describe

Slide 10

Slide 10 text

Understand underlying mechanisms of a city 10 Arbesman, Samuel. Overcomplicated: Technology at the Limits of Comprehension. Penguin, 2016. Describe Predict

Slide 11

Slide 11 text

Understand underlying mechanisms of a city 11 Arbesman, Samuel. Overcomplicated: Technology at the Limits of Comprehension. Penguin, 2016. Describe Predict Generate

Slide 12

Slide 12 text

Understand underlying mechanisms of a city 12 Arbesman, Samuel. Overcomplicated: Technology at the Limits of Comprehension. Penguin, 2016. Describe Predict Generate

Slide 13

Slide 13 text

Multi-modal understanding 13

Slide 14

Slide 14 text

Multi-modal understanding 14

Slide 15

Slide 15 text

A multitude of dimensions and aspects! 15 Urban vitality 1 DONE ON-GOING / FUTURE

Slide 16

Slide 16 text

A multitude of dimensions and aspects! 16 Urban vitality 1 Security perception 2 DONE ON-GOING / FUTURE

Slide 17

Slide 17 text

A multitude of dimensions and aspects! 17 Crime 3 Urban vitality 1 Security perception 2 DONE ON-GOING / FUTURE

Slide 18

Slide 18 text

A multitude of dimensions and aspects! 18 Crime 3 Urban vitality 1 Security perception 2 Structural design 4 DONE ON-GOING / FUTURE

Slide 19

Slide 19 text

A multitude of dimensions and aspects! 19 DESCRIBE & PREDICT Crime 3 Urban vitality 1 Security perception 2 Structural design 4 DONE ON-GOING / FUTURE

Slide 20

Slide 20 text

A multitude of dimensions and aspects! 20 DESCRIBE & PREDICT PREDICT & GENERATE Crime 3 Urban vitality 1 Security perception 2 Structural design 4 DONE ON-GOING / FUTURE

Slide 21

Slide 21 text

Urban Vitality OBJECTIVE CHARACTERISTICS Q: Can we describe and predict vitality from urban physical characteristics?

Slide 22

Slide 22 text

The theory: Jane Jacobs One of the most influential books in city planning • Death: caused by the elimination of pedestrian activity • Life: created by a vital urban fabric at all times of the day 22 Jacobs, Jane. The death and life of great American cities. Vintage, 1961 2 1 3 URBAN VITALITY 4

Slide 23

Slide 23 text

The theory: Jane Jacobs Diversity => Urban vitality There are 4 diversity conditions Operationalize the theory 23 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY 2 1 3 URBAN VITALITY 4

Slide 24

Slide 24 text

The theory: Jane Jacobs 2+ primary uses (contemporarily) 24 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY 2 1 3 URBAN VITALITY 4

Slide 25

Slide 25 text

The theory: Jane Jacobs 2+ primary uses (contemporarily) 25 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY For district : % = − ( %,+ log (%,+ ) log || +∈5 %,+: % square footage of land use : {residential, commercial, recreation} 1 0 2 1 3 URBAN VITALITY 4

Slide 26

Slide 26 text

The theory: Jane Jacobs City blocks should be small/short 26 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY BLOCKS 2 1 3 URBAN VITALITY 4

Slide 27

Slide 27 text

The theory: Jane Jacobs City blocks should be small/short 27 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY BLOCKS For district : |% | % 2 1 3 URBAN VITALITY 4

Slide 28

Slide 28 text

The theory: Jane Jacobs Buildings mixed (age and types) 28 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY 2 1 3 URBAN VITALITY 4

Slide 29

Slide 29 text

The theory: Jane Jacobs Buildings mixed (age and types) 29 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY Standard deviation of building ages 2 1 3 URBAN VITALITY 4

Slide 30

Slide 30 text

The theory: Jane Jacobs Concentration of people and enterprises 30 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY 2 1 3 URBAN VITALITY 4

Slide 31

Slide 31 text

The theory: Jane Jacobs Concentration of people and enterprises 31 LAND USE SMALL BLOCKS AGED BUILDINGS DENSITY Population density: |% | % 2 1 3 URBAN VITALITY 4

Slide 32

Slide 32 text

Vitality Anonymized Mobile phone Internet activity as a proxy for urban vitality 32 1 % || ( | % | G∈H : set of hours (180 days x 24h) : area of district 2 1 3 URBAN VITALITY 4

Slide 33

Slide 33 text

The Death and Life of Great Italian Cities 33 DATA • Web and Open data (physical characteristics) • Mobile phone data (proxy for vitality) MODEL • Fit with Ordinary Least Squares regression (OLS) • Predict with Cross-validated OLS De Nadai, Marco, et al. "The Death and Life of Great Italian Cities: A Mobile Phone Data Perspective." WWW, 2016.

Slide 34

Slide 34 text

The log Linear Regression 34 Vitality (Ground truth) = K K + N N + ⋯ + P P + 2 1 3 URBAN VITALITY 4

Slide 35

Slide 35 text

The log Linear Regression 35 Vitality (Ground truth) Land Use Mix = K K + N N + ⋯ + P P + 2 1 3 URBAN VITALITY 4

Slide 36

Slide 36 text

The log Linear Regression 36 Vitality (Ground truth) Land Use Mix Employment density = K K + N N + ⋯ + P P + 2 1 3 URBAN VITALITY 4

Slide 37

Slide 37 text

37 Urban metric Beta coefficient Employmentdensity 0.434*** Intersections density 0.191*** Housing types 0.185*** Closeness highways -0.102*** 3rd places x closenesshighways 0.07** Closeness parks x closeness highways -0.07*** adj − RN 0.77 *** p-value < 0.001; ** p-value < 0.01; Describe urban vitality 2 1 3 URBAN VITALITY 4

Slide 38

Slide 38 text

Result 38 De Nadai, Marco, et al. "The Death and Life of Great Italian Cities: A Mobile Phone Data Perspective." WWW, 2016. Physical characteristics describe and predict urban vitality 2 1 3 URBAN VITALITY 4

Slide 39

Slide 39 text

Security perception SUBJECTIVE CHARACTERISTICS Q: how can new sources of data and deep learning models help to link urban visual perception and the behavior of people?

Slide 40

Slide 40 text

Broken windows theory • City mismanagement • Dirty places • Poor infrastructure Lead to misbehavior => Crime 40 Wilson, James Q., and George L. Kelling. "Broken windows." Critical issues in policing: Contemporary readings (1982): 395- 407. 2 1 3 SECURITY PERCEPTION 4

Slide 41

Slide 41 text

Are Safer Looking Neighborhoods More Lively? 41 DATA • Web data (Google Street View imagery) • Mobile phone data (proxy for vitality) MODEL • Convolutional Neural Network (CNN) • Spatial Ordinary Least Squares De Nadai, Marco, et al. "Are Safer Looking Neighborhoods More Lively?: A Multimodal Investigation into Urban Life." ACM MM, 2016. 1 SECURITY PERCEPTION 4 2 3

Slide 42

Slide 42 text

Place Pulse 2.0 42 1 SECURITY PERCEPTION 4 Salesses, P., Schechtner, K., & Hidalgo, C. A. (2013). The collaborative image of the city: mapping the inequality of urban perception. PloS one 2 3

Slide 43

Slide 43 text

Security perception prediction • Learning safety perception, predict in Rome and Milan • Standard architecture AlexNet CNN • Trained on Places205* • Data Augmentation 43 * B. Zhou, A. Lapedriza, J. Xiao, A. Torralba, and A. Oliva. “Learning Deep Features for Scene Recognition using Places Database.” NIPS, 2014. 1 SECURITY PERCEPTION 4 2 3

Slide 44

Slide 44 text

Describe safety perception 44 REGRESSION SAFETY PERCEPTION LIVELINESS 1 SECURITY PERCEPTION 4 2 3

Slide 45

Slide 45 text

Security perception <-> presence of people 45 Urban metric Beta coefficient Population density Employees density Deprivation Distance from the center Safety appearance adj − RN 0.91 ** p-value < 0.001; * p-value < 0.01; 1 SECURITY PERCEPTION 4 2 3

Slide 46

Slide 46 text

46 Urban metric Beta coefficient Population density 0.155** Employees density 0.328** Deprivation -0.022 Distance from the center -0.257** Safety appearance 0.105** adj − RN 0.91 ** p-value < 0.001; * p-value < 0.01; Security perception <-> presence of people 1 SECURITY PERCEPTION 4 2 3

Slide 47

Slide 47 text

47 Urban metric Beta coefficient % of women (from census) 0.001 Deprivation -0.005 Distance from the center -0.003 Safety appearance 0.020** adj − RN 0.65 ** p-value < 0.001; * p-value < 0.01; Security perception <-> women around 1 SECURITY PERCEPTION 4 2 3

Slide 48

Slide 48 text

Visual elements for security perception 48 HIGH SAFETY PERCEPTION RANDOMLY OBSCURE PART OF THE IMAGE AND PREDICT 1 SECURITY PERCEPTION 4 2 3

Slide 49

Slide 49 text

Visual elements for security perception 49 HIGH SAFETY PERCEPTION RANDOMLY OBSCURE PART OF THE IMAGE AND PREDICT 1 SECURITY PERCEPTION 4 2 3

Slide 50

Slide 50 text

Result 50 De Nadai, Marco, et al. "Are Safer Looking Neighborhoods More Lively?: A Multimodal Investigation into Urban Life." ACM MM, 2016. Security perception can predict Presence of people 3 4 1 SECURITY PERCEPTION 2

Slide 51

Slide 51 text

Crime OBJECTIVE CHARACTERISTICS Q: can we describe how physical characteristics influence crime? Q: can crime be predicted from the urban physical characteristics?

Slide 52

Slide 52 text

Crime theory 52 CRIME Felson, M. (1994). Crime and everyday life: Insight and implications for society. Thousand Oaks, CA: Pine. 1 CRIME 4 2 3

Slide 53

Slide 53 text

MOTIVATED OFFENDER Crime theory 53 CRIME Felson, M. (1994). Crime and everyday life: Insight and implications for society. Thousand Oaks, CA: Pine. 1 CRIME 4 2 3

Slide 54

Slide 54 text

SUITABLE VICTIM MOTIVATED OFFENDER Crime theory 54 CRIME Felson, M. (1994). Crime and everyday life: Insight and implications for society. Thousand Oaks, CA: Pine. 1 CRIME 4 2 3

Slide 55

Slide 55 text

ABSENCE OF CAPABLE GUARDIAN SUITABLE VICTIM MOTIVATED OFFENDER Crime theory 55 CRIME Felson, M. (1994). Crime and everyday life: Insight and implications for society. Thousand Oaks, CA: Pine. 1 CRIME 4 2 3

Slide 56

Slide 56 text

Crime: does place matters/predicts? 56 DATA • Web and Open data (physical characteristics) • Mobile phone data (proxy for mobility) MODEL • Fit with Spatial Negative Binomial Model (NB) • Predict with Cross-validated Random Forest 1 CRIME 4 2 3

Slide 57

Slide 57 text

Crime data: describe and predict 57 CITY STRUCTURE OSM, Foursquare Data 1 CRIME 4 2 3

Slide 58

Slide 58 text

Crime data: describe and predict 58 CITY STRUCTURE DEPRIVATION Census Data OSM, Foursquare Data 1 CRIME 4 2 3

Slide 59

Slide 59 text

Crime data: describe and predict 59 PEOPLE’S MOBILITY CITY STRUCTURE DEPRIVATION O/D matrices (Mobile phone data) Census Data OSM, Foursquare Data 1 CRIME 4 2 3

Slide 60

Slide 60 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Negative Binomial Regression 60 1 CRIME 4 2 3

Slide 61

Slide 61 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Negative Binomial Regression 61 1 CRIME 4 2 3 Crime (Ground truth)

Slide 62

Slide 62 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Negative Binomial Regression 62 1 CRIME 4 2 3 Offset (population) Crime (Ground truth)

Slide 63

Slide 63 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Negative Binomial Regression 63 1 CRIME 4 2 3 Offset (population) Crime (Ground truth) Features (e.g. land use mix, deprivation)

Slide 64

Slide 64 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Negative Binomial Regression 64 1 CRIME 4 2 3 “everything is related to everything else, but near things are more related than distant things.” Tobler's first law of geography

Slide 65

Slide 65 text

log ((% ) = ( ) + W + ( Y Y P Y[K + ( + P\+ ] +[K The Spatial Negative Binomial Regression 65 (significant) Spatial Eigenvectors 1 CRIME 4 2 3 Getis, Arthur, and Daniel A. Griffith. "Comparative spatial filtering in regression analysis." Geographical analysis 34.2 (2002): 130-140. • Eigenvector Spatial Filtering

Slide 66

Slide 66 text

66 Metric Social disorganization Daily routine City structure Full RMSE 231.93 312.70 145.04 127.76 McFadderPseudo-R^2 0.077 0.085 0.113 0.143 Predict (nowcast) crime * 5-fold Cross-validation 1 CRIME 4 2 3 PRELIMINARY RESULTS

Slide 67

Slide 67 text

Preliminary result 67 Physical characteristics better describe and predict crime 4 1 CRIME 2 3

Slide 68

Slide 68 text

Next? 68 4 • Add ambient population • Model Los Angeles, Boston, Providence • Describe commonality and differences 1 CRIME 2 3

Slide 69

Slide 69 text

Structural layout ‘GENERATE’ THE CITY Q: can we formalize the desired qualities of a neighborhood and prototype it?

Slide 70

Slide 70 text

Why simulate/generate a city • Endless discussions between stakeholders • Describe, predict => play • New insights 70 2 1 3 STRUCTURAL LAYOUT 4

Slide 71

Slide 71 text

Why simulate/generate a city • Endless discussions between stakeholders • Describe, predict => play • New insights 71 2 1 3 STRUCTURAL LAYOUT 4

Slide 72

Slide 72 text

Structural layout Design and enhance existing layouts • Learn from thousands of examples • Respecting the existing constraints • Build neighborhoods that work 72 2 1 3 STRUCTURAL LAYOUT 4

Slide 73

Slide 73 text

Constructive Machine Learning Traditional approaches are limited • Model complex relations • Predict structured objects • Hard and soft constraints on the output 73 2 1 3 STRUCTURAL LAYOUT 4

Slide 74

Slide 74 text

Constructive Machine Learning Traditional approaches are limited • Model complex relations • Predict structured objects • Hard and soft constraints on the output 74 2 1 3 STRUCTURAL LAYOUT 4 In collaboration with Andrea Passerini and MIT Media Lab

Slide 75

Slide 75 text

Why does it matter?

Slide 76

Slide 76 text

Data mining to understand • Data mining as an inexpensive way to understand urban mechanisms • Predict social outcome from newly arise data • Deep understanding of city life through multi-modal data 76

Slide 77

Slide 77 text

Limitations • Presence of data • Bias on data and models • Partial view • Domain adaptation (is it?) 77

Slide 78

Slide 78 text

Multi-disciplinariety means discussions We collaborate with • Studio Carlo Ratti (MIT) • Humnetlab Marta Gonzalez (MIT) • Data-pop alliance, World Bank • Criminology researcher • MIT Media Lab Changing places 78

Slide 79

Slide 79 text

Multi-disciplinary investigation into urban life 79 Describe Predict Generate

Slide 80

Slide 80 text

We published CONFERENCES • De Nadai, M., et al. "The Death and Life of Great Italian Cities: A Mobile Phone Data Perspective." WWW, 2016. • De Nadai, M., et al. "Are Safer Looking Neighborhoods More Lively?: A Multimodal Investigation into Urban Life." ACM MM, 2016. JOURNALS • Barlacchi, G., De Nadai M., et al. A multi-source dataset of urban life in the city of Milan and the Province of Trentino. Scientific data, 2 (2015). • Centellegher, S., De Nadai M., et al. "The Mobile Territorial Lab: a multilayered and dynamic view on parents’ daily lives." EPJ Data Science 5.1 (2016). 80

Slide 81

Slide 81 text

The plan for the (next) future 81

Slide 82

Slide 82 text

Thanks!

Slide 83

Slide 83 text

Safety perception: fix sparse votes • Learning safety perception, predict in Rome and Milan • AlexNet CNN trained on Places205* 83 * B. Zhou, A. Lapedriza, J. Xiao, A. Torralba, and A. Oliva. “Learning Deep Features for Scene Recognition using Places Database.” NIPS, 2014. ** Ordonez, Vicente, and Tamara L. Berg. "Learning high-level judgments of urban perception.” ECCV, 2014. Model type State of art [**] Our model NY - NY 0.687 0.718 NY - Boston 0.701 0.734 Boston - Boston 0.718 0.744 Boston - NY 0.636 0.693 1 2 MULTI- MODAL APPROACH

Slide 84

Slide 84 text

“What this [paper] does is put the facts on the table, and that’s a big step” … “It will bring up a lot of other research, in which, I don’t have any doubt, this will be put up as a seminal step” Luis Valenzuela, Urban Planner Harvard University Source: http://news.mit.edu/2016/quantifying-urban-revitalization-1024

Slide 85

Slide 85 text

Safety perception: aggregation 85 1 2 MULTI- MODAL APPROACH 3.5 3.7 3.9 4.2 4.4 4.6 4.8 5.0 Safety score DUOMO SAN SIRO QUARTO OGGIARO CITTA' STUDI BICOCCA TRASTEVERE TIBURTINO OSTIENSE PRIMAVALLE LESS SAFE SAFER ROME MILAN