Machine Learning Family Routines
Download as PPTX, PDF0 likes305 views
The document discusses the role of routines in family logistics and how machine learning can be applied to model these routines to enhance awareness and control in family planning. It highlights the challenges families face in managing their schedules due to incomplete knowledge of each other's routines and the perception of control over daily activities. The research proposes using sensing technologies and modeling to synthesize crucial information and improve family coordination.
![Because the calendar [is] all words and numbers, so you have to really think about …how long everything takes…Visual Simplifies Calculation“−P1095](https://image.slidesharecdn.com/defense-110322163054-phpapp02/85/Machine-Learning-Family-Routines-96-320.jpg)
Machine Learning Family Routines
- 1. Routine as Resource for the Design of Learning SystemsScott Davidoff Dissertation Defense
- 2. A pattern of behavior that is followed repeatedly, but is subject to change if conditions changeWinter 1964““”1
- 3. Koestler 1967Routines reduce our attention needs2
- 4. Zerubavel 1981Reduce attention needed for tasks3
- 5. Wakkary & Maestri 2007Free attention for bigger challenges4
- 6. Wolin + Bennett 19845
- 7. The cyclic nature of routine makes it a natural target for machine learning
- 8. Machine learning can use this order7
- 9. Suchman 1983, Tolmie et al. 2002Idiosyncrasies are hard to model8
- 10. Certain human routines can be modeled, increasing the scope of activity recognitionMachine learningOpportunities9
- 11. The ability to use learned routines in end-user applications would solve a variety of human problemsHCIOpportunities10
- 13. Sequence of place and transportation (rides) that occur on daily, weekly, and seasonal cycles““”12
- 14. Darrah et al. 2000Managing details can be difficult13
- 15. Frissen 2000Routines give a feeling of control14
- 16. Beech et al. 2004Life does not always follow routines15
- 17. Breakdowns lead to loss of control16
- 18. Perry et al. 2001, Ling + Campbell 2003A constant need to follow updates17
- 19. Gneezy + Rustichini 1998, Darrah 2009A constant source of anxiety18
- 20. We can learn a model of family logistical routines, and present that information to families to help them feel more in control of their lives19
- 21. 20
- 22. 21
- 23. 2We can use sensing and modeling to synthesize missing information resources3Show how to use the model and evaluate the impact of the information1We can use fieldwork to identify missing but needed information resourcesFieldworkModelingValidation22
- 24. Create an ordered list of places and ridesAttend to the details of a plan as it unfoldsLogisticsPlanning and coordinationCoordinatePlan23
- 25. Fieldwork124
- 26. 2212Davidoff et al Ubicomp 2006MonthsFamilies1806Davidoff et al Ubicomp 2007MonthsFamilies0506Davidoff, Dey + Zimmerman CHI 2009MonthsFamilies4524MonthsFamilies25
- 27. Semi-Structured InterviewsDavidoff et al. Ubicomp 200626
- 28. Needs Validationscott davidoff, min kyung lee, anind dey + john zimmerman
- 29. Davidoff et al. Ubicomp 2007User Enactments28
- 30. User EnactmentsDavidoff et al. Ubicomp 200729
- 31. GPSPhone CallsEmailSMSCalendarsDavidoff, Dey + Zimmerman CHI 200930
- 32. 528 phone interviews109 activity Interviews108 calendar monthsDavidoff, Dey + Zimmerman CHI 200931
- 33. Less than 20% of days go exactly as planned1Routines are not documentedPeople have incomplete knowledge of other people’s routinesPeople make plans that depend on incorrect information23432
- 34. 33
- 35. Family ERoutineDeviation, ScheduledDeviation, Unscheduled34
- 36. Family EAll FamiliesRoutineDeviation, ScheduledDeviation, Unscheduled35
- 37. Less than 20% exactly as planned36
- 38. Routineness of Family E Activities37
- 39. Routineness of Family E Activities38
- 40. Routineness of Family E Activities39
- 41. Routines are not documentedFieldwork FindingDavidoff, Dey + Zimmerman CHI 201040
- 42. AugustNovemberSeptemberOctoberDavidoff, Dey + Zimmerman CHI 201041
- 43. People have incomplete knowledge of other people’s routinesFieldwork FindingDavidoff, Dey + Zimmerman CHI 201042
- 44. Davidoff, Dey + Zimmerman CHI 201043
- 45. Davidoff, Dey + Zimmerman CHI 201044
- 46. Davidoff, Dey + Zimmerman CHI 201045
- 47. Davidoff, Dey + Zimmerman CHI 201046
- 48. Davidoff, Dey + Zimmerman CHI 201047
- 49. Davidoff, Dey + Zimmerman CHI 201048
- 50. Davidoff, Dey + Zimmerman CHI 201049
- 51. Davidoff, Dey + Zimmerman CHI 201050
- 52. People make plans that depend on incorrect informationFieldwork FindingDavidoff, Dey + Zimmerman CHI 201051
- 53. View of Plan: Dad4:005:006:007:003:002:008:00SchoolHomeTrack PracticeDad WorkOrthodontistCheck-upScoutsS16Dad52
- 54. View of Plan: S164:005:006:007:003:002:008:00SchoolHomePaper RouteTrack PracticeDad WorkOrthodontistScoutsS16Dad53
- 55. Who Will do the Paper Route?4:005:006:007:003:002:008:00SchoolHomePaper RouteTrack PracticeDad WorkOrthodontistCheck-upScoutsS16Dad54
- 56. Information gaps can break down coordinationNo resources exist to find needed informationDavidoff, Dey + Zimmerman CHI 201055
- 57. Modeling256
- 58. 231Ride DetectionDriver PredictionPredict Lateness57
- 59. In Situ ObservationMaintain Current BehaviorsUbiquitous SensingGPS58
- 60. Parent drives kid to an activityParent drives kidfromactivityRidePick-ups and drop-offsPick-upDrop-off59
- 61. Ride Detection
- 62. Using GPS to Sense a Drop-Offt3t4t5t6t2t1t7Day CareWorkParentChild61
- 63. Using GPS to Sense a Drop-Offt3t4t5t6t2t1t7Day CareWorkParentChild62
- 64. Using GPS to Sense a Drop-Offt3t4t5t6t2t1t7Day CareWorkParentChild63
- 65. Using GPS to Sense a Drop-Offt3t4t5t6t2t1t7Day CareWorkParentChild64
- 66. Using GPS to Sense a Pick-Upt3t4t5t6t2t1t7Day CareWorkParentChild65
- 67. Using GPS to Sense a Pick-Upt3t4t5t6t2t1t7Day CareWorkParentChild
- 68. Using GPS to Sense a Pick-Upt3t4t5t6t2t1t7Day CareWorkParentChild
- 69. GPS Ride Detection Evaluation68
- 70. Smarter power managementWi-Fi, Bluetooth, cell tower IDApply heuristics to modelingCultural norms, Individual BehaviorSingle Location SensorGPS Sampling Rate69
- 72. Driver Prediction Model Features71
- 73. Driver Prediction Performance x Number of Training Weeks of Data1 241 241 241 241 241 241 241 241 241 241 241 241 241 2472
- 74. Driver Prediction Performance x Number of Training Weeks of Data1 241 241 241 241 241 241 241 241 241 241 241 241 241 2473
- 75. Driver Prediction Performance x Number of Training Weeks of Data1 241 241 241 241 241 241 241 241 241 241 241 241 241 2474
- 77. Distribution of Parent Late ArrivalsNumber of ExamplesTnowTidealLateness in MinutesN = 8376
- 78. 77
- 79. 78
- 80. 79
- 81. 80
- 82. 81
- 83. tideal- 30.6A’ = 0.659.582
- 86. We can learn a model of family logistical routines, and present that information to families to help them feel more in control of their lives85
- 87. Validation386
- 88. 87
- 89. Do parents understand the view?Do parents perceive that the information is valuable?Do parents feel more in control having this view?88
- 90. 4524Davidoff et al Ubicomp 2006Davidoff et al Ubicomp 2007Davidoff, Dey + Zimmerman CHI 2009Davidoff et al. CHI 2010MonthsFamilies12Davidoff et al. Ubicomp 2011(in preparation)Families5724MonthsFamilies89
- 91. Davidoff et al. Ubicomp 2011 in preparationExperience prototype: Doctor90
- 92. 91
- 93. 92
- 94. Davidoff et al. Ubicomp 2011 in preparationExperience prototype: Kitchen93
- 95. It's nice to have a map of the day. I don't literally write all this stuff down, so sometimes it's just hard to keep in my head. Visual Distills Details““”−P294
- 96. Because the calendar [is] all words and numbers, so you have to really think about …how long everything takes…Visual Simplifies Calculation“−P1095
- 97. …With the visual you can see who's doing what at what time. You can make a decision about who's activity to change.Visual Simplifies Calculation“”−P1096
- 98. I would be able to know where everyone was gonna be, instead of having to ask around.Visual Clarifies Intentions““”−P1197
- 99. Venkatesh + Davis 2003Technology Acceptance Model-398
- 100. …helps me do my job “as a parent.”…would want my family to use it.…more in control of details, more ready for changesPerception of ControlBehavioral IntentionPerceived Usefulness99
- 101. Perceived UsefulnessBehavioral IntentionPerception of ControlSupports PlanningSupports Awareness7654321100
- 102. We can learn a model of family logistical routines, and present that information to families to help them feel more in control of their lives101
- 103. Contributions102
- 104. Suchman 1983, Tolmie et al. 2002Certain routines can be modeled103
- 105. 231Ride DetectionDriver PredictionPredict Lateness104
- 106. Commodity GPS with real power constraints105
- 107. 106
- 108. Breakdowns lead to loss of control107
- 109. Perceived UsefulnessBehavioral IntentionPerception of ControlSupports PlanningSupports Awareness7654321108
- 110. Gneezy + Rustichini 1998, Darrah 2009A constant source of anxiety109
- 111. Empower ParentsDavidoff et al. Ubicomp 2007110
- 112. Future Work111
- 113. Improve algorithmsUse algorithms to enable other family coordination systemsAdd additional sensor inputsExtend to other forms of routine112
- 114. 113Neustaedter + Brush CHI 2006
- 115. 6pmOrthodontist114Neustaedter + Brush CHI 2006
- 116. 6pmOrthoPaper Route115Neustaedter + Brush CHI 2006
- 117. Baseball116Brown et al. Ubicomp 2007
- 118. Baseball117Brown et al. Ubicomp 2007
- 119. The project onFAMILY CONTROLSMARTHOME
- 120. Thanks for your time, brain power, and support
- 121. Thanks to our sponsorsNSF IIS 1017429Google Research Award
- 122. Thanks to our supporters
- 123. Thanks to my collaborators
- 124. Thanks to my committee
- 125. Thanks to my advisors
- 126. Questions
- 128. QUESTIONSLIDES127
- 130. FORMULAS129
- 132. LWDT131
- 133. Bayesian Network132
- 134. RIDE SENSING TECHNICAL VERSION133
- 135. Davidoff, Dey & Zimmerman CHI 2011134
- 136. Two Kinds of Drop-Offs135
- 137. Two Kinds of Drop-Offst3t2 ,t1 ,5:155:305:456:005:004:456:15Day CareWorkPC136
- 138. Two Kinds of Drop-Offst3t2 ,t1 ,5:155:305:456:005:004:456:15Day CareWorkPC137
- 139. Two Kinds of Pick-Ups138
- 140. Using GPS to Sense a Pick-Upt3t2 ,t1 ,5:155:305:456:005:004:456:15Day CareWorkPC139
- 141. Using GPS to Sense a Pick-Upt3t2 ,t1 ,5:155:305:456:005:004:456:15Day CareWorkPC140
- 142. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWork141
- 143. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWork142
- 144. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWork143
- 145. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWorkPC144
- 146. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWorkPC145
- 147. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWorkPC146
- 148. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWorkPC147
- 149. Using GPS to Sense a Drop-Off5:155:305:456:005:004:456:15Day CareWorkPC148
- 150. Using GPS to Sense a Pick-Up5:155:305:456:005:004:456:15Day CareWorkPC149
- 151. Using GPS to Sense a Pick-Up5:155:305:456:005:004:456:15Day CareWorkPC150
- 152. Using GPS to Sense a Pick-Up5:155:305:456:005:004:456:15Day CareWorkPC151
- 153. Using GPS to Sense a Pick-Up5:155:305:456:005:004:456:15Day CareWorkPC152
- 154. Using GPS to Sense a Pick-Up5:155:305:456:005:004:456:15Day CareWorkPC153
- 155. SCHOOL HALF DAY BREAKDOWN154
- 156. View of Plan: S164:005:006:007:003:002:008:00SchoolHomeTrack PracticeDad WorkBaseballS16Dad155
- 157. View of Plan: S164:005:006:007:003:002:008:00SchoolHomeTrack PracticeDad WorkBaseballS16Dad156
- 158. Dad Started Baseball League4:005:006:007:003:002:008:00SchoolHomeTrack PracticeDad WorkBaseballS16Dad157
- 160. Detect conflict with unlabeled routine4:005:006:007:003:002:008:00SchoolHomePaper RouteTrack PracticeDad WorkOrthodontistCheck-upScoutsS16Dad159
- 161. Dad Started Baseball League4:005:006:007:003:002:008:00SchoolHomeTrack PracticeDad WorkBaseballS16Dad160
- 164. Smarter power managementGPS Sampling Rate163
- 165. Smarter power managementWi-Fi, cell tower, bluetoothSingle Location SensorGPS Sampling Rate164
- 167. Smarter power managementWi-Fi, cell tower, bluetoothApply heuristics to modelingCultural norms, Individual BehaviorSingle Location SensorGPS Sampling Rate166
- 168. Lost Signal During Daytime11:001:00pm2:003:004:0012:005:00SchoolHomeWorkParentChild167
- 170. Driver Prediction Performance x Number of Training Weeks of Data169
- 172. User EnactmentsDavidoff et al. Ubicomp 2007171
- 173. User EnactmentsDavidoff et al. Ubicomp 2007172
- 174. UNUSED PICS173
- 175. Koestler 1967Routines are not documented174
- 176. Darrah et al. 2000, Frissen 2000, Beech et al. 2004People can’t see all the facts175
- 177. Davidoff et al. Ubicomp 2006Managing Rides Causes Anxiety176
- 178. Pentland + Reuter 1994Graceful177
- 179. Wolin & Bennett 1984Path for exercise178
- 181. HCI Approach to Machine Learning180
- 183. Frissen 2000, Fiese et al. 2002, Davidoff et al. 2007Routines help parents feel in control182
- 184. Frissen 2000, Fiese et al. 2002, Davidoff et al. 2007Just surviving the day183
- 185. Darrah et al. 2000, Frissen 2000, Beech et al. 2004Non-Routine Events Cause Anxiety184
- 186. Gneezy + Rustichini 1998, Darrah 2009A constant source of anxiety185
- 187. Gneezy + Rustichini 1998, Darrah 2009A constant source of anxiety186
- 188. How do routines provide support?Davidoff et al. Ubicomp 2007187
- 190. We can use fieldwork to identify missing but needed information resourcesFieldwork1189
- 191. We can use sensing and modeling to synthesize missing information resourcesModeling2190
- 192. We can develop an application to deliver synthesized resources and evaluate how it helps coordinationShow how to use the model and validate that the application potentially solves the problem by givengfpeople the information they wantBuild and validation3191
- 193. We can exploit routine as a resource to design learning systemsThesis Statement192
- 194. Understanding how routines support daily life generate new design ideasGenerative Resource193
- 195. Model information about family transportation routines to supplement family knowledgeInformation Resource194
- 196. We can exploit routine as a resource to design learning systemsThesis Statement195
- 197. context-aware, adaptive, intelligent, cognitive, expert, decision-supportsystem196