"Where, When, Why, and How: Directi
summary
A portfolio of location systems that combine Bayesian models, crowd GPS and device streams to predict destinations, route traffic, and schedule services. A donated dataset of 700,000 km of GPS and tens of thousands of trips powers predestination models that give about a 2‑km error halfway through a trip. An opportunistic planner prototypes fuel and coupon offers; an ABC rideshare simulation with thousands of commutes shows 40→50% commute savings as adoption grows. ClearFlow runs live in 72 cities, updating millions of street estimates every five minutes; Coordinate uses five years of presence data and a learned cost‑of‑interruption model. Privacy work focuses on local on‑device learning.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:54Proliferation of "Intention Machines"
- 1:16Proliferation of "Preference Machines"
- 1:49Geocentric Collaborative Filtering
- 2:42Theme: Streaming Intelligence Continual sensing, learning, and reasoning available in stream of daily life
- 3:35Predestination: Destination from Partial Trajectories Assign probabilities to each cell as trip progresses
- 6:17Example Scenario: Fuel Purchases
- 9:26Ideal Coalescence
- 13:15Leaming from Streams of Sensor Data
- 16:39Models of Future Surprises
- 22:08Forecasting Presence & Availability
- 23:42Unify Multiple Services: Assistant for Presence & Availability
related talks
34mData Drives Business
16mOrchestrating Ecosystems with Location Intelligence
25mAtlas of Popular Transport: Democratizing mobility data
14mTelling Traces
40mPowering Location Data for Good
38mSampling and Modelling the World in Real-Time for Mobility
15mThe turf wars on data - can be a matter of life and death
41mPredicting the Traffic Jam Congestion Aware Routing