Finding your custom audiences in the IoT world
- 1. Finding Your Audience in the Internet of Things Arpan GhoshDhruv Choudhary
- 4. • GPS readings • Battery voltage, check engine light and DTC codes • Speed, RPM, mass air-flow, fuel level and hundreds of other vehicle sensors every second Vehicles Are a Rich Data Source
- 6. Same Vehicle Three Different Drivers
- 8. Velocity Acceleration Frequency Histogram Time Spent Driving (hrs) Automatic’s User Data Model
- 9. Sudden Accel & Decel Consistent Velocities
- 13. Raw Image with Noise Processed Image Compressed Representation Stacked Denoising Autoencoder
- 14. Stacked Denoising Autoencoder Raw Histograms with Noise Processed Histograms
- 15. Original Non Negative Matrix Factorization W H Reconstructed Factor Membership Latent Factors
- 18. Wide Spread in Acceleration Vectors
- 19. Significant Spread at Neighborhood Velocities
- 20. Correlation to Relatable Personas
- 21. • Behavior at City Velocities (Impatience) • Behavior at Highway Velocities (Aggression) • Local Conditions (Traffic, Weather, Terrain, Temperature) affect velocity distributions. Possible Hidden Variables
- 22. Patient, no traffic, mostly freeway driver. Calm Calm Mild Aggressive and impatient all around city and highway. Aggressive Impatient Extreme Aggressive, in traffic, highway (40-60mph) Moderate Moderate Extreme Impatient, in traffic, in city/ neighborhood Calm Impatient Extreme Patient, in mild traffic, expressway driver. Calm Calm Moderate Aggressive, little traffic, drives on the highway and expressway. Aggressive Calm Moderate Impatient, in traffic, drives mostly in city and highway. Calm Moderate Extreme AGGRESSION PATIENCE LOCAL CONDITIONSDRIVER CHARACTERISTICS
- 23. Patient, no traffic, mostly freeway driver. Calm Calm Mild Aggressive, little traffic, drives on the highway and expressway. Aggressive Calm Moderate AGGRESSION PATIENCE LOCAL CONDITIONSDRIVER CHARACTERISTICS
- 25. Driving Behavior is Complex! ! Highly dependent on our surroundings. ! This is just a first attempt in trying to understand our users and their behavior. ! Many other factors affect user behavior • Length of the trip • Time of Day • Mood • Urgency
- 27. Vehicles as an ‘Audience’ Velocity RevolutionsPerMinute
- 28. store.automatic.com FOR 20% OFF, ENTER PROMO CODE DataByTheBay20
- 29. • Principal Component Analysis • Non Negative Matrix Factorization • Matrix Completion • Robust PCA • Sparse Autoencoders Model Selection Low Rank Representation Clustering • k-means • Non Negative Matrix Factorization • Sparse Autoencoders • Hierarchical k-means • LDA