Doing data science on a device

At Snips, we are working on the next generation of context-aware interfaces for mobile devices. 

By looking at the user's location pattern and enriching it with their personal, social and urban contexts, we can predict their intentions, and thus create apps that dynamically adapt to what they actually want to do. For reasons of speed, battery efficiency and privacy, most of the data processing and machine learning involved is performed directly on-device. 

In this talk, we show how we work with Swift and ReactiveCocoa, and how this allows for fast prototyping and profiling of algorithms in a real-world setting, while ensuring a robust data-processing model that scales and can be passed to production.

Transcript

1. Doing data science on a device @michaelfester
5. :snips
7. Devices must learn and adapt to our needs
8. BAR BANK WAITING TIME: 15 MIN FOOD COURT RATING: 4/5 WAITING TIME: 2 MIN METRO STATION LINE: 1, TRAFFIC: FLUID NEXT PEAK: 17:45 TUESDAY, 14:25 CLOUDY, RAIN: 15% TEMPERATURE: 18°C NATIONAL HOLIDAY CAR TRAFFIC FLUID, NO DANGER CONSTRUCTION SITE NOISY, FREQUENT TRUCK STOPS POLICE STATION BUS STOP LINES: 45, 72, 96, 108 NEXT ARRIVAL: 4 MIN
13. Swift bears familiarity with scripting languages
14. » Reduced state / side effects » Immutability & purity » Testability » Thread safety Swift has a functional flavor
15. Imperative / stateful Functional reactive
17. It runs on a device!
18. » Privacy » Speed » Battery » Cost On-device machine learning
19. The future: an AI in all our devices