Machine learning

1) semantic

2) images

3) space

Tranquilien

Scope: waste management

1 ton / per person /per year

However: millions of tons are still buried or burned.

Recycling works

Recycling centers

Predictive API

Designing a sensor

The sensor should:

1. measure affluence
2. measure the level of each containers
3. low cost
4. improvable

Image processing:

1. Haar-Cascade for car recognition
2. Random Forest for bin levels

Positives

Negatives

Counting the number of cars

The level of each container

Laser cutting

Protection

Ant in the sky

We also measure:

giving us more context for our predictions.

What's new? Machine learning in the sensor

1) privacy

2) connectivity

Problems of the v1:

1) difficult to build

2) difficult to mount

3) depending on local wifi

4) stays still

New in v2

1) orientable cam

2) brocker.pi

Demo time

Culture

1. we believe in collective intelligence
2. we build simple
3. we think big
4. we do open source
5. we do beautifully

We fix the interface of public services

Why do we need interfaces ?

For users: too much information

For infrastructure: too much people

Each point of view has a different focus ?

1) change the user

2) change the infrastructure

Uber