Our process for client-side reverse geocoding with machine learning

To make website visitors' coordinate data actionable, digital services need to pass it, over the internet, to 3rd party services. This is not 100% safe because as long as coordinate data leaves the device and travels through the internet to 3rd parties, there are risks.

We present a new way of identifying users' location directly on the browser or mobile device by using neural networks in the client interface. 

  • 1

    1. Location control

    pointNG location management widget allows the user to determine the level she wishes her location to be identified (eg. continent, country, state, city).

    1 etrabqrcq4 ktunaxedq4a
  • 2

    2. Location permission

    The browser prompts permission to use location data. 

    Location prompt
  • 3

    3. Fetch neural networks

    Instead of sending the user's latitude and longitude coordinates to 3rd parties, we send a chain of neural networks to the client that determines the users' location directly in the browser.

    Ezgif.com gif maker (4)
  • 4

    4. Magic!

    The user receives the location-based service without the worries of her location being tracked by anyone.

    Secure local weather 2

Comparison to the "traditional" way of doing reverse geocoding

Pointng process

Chaining of Neural Networks

To make the location detection process as light as possible, we use a "chaining" method to fetch small neural networks to the client interface. Each neural network is trained to identify its specific place in the location hierarchy that currently starts from the continent and ends at the city-level. 

  • Europe
    1st neural network calculates the continent. Eg. "Europe"
  • Finland
    Use 2nd "Europe" neural network that predicts the country. Eg. "Finland"
  • Uusimaa
    Use 3rd "Finland" neural network that predicts the state. Eg "Uusimaa"
  • Helsinki
    Use 4th neural network that predicts the city. Eg. "Helsinki"

Speed from edge machine learning

The current solution consists of 5000 pre-trained neural networks that we have stored in AWS S3 storage. To make the distribution of neural networks fast, we use Stack Path Edge CDN to ship the neural networks to the browser. 

Join our BETA

You can test the machine learning process on your website by joining our BETA program. Leave your email, and we'll send you the instructions. 

Error. Your form has not been submittedEmoji
This is what the server says:
There must be an @ at the beginning.
I will retry