Banking classification collaboration

What Cozy's collaborative AI project for banking classification?

We would like to move towards autonomy in terms of automatic classification; our aim is to use this project in order to develop another project with a wider scope on collaborative and decentralized AI. 

Why suggest our own automatic banking classification service?

The banking expenses classification used on your Cozy via a bank connector is a key Bank service, but it's a complicated operation to automate. Several third-party services - such as Linxo – offer this service subject to sending your bank data to their servers. This touches on one of the core principles that we defend here at Cozy Cloud: your data is yours, and it must remain within your control and subject to your consent. Also, recourse to a proprietary service risks sidelining our self-hosting users, which we consider unacceptable. That's why we are leaning towards our own automatic banking classification algorithm.

How is Cozy Cloud approaching this service?

Our aim is to ensure Cozy server autonomy in terms of this automated banking classification task. This requires a certain algorithm based on a mathematical model, as well as a calibration parameter. These two elements together produce a predictive algorithm, an AI dedicated to automated banking classification. The classification system we are designing bears Cozy's DNA: it's open, (partially) decentralized and respects your private life. Also, by including it in each of your Cozy servers, we can guarantee the autonomy of this task: your banking data may be processed at home using your Cozy. This autonomy is a step towards decentralization!

We determined the algorithm's initial calibration on the basis of examples we found on our own side. The classification obtained using this parameter will not be perfect and it's possible that the algorithm may require adjustments. In order to ensure that it develops, we would like to capitalize on collective intelligence: we invite volunteers to enrich the shared AI with any corrections they may provide to their Cozy predictions whilst respecting their personal data. 

It goes without saying: you will benefit from the shared intelligence thus created even if you don't want to share your corrections. This sharing of new information is a step towards collaborative AI!

Why volunteer to participate in a collaborative project?

By volunteering to help Cozy develop its automatic classification, users help:

• Enrich the collaborative algorithm
• Build a shared and open AI
• Develop an AI that meets users' needs 

We are currently required to use a server dedicated to recovering corrections sent by volunteers. Our data protection philosophy requires extreme caution in gathering this data and we have planned several technical safeguards in order to ensure that we can never disclose or use your personal data identifying you by name, or even reconstitute an anonymous profile. Volunteers are offered the following guarantees:

• Data transmitted from your Cozy shall only contain the strict minimum required to fine tune the calibration
  • Bank transaction name
  • Category initially predicted by the algorithm
  • Category that you modified
  • Transaction date
  • Transaction amount
  • Information on the version of information extracted
• For reasons of anonymity, this data sharing shall not contain your Cozy address, nor any other user-specific information.

For more information on our security guarantees, please read this article.