Trus­ted Data Hub

Share sensi­tive data for multi-party Machine Lear­ning appli­ca­ti­ons and analy­sis without disclo­sing raw data

Trus­ted Data Hub in video

The video shown here gives you a brief over­view of the func­tion­a­lity of the ADVANEO Trus­ted Data Hub with appli­ca­tion examples

Privacy Preser­ving Multi-Party-Compu­­ting (MPC):

Share data for multi-party Machine Lear­nig appli­ca­ti­ons and analy­sis without disclo­sing access to sensi­tive data

Privacy Preser­ving Machine Learning

The joint analy­sis and use of data from diffe­rent parties, e.g. from machi­nes, usually fails due to issues of confi­den­tia­lity, compe­ti­tion and direct compe­ti­tion. The “Trus­ted Data Hub”, a novel combi­na­tion of hard­ware and soft­ware tech­no­lo­gies, provi­des the solu­tion to protect company secrets while sharing data for AI-based analysis.

Data remains private, public analy­ses become possible

“Trus­ted Hub Solu­tion” offers manu­­fac­­tu­­rer-inde­­pen­­dent, data sove­reign machine lear­ning (PPML) and multi-party compu­ting (MPC). PPML enables the trai­ning of the arti­fi­cial intel­li­gence (AI) model and machine lear­ning (ML) in a way that preser­ves the privacy of the data.

Analyse data without disclo­sing it

The MPC allows insti­tu­ti­ons to carry out analy­ses of private data held by seve­ral other insti­tu­ti­ons without ever disclo­sing the data. In other words, MPC is a tech­ni­que where calcu­la­ti­ons and data analy­sis are perfor­med with input from various parties. The parties are not given insight into the data provi­ded by the other parties – only the results of the data analy­sis are shared.

Guaran­tee of confidentiality

The Trus­ted Data Hub solu­tion guaran­tees that the parti­ci­pa­ting parties only receive the results, but no infor­ma­tion about the input of others.

Use cases in verti­cal domains

Many appli­ca­ti­ons in verti­cal domains can bene­fit from the Trus­ted Data Hub. An important exam­ple is the area of predic­tive main­ten­ance. To imple­ment this, machine manu­fac­tu­r­ers usually need access to raw data from multi­ple users. At the same time, data owners do not simply want to disc­lose their trade secrets or machine-gene­ra­­ted data (MGD).

Data control and data sovereignty

The Trus­ted Data Hub enables analy­sis of the combi­ned vulnerable data from diffe­rent compa­nies without losing control of their data. At the same time, machine manu­fac­tu­r­ers can signi­fi­cantly improve their products and services by evalua­ting or analy­zing the combi­ned data of their custo­mers. For exam­ple, machine owners bene­fit from impro­ved machine avai­la­bi­lity and machine settings or from redu­ced wear and tear.

State-of-the-art tech­no­logy

In contrast to all other exis­ting solu­ti­ons, the Trus­ted Hub is comple­tely inde­pen­dent of the analy­tic tools used. While offe­ring the same accu­racy and runtime, the Trus­ted Hub supports not only Python, but also all other ML frame­works, such as Rapidmi­ner, Knime, H2O and Matlab.

Easy inte­gra­tion with full ML compatibility

The Trus­ted Hub is rela­tively easy to inte­grate into exis­ting systems, so it addres­ses the issue of data protec­tion without impac­ting the appli­ca­tion and custo­mer expe­ri­ence. The Trus­ted Hub supports all data types (i.e. nume­ric and cate­go­ri­cal data) and all ML func­tions (i.e. linear and non-linear func­tions) and is ther­e­fore compa­ti­ble with all types of ML algorithms.

Uncom­pro­mi­sing data sovereignty

Further­more, the Trus­ted Hub not only protects the data, but can also protect the ML algo­rithm itself, i.e. only the crea­tor of the ML algo­rithm can access the code.

Uncom­pli­ca­ted implementation

The imple­men­ta­tion and setup of the Trus­ted Hub is very fast and uncom­pli­ca­ted. Further­more, no indi­vi­dual adapt­a­tion is necessary.

Comparison table Multi Party Computation (MPC)

Privacy Preser­ving and Multi Party Compu­ta­tion Tech­ni­queTrus­ted Data Hub 
Addi­tive key/secret SharingHomo­mor­phic Encryp­tion (HE)Diffe­ren­tial privacy (DP)Fede­ra­ted Lear­ning (FL)
Centralized/DecentralizedCentra­li­zedDecen­tra­li­zed with a Trus­ted ServerCentra­li­zedCentra­li­zedDecen­tra­li­zed with a Trus­ted Server
Multi Party Comm­pu­ta­tion SupportYesYesNoYesYes
Data Privacy LevelHighHighVery HighHighHigh
ML Algo­rithm Privacy LevelYesNoYesYesNo
Runtime & ScalabilityHighMediumVery lowMediumMedium
Accu­racyHighMediumMediumMediumMedium to High
Programming Language & ML FrameworkAll Frame­worksPythonVendor-depen­­dentBased on Python and TensorflowTensor­flow and PySyft
Supported Data TypesNume­ri­cal and categoricalNume­ri­cal dataNume­ri­cal dataNume­ri­cal and categoricalNume­ri­cal and categorical
Supported Func­tionsAll func­tionsNot allNot allAll Func­tionsAll func­tions
Network Over­headMediumMediumMediumMediumHigh


Alle hier gezeig­ten Inhalte und Grafi­ken unter­lie­gen dem Urhe­ber­recht und dürfen ohne unsere Erlaub­nis nicht kopiert oder verwen­det werden.