Federated Learning for Trusted Data Spaces

by Antoi­ne Gar­nier

MUSKETEER has released initi­al open source libra­ries enab­ling users to share and train AI models without com­pro­mi­sing the sov­er­eig­n­ty of their data. The H2020 fun­ded pro­ject offers an indus­tri­al data plat­form lever­aging fede­r­a­ted lear­ning and pri­va­cy pre­ser­ving tech­no­lo­gies ali­gned on IDSA con­cepts.

The pre­sence of a huge num­ber of machi­nes in indus­tri­al auto­ma­ti­on fac­to­ries and the ele­va­ted cost of down­ti­me, pro­du­ce lar­ge expen­ses for pro­duc­tion line main­ten­an­ce. Get­ting a more accu­ra­te eva­lua­ti­on of robot per­for­mance hel­ps to avoid dama­ging the pro­duc­tion capa­ci­ty con­tin­gent­ly by 5 to 20 % in cer­tain cases[1].

But collec­ting data from dif­fe­rent fac­to­ries to build power­ful AI tools can rai­se pri­va­cy issu­es. Data used to train AI models can be sen­si­ti­ve com­pa­ny data but also lead to per­so­nal data con­cerns, for instance data inclu­de infor­ma­ti­on about ope­ra­tors working at the plant.

Fig. 1 Data collection and processing in federated learning environment

Using the MUSKETEER plat­form pro­vi­des a robust solu­ti­on mixing a fede­r­a­ted machi­ne lear­ning approach (local trai­ning using the Mus­ke­teer cli­ent) with pri­va­cy pre­ser­ving tech­no­lo­gies (high­ly cus­to­mi­zed encryp­ti­on, data poi­so­ning attacks miti­ga­ti­on) while respec­ting sov­er­eig­n­ty of the users based on IDSA con­cepts.

The IDS refe­rence archi­tec­tu­re is the de fac­to stan­dard for crea­ting and ope­ra­ting data eco­sys­tems. Its approach is to ensu­re data sov­er­eig­n­ty by faci­li­ta­ting the secu­re exchan­ge of data bet­ween trus­ted par­ties. Cer­ti­fied par­ti­ci­pants are gran­ted access to the data eco­sys­tem, in which they can attach usa­ge poli­ci­es to their data befo­re they make it avail­ab­le to other par­ti­ci­pants.

The Mus­ke­teer cli­ent acting as a poten­ti­al IDS con­nec­tor and the over­all eco­sys­tem is cur­r­ent­ly asses­sing the pro­per com­pli­an­ce with the IDSA stan­dard[2]. This approach has been suc­cess­ful­ly tes­ted in a manu­fac­tu­ring use case hel­ping to sol­ve com­plex chal­len­ges for the auto­mo­ti­ve sec­tor:

  1. Impro­ving the wel­ding qua­li­ty assess­ment to deve­lop pre­dic­ti­ve main­ten­an­ce for robots while incre­a­sing pro­duct safe­ty at the same time.
  2. Trai­ning a wel­ding qua­li­ty assess­ment algo­rithm on lar­ge data­sets from mul­ti­ple fac­to­ries

The pro­ject has recei­ved fun­ding from the European's Hori­zon 2020 rese­arch and inno­va­ti­on pro­gram under grant agree­ment No. 824988.

To find out more about this sto­ry and dis­co­ver the com­ple­te sce­n­a­rio along with user tes­ti­mo­ni­als, visit: https://musketeer.eu/publications/

 

Sources

[1] https://www2.deloitte.com/content/dam/Deloitte/us/Documents/process and operations/us cons pre­dic­ti­ve maintenance.pdf

[2] https://www.internationaldataspaces.org/wp content/uploads/2019/03/IDS Refe­rence Archi­tec­tu­re Model 3.0.pdf