PRIVACY-AWARE COLLABORATION

Federated threat detection

Keep security logs local. Make every training contribution accountable.

Illustrative data

Round budget ↗

30credits

Financial budget, separate from privacy

Settled payment ▣

0.00credits

Awaiting offers and execution

Accepted offers ◇

0/ 12 organizations

Confirmed locally by each client

Released updates ⇡

0updates

Training and uploads are simulated

01 / NETWORK

A network of independent participants

Round 1 · Ready

PendingAcceptedDeclinedSelect a node to inspect its local view

Privacy as
CommodityProcurement · Coordination

Waiting for a task

03 / HOW IT WORKS

From a private preference to a traceable contribution

00 / 06

01Publish taskObjective & budget
02Collect bidsCost & privacy preference
03Make offersAllocation & payment
04Check locallyCapacity, ledger & utility
05Train & releaseProtected model updates
06Settle roundAggregation & audit

Play a round or advance one step at a time. Select an organization to inspect its local decision.

⌘

What happens inside the method?Build a mean field, explore equilibrium, and understand the learning objective.

Explore the method ↗

04 / MECHANISM

Where MFG-RegretNet fits

Method overview

⇥Client reportsCost · Privacy capacity

→

◉Population contextSummary of current bids

→

⌘Allocation / PaymentLearned offer mapping

→

✓Budget projectionTotal offered payment ≤ B

Mean reported cost—Mean reported capacity—Total offered payment—

Training

Search for profitable misreports and penalize regret. A local rejection check protects execution; incentive guarantees require their own analysis.

CLIENT VIEW

Client 01 · Campus gateway

Local constraints and decisions

Reported / true capacity—

Allocated ε / offered credits—

Realized utility this round—

Cumulative ε · Demo0.00 / 2.00

Waiting for an offer

05 / CONTRACTS

Every offer, explained

Credits and ε have different units · Illustrative records

Participant	Bid v	Reported cap	Allocated ε	Offered credits	Local decision	Settled credits

06 / AUDIT TRAIL

A record of every step

FROM METHODS TO EVIDENCE

Compare the manuscript results

Explore accuracy, incentive security, and welfare across the proposed method and its baselines.

Open results comparison ↗

MNISTCIFAR-10Mean ± SD

A fixed snapshot of the manuscript tables. The live workflow above remains a synthetic illustration.

▤

The research behind this demoDependable and Secure Privacy Trading in Federated Learning via MFG-RegretNet · Sun et al., 2026.

Read the paper ↗