Tokenized Reward Systems for Federated Learning: The Privacy vs. Fairness Tradeoff
Organizations: Department of Electrical and Computer Engineering, Democritus University of Thrace, 67100 Xanthi, Greece · Institute for Language and Speech Processing, Athena Research Center, 67100 Xanthi, Greece
Abstract
Federated learning (FL) systems increasingly rely on blockchain based reward mechanisms to incentivize participation while preserving data privacy. We design and empirically evaluate two complementary Solidity smart contracts for FL reward distribution: the first one that allocates rewards proportionally to a publicly disclosed quality score, and a privacy preserving variant that hides individual quality scores behind a zero-knowledge (ZK) proof of a threshold condition. Both contracts mint a non transferable, use-and-burn token maintaining a trustless incentive structure. We evaluate the two designs through MNIST training with adversarial behavior, including Sybil attacks, zero gradient and random noise, across three independent random seeds. Our results reveal a fundamental tradeoff between cryptographic privacy and incentive fairness. The privacy preserving architecture strictly degrades system robustness, doubling the adversary's average reward capture while introducing severe outcome variance. This vulnerability is driven by a structural self-inclusion bias within cosine-similarity scoring: as the global model converges, legitimate gradient updates naturally diminish, rendering them statistically indistinguishable from bounded noise. Adversaries exploit this dynamic to systematically breach the ZK threshold, triggering a late stage failure cascade that culminates in absolute adversary monopoly over the reward pool and the total exclusion of honest contributors. Alongside a closed-form theorem formalizing this privacy--Sybil resistance tradeoff, we map strict mainnet deployment boundaries via linear gas scaling models, explicitly incorporating the post-Fusaka EIP-7825 per transaction gas cap. Ultimately, while cryptographically secure, static threshold ZK gates fail as incentive layer fairness mechanisms under realistic FL convergence dynamics on standard benchmarks.
Figures & tables
| Operation | Standard | ZK | |
|---|---|---|---|
| deploy | 1,847,219 | 2,198,411 | +351,192 |
| registerParticipant | 92,647 | 92,647 | 0 |
| updateStake | 30,812 | 30,812 | 0 |
| submitQualityProof | — | 127,640 | — |
| distributeRewards | 60,738 | 61,509 | +771 |
| (slope, gas / participant) | (+1.27%) |
| Component | Phase 2 Std | Phase 3 ZK (stub) | Phase 3 ZK (real) |
|---|---|---|---|
| distributeRewards (total) | 13,028,590 | 7,210,890 | 7,210,890 |
| submitQualityProof (total) | 0 | 8,946,684 | 22,596,684 |
| Total gas (20 rounds) | 13,028,590 | 16,157,574 | 29,807,574 |
| Multiplier vs. Standard | 1.0 | 1.24 | 2.29 |
| Per-round average | 651,430 | 850,399 | 1,568,820 |
| Round | Predicted bias | Measured | Match | ||
|---|---|---|---|---|---|
| 2 | 2.10 | 1.73 | 0.121 | 0.117 | 96.7% |
| 4 | 2.11 | 0.84 | 0.252 | 0.243 | 96.6% |
| 5 | 2.12 | 0.68 | 0.310 | 0.303 | 97.6% |
| 7 | 2.11 | 0.65 | 0.323 | 0.328 | 98.5% |
| 10 | 2.10 | 0.43 | 0.491 | 0.496 | 98.9% |
| 12 | 2.11 | 0.42 | 0.506 | 0.513 | 98.6% |
| Round | Adv. verified | Honest verified | Outcome | |
| 1–5 | 7–9 | 0 | 7–9 | Gate functions: only adversaries blocked |
| 6–9 | 5–8 | 0 | 5–8 | Honest clients begin to fail |
| 10–13 | 1–2 | 0–1 | 0–2 | Gate severely degraded |
| 14 | 0 | 0 | 0 | Total gate failure: |
| 15–16 | 1–2 | 0–1 | 1–2 | Partial recovery |
| 17–20 | 1–2 | 1–2 | 0 | Adversary monopoly: 100% capture |
| Metric | Standard | ZK | Multiplier |
|---|---|---|---|
| Adversary share (mean) | 10.27% | 24.51% | 2.4 |
| Adversary share (std) | 0.71% | 6.93% | 9.8 |
| Adversary share (range) | 9.30–10.90% | 15.09–31.58% | — |
| Honest reward Gini | 0.2171 | 0.4616 | 2.1 |
| Skipped rounds (per 20) | 0 | 1.33 | — |
| Adversary monopoly rounds | 0 | 3.0 | — |