Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Proof-of-Learning with Incentive Security

Zishuo Zhao, Zhixuan Fang, Xuechao Wang, Xi Chen, Hongxu Su, Haibo Xiao, Yuan Zhou

TL;DR

This work tackles the sustainability and security limitations of traditional blockchain consensus by proposing an incentive-secure Proof-of-Learning (PoL) mechanism within Proof-of-Useful-Work (PoUW). It develops both trusted-verifier and untrusted-verifier designs, proving theoretical incentive-security guarantees and achieving substantially lower verification overheads ($O\left(\frac{\log E}{E}\right)$ or $O\left(\frac{1}{E}\) with staking) than prior PoL schemes. Central to the approach is a Capture-The-Flag protocol that injects verifiable flags to deter lazy or malicious verifiers and to align economic incentives for all parties, including front-end secure operation under untrusted problem providers. The experimental results on standard ML tasks demonstrate robustness against a range of attacks and practical overheads, highlighting the potential to enable a decentralized, verifiable ML-as-a-Service ecosystem and a decentralized computing power market in the AI era.

Abstract

Most concurrent blockchain systems rely heavily on the Proof-of-Work (PoW) or Proof-of-Stake (PoS) mechanisms for decentralized consensus and security assurance. However, the substantial energy expenditure stemming from computationally intensive yet meaningless tasks has raised considerable concerns surrounding traditional PoW approaches, The PoS mechanism, while free of energy consumption, is subject to security and economic issues. Addressing these issues, the paradigm of Proof-of-Useful-Work (PoUW) seeks to employ challenges of practical significance as PoW, thereby imbuing energy consumption with tangible value. While previous efforts in Proof of Learning (PoL) explored the utilization of deep learning model training SGD tasks as PoUW challenges, recent research has revealed its vulnerabilities to adversarial attacks and the theoretical hardness in crafting a byzantine-secure PoL mechanism. In this paper, we introduce the concept of incentive-security that incentivizes rational provers to behave honestly for their best interest, bypassing the existing hardness to design a PoL mechanism with computational efficiency, a provable incentive-security guarantee and controllable difficulty. Particularly, our work is secure against two attacks, and also improves the computational overhead from $Θ(1)$ to $O(\frac{\log E}{E})$. Furthermore, while most recent research assumes trusted problem providers and verifiers, our design also guarantees frontend incentive-security even when problem providers are untrusted, and verifier incentive-security that bypasses the Verifier's Dilemma. By incorporating ML training into blockchain consensus mechanisms with provable guarantees, our research not only proposes an eco-friendly solution to blockchain systems, but also provides a proposal for a completely decentralized computing power market in the new AI age.

Proof-of-Learning with Incentive Security

TL;DR

This work tackles the sustainability and security limitations of traditional blockchain consensus by proposing an incentive-secure Proof-of-Learning (PoL) mechanism within Proof-of-Useful-Work (PoUW). It develops both trusted-verifier and untrusted-verifier designs, proving theoretical incentive-security guarantees and achieving substantially lower verification overheads ( or $O\left(\frac{1}{E}\) with staking) than prior PoL schemes. Central to the approach is a Capture-The-Flag protocol that injects verifiable flags to deter lazy or malicious verifiers and to align economic incentives for all parties, including front-end secure operation under untrusted problem providers. The experimental results on standard ML tasks demonstrate robustness against a range of attacks and practical overheads, highlighting the potential to enable a decentralized, verifiable ML-as-a-Service ecosystem and a decentralized computing power market in the AI era.

Abstract

Most concurrent blockchain systems rely heavily on the Proof-of-Work (PoW) or Proof-of-Stake (PoS) mechanisms for decentralized consensus and security assurance. However, the substantial energy expenditure stemming from computationally intensive yet meaningless tasks has raised considerable concerns surrounding traditional PoW approaches, The PoS mechanism, while free of energy consumption, is subject to security and economic issues. Addressing these issues, the paradigm of Proof-of-Useful-Work (PoUW) seeks to employ challenges of practical significance as PoW, thereby imbuing energy consumption with tangible value. While previous efforts in Proof of Learning (PoL) explored the utilization of deep learning model training SGD tasks as PoUW challenges, recent research has revealed its vulnerabilities to adversarial attacks and the theoretical hardness in crafting a byzantine-secure PoL mechanism. In this paper, we introduce the concept of incentive-security that incentivizes rational provers to behave honestly for their best interest, bypassing the existing hardness to design a PoL mechanism with computational efficiency, a provable incentive-security guarantee and controllable difficulty. Particularly, our work is secure against two attacks, and also improves the computational overhead from to . Furthermore, while most recent research assumes trusted problem providers and verifiers, our design also guarantees frontend incentive-security even when problem providers are untrusted, and verifier incentive-security that bypasses the Verifier's Dilemma. By incorporating ML training into blockchain consensus mechanisms with provable guarantees, our research not only proposes an eco-friendly solution to blockchain systems, but also provides a proposal for a completely decentralized computing power market in the new AI age.
Paper Structure (36 sections, 7 theorems, 64 equations, 4 figures, 4 tables, 4 algorithms)

This paper contains 36 sections, 7 theorems, 64 equations, 4 figures, 4 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Proof-of-Useful-Work in Literature
  4. Settings of Trusted or Untrusted Problem Providers in PoUW Protocols
  5. Trustworthy AI and MLaaS on the Blockchain Platform
  6. Preliminaries
  7. Modeling of ML Training Tasks
  8. Credible (Pseudo-)Randomness Generator
  9. Modeling of Prover's Incentive
  10. Threat Model
  11. Basic Mechanism for Trusted Verifiers
  12. Generation of PoL Certificate
  13. Verification
  14. Full Mechanism for Untrusted Verifiers
  15. Verifier's Strategy Space
  16. ...and 21 more sections

Key Result

theorem 1

In a verification game in which the only information the verifier(s) report is "Success" or "Fail", i.e. $|\mathcal{I}|=1$, and honest verification has a strictly positive cost, i.e. it is impossible to design a verification mechanism with a pure-strategy Nash equilibrium that the prover and verifier(s) simultaneously act honestly.

Figures (4)

  • Figure 1: Experimental Results.
  • Figure 2: Prover Net Utilities.
  • Figure 3: Verifier's Utility
  • Figure 4: Verifier's Utility without CTF Protocol

Theorems & Definitions (13)

  • definition 1: Strict interim individual-rationality
  • definition 2: Strict interim basic incentive-security
  • definition 3: Verification Game
  • theorem 1: Verifier's Dilemma
  • definition 4: Benign verification strategy
  • definition 5: Symmetric-cheating prover
  • theorem 2
  • definition 6: Verifier incentive-security
  • theorem 3: Main Theorem
  • theorem 4
  • ...and 3 more