pith. sign in

arxiv: 2307.07066 · v2 · submitted 2023-07-13 · 💻 cs.CR · cs.CE· cs.DC· cs.LG

Can Blockchains Reliably Train Machine Learning Models?

Peihao Li , Nadia Dahmani This is my paper

Pith reviewed 2026-05-24 07:46 UTC · model grok-4.3

classification 💻 cs.CR cs.CEcs.DCcs.LG
keywords proof of trainingblockchainmachine learningproof of workverifiable computationdecentralized trainingconsensus protocolenergy repurposing
0
0 comments X

The pith

Proof of training lets blockchains redirect mining power to verifiable machine learning tasks instead of hash puzzles.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces proof of training, a protocol that repurposes the computation in proof of work blockchains for training machine learning models. It achieves this by making the training steps verifiable so the network can confirm the work was done correctly. A sympathetic reader would care because current mining wastes enormous electricity on puzzles with no external value, and this approach could convert that power into models useful in many domains while keeping the same economic incentives for miners. The authors analyze which blockchain structures best support reliable training, security, and scale, then build and test a decentralized implementation. Their findings show the setup supports high task throughput, resists problems, and strengthens the network overall.

Core claim

We introduce proof of training (PoT), a protocol that directs mining power toward verifiable training of machine learning models while preserving PoW's incentives for participation and growth. We study PoT by theoretically identifying the blockchain structure that best meets the goals of training reliability, security, and scalability, and we further evaluate it by implementing a decentralized training network. Our results indicate considerable potential, including high task throughput, strong robustness, and improved network security.

What carries the argument

The proof of training (PoT) protocol, which integrates verifiable machine learning training steps into blockchain consensus so miners receive rewards for useful work rather than hash computations.

If this is right

  • Mining rewards flow to participants who complete verifiable ML training tasks rather than hash puzzles.
  • The blockchain network gains robustness and security improvements from the new consensus structure.
  • Training tasks achieve high throughput in a decentralized setting while models remain reliable.
  • Economic incentives for network growth and participation stay intact under the PoT design.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Networks using this approach could lower overall energy consumption by turning computation into trained models with external value.
  • Similar verification methods might extend to other intensive computations beyond machine learning if the same reliability properties hold.
  • Decentralized model training could become more accessible if multiple blockchains adopt comparable protocols.

Load-bearing premise

Machine learning training can be structured so that its correctness is verifiable by the blockchain network in a decentralized way without creating fresh security holes or scalability limits.

What would settle it

A concrete attack or test case in which a miner submits a training result that passes the PoT verification checks yet does not correspond to actual correct training performed on the claimed data and model.

read the original abstract

Large proof of work (PoW) networks allow anyone to earn rewards by running computation-intensive hash puzzles for profit, yet they typically consume electricity comparable to that of medium-sized countries. Repurposing computing resources from hash puzzles to machine learning training can benefit the energy sector as a whole, since this computing power is no longer wasted on solving hash puzzles but is instead used to train machine learning models that provide value across different application domains. However, major technical gaps currently prevent this integration. To bridge these gaps, we introduce proof of training (PoT), a protocol that directs mining power toward verifiable training of machine learning models while preserving PoW's incentives for participation and growth. We study PoT by theoretically identifying the blockchain structure that best meets the goals of training reliability, security, and scalability, and we further evaluate it by implementing a decentralized training network. Our results indicate considerable potential, including high task throughput, strong robustness, and improved network security.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 1 minor

Summary. The manuscript introduces Proof of Training (PoT), a protocol to repurpose PoW mining computation for verifiable machine learning model training while preserving participation incentives. It theoretically determines the blockchain structure best suited to training reliability, security, and scalability, then evaluates the design via implementation of a decentralized training network, reporting high task throughput, strong robustness, and improved network security.

Significance. If the protocol's verifiability construction and integration hold, the work could meaningfully address PoW energy consumption by converting it into useful ML computation, while potentially strengthening blockchain security. The combination of theoretical blockchain-structure analysis and a concrete implementation provides both conceptual grounding and practical evidence, which is a clear strength of the manuscript.

minor comments (1)
  1. The abstract would benefit from explicit numerical results (e.g., measured throughput in tasks per second or robustness percentages) rather than qualitative descriptors such as 'high' and 'strong'.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their careful summary of our manuscript and for recognizing its potential significance in repurposing PoW computation for verifiable ML training. The report accurately reflects the protocol design, theoretical analysis, and implementation results. No specific major comments were provided in the referee report, so we have no point-by-point responses or revisions to address at this stage. We remain available to clarify any aspects of the work or respond to additional feedback.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces a novel PoT protocol as a new construction, identifies suitable blockchain structures theoretically, and evaluates via implementation. No load-bearing steps reduce by definition, fitted inputs renamed as predictions, or self-citation chains to the target result. The derivation chain is self-contained against external benchmarks of protocol design and empirical evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim depends on the domain assumption that verifiable decentralized training is feasible, which is not evidenced in the abstract.

axioms (1)
  • domain assumption It is possible to verify the correctness of machine learning model training in a decentralized setting suitable for blockchain integration.
    This is required for the PoT protocol to function as described.
invented entities (1)
  • Proof of Training (PoT) no independent evidence
    purpose: A protocol to direct mining power to verifiable ML training
    Newly introduced in this paper to address the integration gap.

pith-pipeline@v0.9.0 · 5693 in / 1187 out tokens · 34908 ms · 2026-05-24T07:46:03.704356+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages · 1 internal anchor

  1. [1]

    Ocean Protocol: A Decentralized Substrate for AI Data and Services

    2018. Ocean Protocol: A Decentralized Substrate for AI Data and Services . Accessed: [Date Accessed]

    work page 2018

  2. [2]

    Fetch.AI Whitepaper

    2019. Fetch.AI Whitepaper. Accessed: [Date Accessed]

    work page 2019

  3. [3]

    Ethereum Network Hashrate Chart

    2023. Ethereum Network Hashrate Chart. Accessed: [Date Accessed]

    work page 2023

  4. [4]

    Singularitynet

    2023. Singularitynet. Accessed: 2023-04-05

    work page 2023

  5. [5]

    Alshahrani, N

    H. Alshahrani, N. Islam, D. Syed, A. Sulaiman, M. Al Reshan, K. Rajab, A. Shaikh, J. Shuja-Uddin, and A. Soomro, Sus- tainability in blockchain: A systematic literature review on scalability and power consumption issues , Energies 16 (02 2023), 1510

    work page 2023

  6. [6]

    Baldominos and Y

    A. Baldominos and Y. Saez, Coin.ai: A proof-of-useful-work scheme for blockchain-based distributed deep learning , Entropy 21 (2019)

    work page 2019

  7. [7]

    IPFS - Content Addressed, Versioned, P2P File System

    J. Benet, IPFS - content addressed, versioned, P2P file system , CoRR abs/1407.3561 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  8. [8]

    Boyle, 2021

    J. Boyle, 2021. BTC Mining Used More Electricity than Sweden. Accessed: [Date Accessed]

    work page 2021

  9. [9]

    Brassard, D

    G. Brassard, D. Chaum, and C. Crépeau, Minimum disclosure proofs of knowledge, J. Comput. System Sci. 37 (1988), 156– 189

    work page 1988

  10. [10]

    Bravo-Marquez, S

    F . Bravo-Marquez, S. Reeves, and M. Ugarte, Proof-of-learning: A blockchain consensus mechanism based on machine learn- ing competitions, 2019 IEEE International Conference on Decentralized Applications and Infrastructures (DAPPCON), 2019, pp. 119–124

    work page 2019

  11. [11]

    Castro and B

    M. Castro and B. Liskov, Practical byzantine fault tolerance, OSDI 99 (1999), 173–186

    work page 1999

  12. [12]

    Cohney, J

    S. Cohney, J. Sklaroff, and D. Wishnick, The dao controversy: The case for a new legal framework for daos , Stanford T ech- nology Law Review 24 (2021)

    work page 2021

  13. [13]

    Cryptocurrency miners may lead the next stage of ai

    Cointelegraph, 2023. Cryptocurrency miners may lead the next stage of ai . Accessed: [Date Accessed]

    work page 2023

  14. [14]

    De Vries, Cryptocurrencies on the road to sustainability: Ethereum paving the way for bitcoin, Patterns 4 (2023), 100633

    A. De Vries, Cryptocurrencies on the road to sustainability: Ethereum paving the way for bitcoin, Patterns 4 (2023), 100633

    work page 2023

  15. [15]

    Eos.io technical white paper

    EOS.IO, 2017. Eos.io technical white paper

    work page 2017

  16. [16]

    Fischer, N.A

    M.J. Fischer, N.A. Lynch, and M.S. Paterson, Impossibility of distributed consensus with one faulty process , J. ACM 32 (apr 1985), 374–382

    work page 1985

  17. [17]

    Foundation., Ethereum 2.0 Specifications

    E. Foundation., Ethereum 2.0 Specifications. Accessed: [Date Accessed]

    work page

  18. [18]

    Langer, Z

    M. Langer, Z. He, W. Rahayu, and Y. Xue, Distributed training of deep learning models: A taxonomic perspective, IEEE T rans. Parallel Distrib. Syst. 31 (2020), 2802–2818. 27

    work page 2020

  19. [19]

    Y. Liu, Y. Lan, B. Li, C. Miao, and Z. Tian, Proof of learning (pole): Empowering neural network training with consensus building on blockchains, Comput. Networks 201 (2021), 108594

    work page 2021

  20. [20]

    Nguyen, H

    C. Nguyen, H. Dinh Thai, D. Nguyen, D. Niyato, H. Nguyen, and E. Dutkiewicz, Proof-of-stake consensus mechanisms for future blockchain networks: Fundamentals, applications and opportunities , IEEE Access PP (06 2019), 1–1

    work page 2019

  21. [21]

    Nguyen and K

    T. Nguyen and K. Kim, A survey about consensus algorithms used in blockchain , J. Informat. Process. Syst. 14 (01 2018), 101–128

    work page 2018

  22. [22]

    Sergey and S

    N. Sergey and S. Johnson, 2017. Chainlink: A decentralized oracle network. Accessed: yyyy-mm-dd

    work page 2017

  23. [23]

    Federal Act on the Adaptation of Federal Law to Developments in Distributed Electronic Register T echnology

    Swiss Government, 2021. Federal Act on the Adaptation of Federal Law to Developments in Distributed Electronic Register T echnology. Accessed: [Date Accessed]

    work page 2021

  24. [24]

    W. Wang, H. Dinh Thai, P . Hu, Z. Xiong, D. Niyato, P . Wang, Y. Wen, and D.I. Kim, A survey on consensus mechanisms and mining strategy management in blockchain networks, IEEE Access PP (01 2019), 1–1

    work page 2019