Author:
(1) Martin Kolar, Brno University of Technology (kolarmartin@fit.vutbr.cz). Author: Author: (1) Martin Kolar, Brno University of Technology (kolarmartin@fit.vutbr.cz). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction Volunteer Computing


PNPCoin
3.1 Blockchain Integration and 3.2 Bounded Complexity
3.3 Runtime Authority and 3.4 Back-compatibility


Use case - Cellular Docking


Conclusion and References Volunteer Computing Volunteer Computing Volunteer Computing PNPCoin
3.1 Blockchain Integration and 3.2 Bounded Complexity
3.3 Runtime Authority and 3.4 Back-compatibility PNPCoin PNPCoin 3.1 Blockchain Integration and 3.2 Bounded Complexity 3.1 Blockchain Integration and 3.2 Bounded Complexity 3.3 Runtime Authority and 3.4 Back-compatibility 3.3 Runtime Authority and 3.4 Back-compatibility Use case - Cellular Docking Use case - Cellular Docking Use case - Cellular Docking Conclusion and References Conclusion and References Conclusion and References Abstract Research and applications in Machine Learning are limited by computational resources, while 1% of the world’s electricity goes into calculating 34 billion billion SHA-256 hashes per second[5], four orders of magnitude more than the 200 petaflop power of the world’s most powerful supercomputer. The work presented here describes how a simple soft fork on Bitcoin can adapt these incomparable resources to a global distributed computer. By creating an infrastructure and ledger fully compatible with blockchain technology, the hashes can be replaced with stochastic optimizations such as Deep Net training, inverse problems such as GANs, and arbitrary NP computations. 1 Introduction The existence of a powerful global computing infrastructure is paramount in the deployment of Artificial Intelligence. There exists a global computing system, the blockchain [4]. Based on anonymous SHA-256 hash computation, it is a unified ledger of verified transactions. In addition to being secure, Bitcoins are distributed for calculating the hashes necessary to keep the system running, which creates sufficient incentive for millions of people to participate in the global computation. In fact, they perform more computations than the combined power of the 500 most powerful supercomputers by several orders of magnitude. Although the hashes may be run on purpose-made hardware which is not Turing-complete, the blockchain computes an estimated 1000 billion billion floating point operations per second[1], or 1000000 petaflops, compared to 200 petaflops of Summit, which cost 300 million dollars to build. If the general-purpose computers used to compute the blockchain were available for research, the scientific community could harness the computational power of 50’000 such supercomputers. This proposal outlines the limitations of existing Volunteer Computing projects, explains how the Bitcoin technology may be adapted to perform useful computations, and lists several problems that can currently only be solved with this new shared global computer. Never has a Volunteer Computing infrastructure of such magnitude been created, and this work presents a way to harness this power. The creation of another project of success comparable to Bitcoin is unlikely, but there are Bitcoin Improvement Proposals[3] in place to allow for updates of the existing framework. This work proposes updating the proof-of-work algorithm used in Bitcoin, currently the SHA-256 hash function, with a flexible jash function allowing useful computations for the benefit of global science. By allowing SHA-256 hashes as well as new general computations, this improvement can be created without jeopardizing the existing Bitcoin structure, only requiring agreement on a soft fork to the consensus rules. The proposed update, called PNPCoin, enables solving certain types of otherwise intractable computational problems to the blockchain, and receive results within minutes. Several of these problems are critical to AI and Deep Learning, such as finding the next optimum in hyperdimensional stochastic gradient descent, computing the inverse of a nonlinear deep network, and finding the appropriate input to a Generator to fit a Discriminator in GAN applications. Among other applications is brute-force theorem proving, such as running Sledgehammer[6] on randomly generated theorems, a critical step in superhuman problem solving. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv [1] Hashes and FLOPS cannot be compared rigorously, we consider 20 FLOPS per hash, but this can be 20000 on a modern CPU

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

A Soft‑Fork Proposal for Blockchain‑Based Distributed AI Computation

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

15 Common Types of Unethical Behavior Found in Open-Source Projects

Integrating Useful Computation into Blockchain: PNPCoin’s Runtime and Compatibility Framework

From Blockchains to Biotech: How Miners Could Help Discover New Medicines

Was The Dead Internet Theory Only a Theory?

8 Ways Artificial Intelligence Is Transforming Web Design and Development

Difference between Artificial Intelligence, Machine learning, and deep learning

15 Common Types of Unethical Behavior Found in Open-Source Projects

Integrating Useful Computation into Blockchain: PNPCoin’s Runtime and Compatibility Framework

From Blockchains to Biotech: How Miners Could Help Discover New Medicines

Was The Dead Internet Theory Only a Theory?

8 Ways Artificial Intelligence Is Transforming Web Design and Development

Difference between Artificial Intelligence, Machine learning, and deep learning

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps