Morphware – Decentralized deep learning

Funding Proposals Grants
1

Name of Project

Morphware

Proposal overview

Morphware crowdsources latent computing capacity from gaming rigs to train machine learning algorithms. Gaming rigs running Morphware bid on workloads posted the network in a sealed-bid, second-price (i.e., Vickrey) auction and are compensated in MorphwareToken. It is currently live on Ethereum mainnet.

The reason this work is important is because the computation capacity required to train a state-of-the-art machine learning algorithm doubles approximately every 3.4 months. In contrast, Moore’s Law observes that the number of transistors in a densely integrated circuit double approximately every 18 months; so there’s an enormous gap that is widening. I should also note that data science budgets are ballooning, and that this trend shows no signs of abating.

This benefits the blockchain ecosystem because it gives ETH miners an economic substitute to solving computations related to Ethereum’s proof-of-work consensus mechanism, and will be even more beneficial once Ethereum migrates its consensus mechanism to proof-of-stake. Ignoring the cost of gas on Ethereum, earning MorphwareToken (hereafter: MWT) is roughly 2-3x more profitable than mining ETH and we are at least 20% cheaper than running the same workload on relevant Amazon Web Services products (EC2 or SageMaker).

Usage

We launched in January, so we currently have less than a dozen users; but we’re planning on hosting a Kaggle competition to build-out the demand-side of the network.

We are in the process of building out the supply-side of the network by deploying worker nodes to blockchain clubs at university campuses. We are deploying one worker-node to the University of British Columbia and have a verbal agreement, at the moment, with Harvard College.

Milestones

We have a variety of milestones that we aim on hitting, this year. The first is getting a miniature data center going in an environment with an extremely low cost of electricity. We have already purchased our first 10 GTX 3060 Ti Founders Edition cards, for this, and are currently in the site selection process for the (server) cabinets.

Another milestone is getting the validator node (see: whitepaper) going on mobile. Right now, Morphware only works on laptops and desktops, but we’d love for people to be able to earn MWT on their phones.

Team

Kenso Trabing

Kenso taught Bitcoin Core in a lecture series at Princeton University in Spring 2018, and Solidity & IPFS at Yale University in Fall 2018, while working for a FinTech-focused coding bootcamp based out of midtown Manhattan. The year before that he opened up a data science bootcamp in Bangalore, for said bootcamp, and taught the first couple of cohorts there.

He previously worked as a senior data scientist at Senary Blockchain Ventures.

Darshan Raju

Darshan is an engineer at Microsoft. He is the project’s engineering lead.

Anthony Albertorio

Anthony is the head of developer relations at ConsenSys and has been helping them build their community, for years, and brings a wealth of experience aboard.

Hassan Malik

Hassan is a smart contract engineer at MetaMask and an active participant in New York’s Ethereum community.

He has previously worked as a Solidity developer at AirSwap.

Rebecca Sealfon

Rebecca has a masters degree in computer science from Columbia University, did her undergrad. at Princeton, and used to work for Google.

She’s been creating content for us, which has been featured in high-traffic blogs like Towards Data Science and CoinMonks.

Robert Kinyon

Robert is a production engineer at Facebook, co-author of a book on the Perl programming language, and contributes to the Ruby programming language.

His role is to take Morphware from being a single machine per workload system to a multiple machine per workload system.

Daniel Iñiguez

Daniel is an electrical engineer and a blockchain developer.

He is helping Morphware with a next generation in-house project related to using FPGAs instead of GPUs for certain (computational) tasks.

Proposal ask

$ 235,790

Metrics for success

Our main metric for success would be 1,000 workloads posted to the network per day.

External links

Website: https://morphware.org
Demo: tnb-hyig-zeg (2021-12-06 at 19:23 GMT-8) - Google Drive
Whitepaper: paper/whitepaper.pdf at main · morphware/paper · GitHub

8 Likes
2

A great project which will connect data scientist and gamers who have gpu to rent their computational power on a single platform

4 Likes
3

Hot project for 2022, high demand use-case for industrial and educational sectors made distributed and decentralised!

3 Likes
4

oh wow really needed this kind of project once eth 2.0 is launched

3 Likes
5

Wow, an exciting development and a unique concept of utilising machine learning within cryptospace!

3 Likes
6

Amazing project and even more amazing team. This has a great future

3 Likes
7

an innovative and promising, i am very excited about the future and potential of this project

2 Likes
8

Overwhelmed by the support this post is getting! Thank you, everyone!

2 Likes
9

Proud to be a part of such a fantastic project with a 5-star team! Keep up the great work guys.

2 Likes
10

Thanks @Ebrahim! We’re glad to have you!

11

Dear Harmony

Kenso is really a great guy, he has put together a stellar team and the effort and dedication he has been putting in his project is second to none. I have been on Harmony for nearly a year and in my opinion a project and team like Morphware’s would add value and fit really well within this ecosystem of excellence

3 Likes
12

A project, for a change, that is actually buidling something, and quite revolutionary! Amazing to be part of such project and with a great team of really smart people! :raised_hands:

1 Like
13

Glad to have you, @danieliniguezv!

1 Like
14

I am so happy to see a project like this. I often think about the GPU requirements for ML and why no one has created a solution like this. My guess is that its pretty tough but this team looks stacked. Excited to see what this project becomes. If it can earn more than Mining it will be an instant supply side success no doubt.

1 Like
15

@kensotrabing can you elaborate on:

  • how you plan to use harmony?
  • what are the smaller milestones (q1-q4 this year)?
    thanks.
2 Likes
16

Hope you’re looking to completely migrate the token over to Harmony to mitigate the high gas costs on Ethereum. Think this project has a lot of potential, wish you the best of luck.

2 Likes
17

Thank you for the kind words!

That’s exactly what we’re planning on doing: completely migrating the token over to Harmony to mitigate the high gas costs on Ethereum

18

We plan on using Harmony primarily because of the rapid finality and low cost of gas

We would migrate our token over to Harmony, along with our JobFactory and AuctionFactory contracts

I am working on cleaning up the roadmap, now, but here is a high level overview of milestones ranked by impact over effort

image
image1541×727 73.1 KB

19

Adding LinkedIn profiles, below, and to the original post, above:

Update: I can’t update the original post anymore, but I’m happy to answer any and all questions about the team :smiley:

20

With regards to use of funds, we would put 100% of any funds Harmony invests in Morphware into hardware.

We are already in the process of mocking up our first few cabinets (see: first picture below), but a quick rundown of the specific components that go into our worker nodes are:

  • [GPU] 6x Nvidia RTX 3060 Ti Founders Edition cards, as they are non-LHR
  • [Motherboard] 1x Asus Prime Z390-P
  • [Disk] 1x Samsung 980 Pro M.2 SSD
  • [RAM] 2x Corsair Vengeance 2400MHz DDR4 sticks
  • [CPU] 1x Intel i5-9400 LGA 1151 (9th Gen)
  • [CPU fan] 1x Noctua NH-L9i-17xx, as they are low-profile
  • [PSU] 2x EVGA 1300 watt, as we would like some redundancy; if one fails. We may actually reconfigure the power distribution within each worker node by getting creative with the harnesses coming from the power supply units, but that’s outside the scope of this comment.

There are also a couple of things I won’t go into that much detail on, like the case fans (5x per case); the case itself; PCIe risers; the actual server rack and literal nuts and bolts that come with the case; the cords & cables; and rack-level equipment like power supply units and network switches: but this is a high-level overview of how we would be investing the money into the business.

The worker nodes would be deployed to Asunción, Paraguay because of the attractive electricity rates, down there. I have been to Asunción before and have successfully set-up a business overseas before (Byte Academy India), so I’m confident that we will be able to get our first three cabinets up by June 30th, 2022.

image
image960×1280 166 KB

1 Like