Current Status, Challenges and the road ahead

Current Status & COMING SOON

Mainnet - 10 IV 2018

Concent (additional verification service)

  • all 5 use cases implemented, first 3 on testnet since June

Brass - further development and UX changes

  • Verification algorithm improvement (more on next slides!) 
  • Network optimizations for scalability and efficiency
  • More user-friendly GUI (more on next slides!)
  • Filter providers based on their machine capacity

GPU support - coming soon (working PoC in repo)

Verification algorithm Improvments

Scene "1971 Mustang Coupe" by Obsidian71 https://www.blendswap.com/blends/view/53262

Decision tree based on:

  • Mean squared error for wavelet transform results on three bands - low, medium and high pseudo-frequency
  • Structural similarity (SSIM)
  • Histograms correlation
  • Edge detection and edge density

New Usecases

New Task API project

New use cases

  • machine learning with rendering (in progress]
  • additional renders
  • video transcoding
  • mining
  • ...


Searching for other

use cases

What is a good golem use case (TODAY)

Requirements:

  • Ability to run significant part of a task in a parallel manner without the need to often synchronize results

  • Small or moderate size of inputs & outputs, so computation takes significantly more time than sending data

contact@golem.network 

Write to us!

Nice to have:

  • open source software

  • non-sensitive input data

  • a deterministic result, simple verification algorithm or Proof of Work

SGX

Intel Software Guard Extensions

GRAPHENE

A library OS for Linux multi-process applications, with Intel SGX support

GRAPHENE-NG

Now with Docker support!

WHY ITS IMPORTANT (AND GREAT!)

vanilla Intel SGX code Graphene-ng with Docker 
C++ Arbitrary binary (any programming language)
Code written and compiled for SGX Arbitrary binary
Static interaction with host (ECALL and OCALL) No need to modify your code
No IO operations IO operations
Linux, Windows, macOs Linux

Same security level

INTEL I7-7500U CPU (4 THREADS)

HASHCASH

EFFICIENCY

Compute-intensive algorithm

IN PROGRESS

CLAY

  • an optimized & more generalized version of Golem

MARKETPLACE

  • improved task selection & provider selection algorithms for Golem

UNLIMITED

  • Golem version for data centers & trusted providers

SGX

  • Graphene-ng based toolkit for decentralizing server solutions (eg. Concent).

... and more

MAREKTPLACE PROVIDER Selection Function

GOLEM TECHNOLOGY

Adversarial Multi-Armed Bandit problem

Exp3C - a generalization of the Exp3 algorithm from [Auer, Cesa-Bianchi, Fischer2002]

p_j = H \cdot \gamma/c_j + (1-\gamma) \cdot (v_j/V)
pj=Hγ/cj+(1γ)(vj/V)p_j = H \cdot \gamma/c_j + (1-\gamma) \cdot (v_j/V)
H = 1/\sum(1/c_j)
H=1/(1/cj)H = 1/\sum(1/c_j)
v_j = w_j / c_j
vj=wj/cjv_j = w_j / c_j
w_i = w_i \cdot exp(\gamma\cdot x/K)
wi=wiexp(γx/K)w_i = w_i \cdot exp(\gamma\cdot x/K)

Thank you

GOLEM.NETWORK

 

WWW.GOLEM.NETWORK

Copy of Dappcon

By The Golem Project

Copy of Dappcon

  • 463
Loading comments...

More from The Golem Project