Global Social Supercomputer Powered by Deep Learning

Open * Innovative * Democratic * Trustworthy

Astra is a social computing platform where people from all over the world share their spare computing resources to build one of the fastest Supercomputer in the world. This universal Supercomputer can be used for any computing task. Initially, it will be used in three under-served markets, which require enormous computing power:

  1. High Performance Medical and Scientific Computing
  2. Trust Automation (Mining) for Promising Crypto-currencies
  3. Building Artificial Intelligence Machines and Robots

This Social Supercomputer made up of equipment contributions from people around the world, by its very nature is dispersed and dynamic. The following are the key characteristics of this Dispersed Dynamic Network:

  1. Computing Nodes with Varied Capabilities
  2. Communication Links with vastly Different Bandwidths
  3. Unpredictable Latency
  4. Unpredictable Availability

Astra is using Deep Learning techniques to transform this Disperse Dynamic Network into a High Performance Computing Network so that it can efficiently run computing tasks in parallel.

  1. High Performance Computing Network
  2. Which Artificial Intelligence techniques is Astra using to transform a Disperse Dynamic Network into a High Performance Computing Network
  3. Brief description of the technique and algorithm
  4. High level pictorial view or diagram of the mechanism and algorithm

Astra is transforming a worldwide “Dispersed Dynamic Network” into a reliable network for parallel scientific computing. We call this reliable network: “Global Social Supercomputer”. The Social in this phrase implies Decentralized or Dispersed. The technique we are using to accomplish this goal is Deep Learning. Our Principal Scientist Arun Chandra has been developing a technical Yellow-paper on this topic. If you can review this Yellow-paper and develop a simple Medium article of 4 paragraphs that crisply defines the problem that we are solving, why we are solving, and the high-level overview of the solution, it will be great.

Traditional cloud computing infrastructure is very homogeneous with identical components in close proximity of each other. Their performance and availability are highly predictable.

In contrast the Decentralized Cloud is made up of equipment contributions from people around the world. Thus, by its very nature it is Dispersed and Dynamic. The following are the key characteristics of this Dispersed Dynamic Network:

  1. Computing Nodes with Varied Capabilities
  2. Communication Links with vastly Different Bandwidths
  3. Unpredictable Latency
  4. Unpredictable Availability

Astra continually monitors the health of this Dispersed Dynamic Network and uses the Deep Learning techniques to identify the most predictable and reliable sub-network for a given parallel computing task. This sub-network is likely to be available with high-speed computing nodes and bandwidth links during the time required for a parallel computing task.

Input Variables:

  1. Digital Fingerprint of the Computing Node or Storage Node
  2. GPS Location
  3. Available Resources (perhaps Constant)
  4. Performance Time-series
  5. Latency Time-series
  6. Availability Time Series
  7. Computing Task (in Future)

Output Variables:

  1. Predict which Nodes are Likely to be Available in Near Future
  2. Predict which Neighboring Nodes are Likely to be Available in the Near Future
  3. Identify Best Sub-network to Execute the Computing Task (in Future)

Share this post

Telegram
Facebook
Twitter
Google+
LinkedIn
Pinterest
Reddit
Email

Leave a Comment

Your email address will not be published. Required fields are marked *