Smartnodes

Main Ideas

expansive user-reputation system for peer-to-peer (P2P) interactions unlocking the potential of online and off-chain work
integrates participation & reputation-based metrics, rewards, and a seamless UI to form a vibrant marketplace of peer-to-peer tasks
- proofs of jobs/interactions are stored with each user to secure the network and aid in the recruitment of users for tasks
Use-cases include but are not limited to:
- The moderation of online and real-world ecosystems (social media, insurance)
- On-demand ML model execution and training
- Marketplace for jobs & job connections, utilizing the unique proofs for better recruitment of employees
- Various tasks whos work can be proven & stored in some way on the blockchain (e.g. homework help, coding problems, data analysis, writing etc.)
APIs in multiple programming languages, allowing users to seamlessly connect their systems to the network
Allows automated systems to manage problems & workflows using collective human intelligence, as well as distributed computations
Multiple blockchains for different scalability/security trade-offs and native reward coins/tokens

Descriptors/filter words (i.e. sex, DoB, nationality, religion, occupation, interests, skills)
Hold reputation and/or hash of previous completed jobs (eg ratings, % accuracy, majority votes)

Co-operative and competitive AI training & model execution, potentially even markets for models and datasets
- "Cloud network architecture" for distributed machine learning execution
- On-demand API calls to these models for execution (e.g. for an LLM-based helper-bot online)
Types of ML task ideas:
- bloom: concurrent random initialized weights
  - layer_dims: input, output
  - workflow: pull data -> model forward pass -> loss -> model backward pass -> update
- cascade: models combine to form one super-model
  - layer_dims: input, fully connected layer(s), output
  - workflow: pull data from source -> model forward pass -> loss -> model backward pass -> update
- ensemble: multiple unique models being trained/executed and provided to the user
  - layer_dims: input, output
- execute: execution of a specific model
  - layer_dims: all dims
types of ML-Users:
- sense: takes input data and feeds it to another user
- inter: takes user data and feeds it to another user (abstract)
- integ: combine previous data to an output whose loss can be calculated
TODO:
- integrate execution & learning types
- job hashing / proof of work
- contract for each ml-net?

The most basic task will be a multiple choice/voting problem and will contain a title, description (problem), possible repsonses, reward, and a list of users
Once a task is cast to the network (i.e. question or censuses), then a random or filtered selection of users are asked to respond to the task
Tasks have a locked reward, which can be distributed among the majority voters or randomly
can be closed manually, after a certain number of votes is reached, or ultimately after a number of blocks to manage contract storage
cost can be fixed/proportional to number or quality of voters
poll description can include recommended format/responses (e.g. multiple choice), this can help the automated system determine when to close the poll and the quality of the answers
If the answer is time consuming, once a user has submitted the task is locked with the reward and the issuer can accept or decline the solution. If declined, the disputed task must be re-instantiated to the tasknet (or maybe a dispute-net) to vote on the outcome (return funds, re-post the task, etc)

Off-chain data stream for fast data transfer, proofs can be prioritized on-chain
Repuation can be a combination of a score provided by the users (i.e. the 'employers'), the number of tasks completed, % accepted solutions/majority vote, and decreasing loss/increasing accuracy of validation data
Contest-net where users can dispute through a reporting system (disputed poll is fed back to a seperate dispute network)
- all users are required to participate, similar to a jury