My Neural Network is open-sourced

We all want to try AI for ideas we have, and we want to manipulate our trained models with a UI that helps us make decisions based on fact. We should securely submit new training data from multiple clients that contribute to the whole of your inferred intelligence space.

My Neural Network is open-sourced
Photo by Adrien Converse / Unsplash

As previously reported here, I have open-sourced my current project:

This is a work in progress since May-2022

... and this is why:

If an embedded engineer or IOT maker wants to use AI without much coding knowledge, their choices depend on their learning curve, too.

We all want to try AI for ideas we have, and we want to manipulate our trained models with a UI that helps us make decisions based on fact. We should securely submit new training data from multiple clients that contribute to the whole of your inferred intelligence space.

Imagine graphing typed clients, their submitted input & outputs that contribute to your fleet of neural networks. Your data across networks for each client starts to get interesting and we use that to solve problems like, "How does it know?"

Ingredients list:

I have fluent experience developing Apollo server APIs and I use React & Ant Design for fast and effective client development. I'm using MongoDB, Redis, and strategic server side resources to build AI using Brain.js.  More information in README.md

Road map:

  • Simulate several clients that will train multiple neural networks
  • Add stability, tests and benchmark to training and inference endpoints
  • Improve client UI with findings
  • Full documentation
  • Use TensorFlow instead of Brain.js (used to prototype)
  • Fully type scripted
  • Dockerize
  • Release to production

Next:

I open-sourced this alpha version because simulation is required before a production version. The released API runs fast and so does the CLIENT.

I shall create a micro-app simulation that will utilize my MQTT broker.  I would love to train AI on backgammon or battleship, but I already have real-time sensor data coming in every 10 seconds. This gives me a chance to test the API key caching, validation and server / services resource usage without having to import data. And with data, I enhance the UI with major eye candy.

AI because its nice / not nice outside?

Maybe the simulation shall send me a Slack message 10 minutes before I should be opening and closing home windows. Think about it. I'll automatically detect when I manually close my windows to provide neural network training. Might use NodeRed for that + MQTT and temperature sensors.

API release:

GitHub - nodejavascript/inputresponse-alpha-api
Contribute to nodejavascript/inputresponse-alpha-api development by creating an account on GitHub.

CLIENT release

GitHub - nodejavascript/inputresponse-alpha-client
Contribute to nodejavascript/inputresponse-alpha-client development by creating an account on GitHub.
Form to enter sample manually
Insomnia query from a single model sample
Sample card from the Client UI, object is completely expandable.
Collapsed JSON with form below data. All forms are congruent.
Synchronized connection to API

So lets see what happens. Its a good start me thinks. Enjoy!