This is a browser library implementing a direct sensor network (with bidirectional numeric data flow).

It focuses on user experience in addition to efficiency.

See documentation and tests & benchmarks & UI.

Sensor network

Tools for connecting humans and AI models to arbitrary sensors: differentiable subscribe/publish.

It can represent streams of any named observations and actions.

Usage

import sn from 'https://unpkg.com/sensor-network/main.min.js'

Design constraints

• Numeric values, optionally -1…1 mostly for humans.
• Position-invariance, of cells into which data is divided. This enables hotswappable and user-defined observations/actions, which is how humans expect computers to operate anyway. In ML, Transformers are currently dominant anyway.
• That's all. A human can use it. AGI can use it. It's boring so as to not get in the way.

You can:

• Collect the required data for machine learning. From a web page.
• Mix-and-match multimodal tasks of variable size, with 0 or more objectives.
• Not worry about efficiency.
• Fully control your computer interaction data: collect it, store it, extend it over the Internet.
• Listen to data as unpleasant raw sounds.

You should eventually be able to:

• Trust in this implementation's production-ready robustness and efficiency.
• Integrate with any OS-level and device-level data.
• Integrate with any machine learning framework and dataset, and allow distilling any ML model. Reachable performance should not deteriorate very significantly compared to non-named representations, either time or objective.
• Equalize capabilities of humans and machines:
  • Enhance machines, by making data about human activity available, and re-using models as much as possible, thus driving the friction and fundamental differences to zero. Thought is largely modeling, and modeling needs data to model human thought, so that data might eventually become obsolete.
  • Enhance humans, merging with AI: adding and removing senses at will, being predicted for 'soft mind uploading', being analyzed in full, and effectively replacing the one real world with a rainbow of possibility.
    • More concretely, transform real-time data into a relevant compressed summary, and translate limited human actions into any machine-accessible actions. And propagate the precedent of some human-AI symbiosis.