AsterMind-ELM: Build Lightweight ELM Models in JavaScript

Overview

AsterMind-ELM provides JavaScript and TypeScript implementations of Extreme Learning Machines (ELMs), which train output weights with a closed-form solve rather than iterative gradient descent. It aims to make compact classification, regression and embedding pipelines practical in browser and Node.js applications.

Beyond a basic ELM, the library includes kernel and online variants, stacked models, embedding storage and model chaining. It is intended for developers who want on-device ML and can work with the library's APIs and model limitations, rather than a hosted training service or a general-purpose deep-learning framework.

Key Features

• Train standard ELMs with a closed-form ridge or pseudoinverse solve.
• Use Kernel ELMs with exact or Nyström modes and several kernel options.
• Update Online ELM models with recursive least squares.
• Stack ELM layers with DeepELM and connect models through ELMChain.
• Build text classification flows with included encoders and preprocessing utilities.
• Store and query embeddings using the included lightweight vector store.
• Offload supported training and prediction operations to Web Workers.
• Import and export model state as JSON.

Use Cases

• A web application developer can build an on-device intent or text classifier when inputs should remain in the browser.
• A team prototyping compact retrieval can generate embeddings and query them with the included embedding store, while evaluating its retrieval quality against other approaches.
• An application that receives feedback over time can explore Online ELM updates instead of retraining the full model for every update.
• A browser-based interactive tool can move supported model work to a Web Worker to avoid blocking the main thread.

Getting Started

Install the npm package:

npm install @astermind/astermind-elm

See the README for model examples, configuration and API details.

License in Plain Words

No standard license was detected. Without a license, the default copyright rules apply: you may read the code, but not reuse it. Check the repository or ask the authors.

Considerations

The project is marked Quiet: it has had no commits in the last 90 days and no releases in the past year. The contributor history is concentrated in one person, and the available health data shows no recent issue closures or merged pull requests. That can mean slower fixes or less continuity; the repository is not marked archived.

The package targets Node.js and also documents browser use. The supplied requirements data reports tests but no CI configuration, so teams should verify their own build and test expectations. The README describes the project as MIT-licensed, while GitHub metadata reports no asserted license; confirm the license in the repository before redistributing or adopting it commercially. The README also labels its I/O utilities experimental.