Files
wifi-densepose/vendor/ruvector/examples/exo-ai-2025/crates/exo-manifold/README.md

2.2 KiB

exo-manifold

Continuous embedding space with SIREN networks for smooth manifold deformation in cognitive AI. Provides the geometric foundation that lets EXO-AI substrates represent and transform concepts as points on learned manifolds.

Features

  • SIREN coordinate network -- uses sinusoidal representation networks (SIREN) to learn implicit neural representations of continuous coordinate spaces with high-frequency detail.
  • Manifold deformation -- smoothly warps the embedding manifold to adapt cognitive geometry in response to new information, preserving local neighbourhood structure.
  • Transfer prior store with domain-pair indexing -- caches learned deformation priors indexed by (source, target) domain pairs so that cross-domain transfers start from an informed initialisation.

Quick Start

Add the dependency to your Cargo.toml:

[dependencies]
exo-manifold = "0.1"

Basic usage:

use exo_manifold::ManifoldEngine;
use exo_core::{ManifoldConfig, Pattern};
use burn::backend::NdArray;

// Create engine with default SIREN parameters
let config = ManifoldConfig::default();
let device = Default::default();
let mut engine = ManifoldEngine::<NdArray>::new(config, device);

// Deform manifold with a high-salience pattern
let pattern = Pattern { /* ... */ };
engine.deform(pattern, 0.9)?;

// Retrieve similar patterns via gradient descent
let query = vec![/* embedding */];
let results = engine.retrieve(&query, 10)?;

// Strategic forgetting of low-salience regions
engine.forget(0.5, 0.1)?;

Crate Layout

Module Purpose
network SIREN network definition and forward pass
retrieval Gradient descent retrieval algorithm
deform Manifold deformation and curvature regulation
forgetting Gaussian smoothing and weight pruning
transfer Prior store with domain-pair indexing

Requirements

  • Rust 1.78+
  • Depends on exo-core, burn, burn-ndarray

License

MIT OR Apache-2.0