Docs / Simple systems ↔ MathNN bridge

Simple systems ↔ MathNN bridge

Integrates simple text / image / audio / video / SimpleML with MathNN / MLApp.

Piece	Role
`_SimpleMathNNBridge`	Feature/Media bundle, encode, fuse, gate, train, predict
`_SimpleText`	Text normalize/tokens/similarity → text features
`_SimpleML`	Naive Bayes + k-means soft features
`_MathNNIntegration`	Perception vision/speech encode
`_MLTask` (MLApp)	Classifier + anomaly autoencoder on fused vectors
`_Conv2D` (MathNNLayers)	Tiny conv on synthetic image patches

Related brainstorms: SimpleML×MathNN · simple media modalities.

Quick start

simple nn doctor
simple nn selftest

simple nn encode text gaming rl status
simple nn encode image camera_frame.png
simple nn encode audio mic.wav -- hello from the microphone
simple nn encode video clip.mp4
simple nn encode perception

simple nn simpleml
simple nn fuse
simple nn train classifier 32
simple nn train anomaly 16
simple nn predict
simple nn gate 0.55
simple nn export data/simple_nn_last.json

Aliases:

simple media status|encode|last|fuse|to nn
nn simple encode text …

Pipeline

simple text / image / audio / video / perception
        │ encode
        ▼
  FeatureBundle (last)
   text[16] image[16] audio[8] video[16] simple_ml[12]
        │ fuse (+ L2 normalize)
        ▼
  fused[]  ──► MLApp classifier / anomaly (MathNN dense / AE)
        │
        ├─ simple nn simpleml  → NaiveBayes scores on transcript/text
        └─ simple nn gate      → high NB conf = "simple", else "neural"

Commands

Command	Effect
`simple nn status\\|doctor\\|selftest\\|help`	Status / checklist / offline test
`simple nn encode text <…>`	Text features via SimpleText
`simple nn encode image <path\\|tag>`	Image features + tiny Conv2D
`simple nn encode audio <path> [-- transcript]`	Speech features (MathNNIntegration if transcript)
`simple nn encode video <path\\|tag>`	Mean of synthetic keyframe encodes
`simple nn encode perception`	Live PerceptionBus → vision+speech
`simple nn simpleml [text]`	NB intent + memory cluster soft features
`simple nn fuse`	Rebuild fused vector
`simple nn train classifier [steps]`	Supervised on fused (label from NB/modality)
`simple nn train anomaly [steps]`	Autoencoder recon on fused
`simple nn predict`	NB + neural class + anomaly
`simple nn gate [threshold]`	Cascade decision
`simple nn features\\|last\\|json`	Dump last bundle JSON
`simple nn export [path]`	Write JSON (default `data/simple_nn_last.json`)

Lua

Registered with script/agent Lua states:

simple_nn_status()
simple_nn_encode_text("hello")
simple_nn_encode_image("frame.png")
simple_nn_encode_audio("a.wav", "transcript")
simple_nn_encode_video("v.mp4")
simple_nn_encode_perception()
simple_nn_simpleml("gaming rl train")
simple_nn_fuse()
simple_nn_train_classifier(32)
simple_nn_train_anomaly(16)
simple_nn_predict()
simple_nn_gate(0.55)
simple_nn_export("data/simple_nn_last.json")
simple_nn_json()
simple_nn_selftest()

Example harness:

print(simple_nn_encode_text("gaming rl status"))
print(simple_nn_simpleml())
print(simple_nn_train_classifier(24))
print(simple_nn_predict())
print(simple_nn_gate(0.5))
print(simple_nn_export())

Feature dimensions

Slice	Dim	Source
text	16	length, tokens, hashes, probe cosine similarities
image	16	path hashes + Conv2D on 8×8 synthetic patch
audio	8	MathNNIntegration speech encode or path hashes
video	16	mean of 4 keyframe image encodes
simple_ml	12	NB confidence + label scores + cluster hash
fused	sum of present slices	L2-normalized concat

Notes / limits (MVP)

Image/video encodes use hash + synthetic patch when no raw pixels are loaded (same pattern as MathNNIntegration). Wire real pixels later from camera paths / OpenCV buffers.
Classifier training uses one-hot from NB label (or modality) on the current fused vector — enough for selftest and wiring; batch datasets come next.
Does not replace full simple image generate / OCR pipelines; it consumes tags, paths, transcripts, and perception for MathNN-ready vectors.
Prefer simple nn gate before expensive LLM analyze.

Files

SimpleMathNNBridge.hpp / .cpp
Wired in SpeechCommands.cpp, AgentRuntime.cpp
Docs: this file · docs/MathNN.md (core) · docs/ChatSession-Features.md

Generated from the project markdown docs on 2026-07-24. This is a static, self-contained site.