Back to corpus
architecturetechnical paper candidatescore 54

DELL (Dual-Equilibrium Latent Learning) Architecture Diagrams

```mermaid graph TB subgraph Input["Input Layer"] X[Motion State x_t<br/>6-dim:<br/>• Left hand velocity<br/>• Right hand velocity<br/>• Torso orientation] C[Context c_t<br/>1-dim:<br/>• Commitment scalar] end

Open full HTMLDownload HTMLCorpus JSONPaper schema

Full HTML reader

Read the full artifact

Open in new tab

Extracted abstract or opening context

## Overview Comprehensive visualization of the DELL neural architecture, training procedure, and inference deployment across Python (training) and Rust (inference). **Interpretation**: Fast equilibrium **fully replaces** its state every frame (α=1.0), making it maximally responsive to immediate motion changes. **Interpretation**: Slow equilibrium **retains 95.8%** of its previous state (1-α), making it smooth and resistant to jitter. | Layer | Shape | Param Count | Memory (f32) | |-------|-------|-------------|--------------| | W_f1 (fast MLP 1) | 135 × 256 | 34,560 | 135 KB | | b_f1 (fast bias 1) | 256 | 256 | 1 KB | | W_f2 (fast MLP 2) | 256 × 128 | 32,768 | 128 KB | | b_f2 (fast bias 2) | 128 | 128 | 0.5 KB | | W_s1 (slow MLP 1) | 135 × 256 | 34,560 | 135 KB | | b_s1 (slow bias 1) | 256 | 256 | 1 KB | | W_s2 (slow MLP 2) | 256 × 128 | 32,768 | 128 KB | | b_s2 (slow bias 2) | 128 | 128 | 0.5 KB | | W_g (gate MLP) | 256 × 128 | 32,768 | 128 KB | | b_g (gate bias) | 128 | 128 | 0.5 KB | | W_y (output) | 129 × 32 | 4,128 | 16 KB | | b_y (output bias) | 32 | 32 | 128 B | | **Total** | - | **172,480** | **~674 KB** | **Inference Performance**: - Matrix multiplications: ~172K FLOPs per step - Measured latency: 120μs on M2 Pro (single core) - Throughput: 8,333 inferences/sec (far exceeds 60 Hz requirement)

Promotion decision

What has to happen next

Promote into a technical note or architecture paper with implementation anchors.

Why this is not always a full paper yet

Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.