README.md

---
license: mit
tags:
- pytorch
- safetensors
- threshold-logic
- neuromorphic
- encoding
- decoder
---

# threshold-binary-to-onehot

2-to-4 binary-to-one-hot encoder. Converts a 2-bit binary value to a 4-bit one-hot representation. The inverse of the one-hot decoder.

## Circuit

```
     a1        a0
      │         │
      ├─────────┤
      │         │
  ┌───┴───┐ ┌───┴───┐
  │       │ │       │
  ▼       ▼ ▼       ▼
┌───┐   ┌───┐   ┌───┐   ┌───┐
│y0 │   │y1 │   │y2 │   │y3 │
│NOR│   │AND│   │AND│   │AND│
│   │   │¬a1│   │a1 │   │a1 │
│   │   │ a0│   │¬a0│   │a0 │
└───┘   └───┘   └───┘   └───┘
  │       │       │       │
  ▼       ▼       ▼       ▼
 y0      y1      y2      y3
(00)    (01)    (10)    (11)
```

## Function

```
binary_to_onehot(a1, a0) -> (y0, y1, y2, y3)
```

Exactly one output is high, corresponding to the binary input value.

## Truth Table

| a1 | a0 | Value | y0 | y1 | y2 | y3 |
|:--:|:--:|:-----:|:--:|:--:|:--:|:--:|
| 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| 0 | 1 | 1 | 0 | 1 | 0 | 0 |
| 1 | 0 | 2 | 0 | 0 | 1 | 0 |
| 1 | 1 | 3 | 0 | 0 | 0 | 1 |

Output encoding: `y[i] = 1` iff input value equals `i`.

## Mechanism

Each output neuron detects a specific binary pattern:

| Output | Pattern | Weights [a1, a0] | Bias | Logic |
|--------|---------|------------------|------|-------|
| y0 | 00 | [-1, -1] | 0 | NOR(a1, a0) |
| y1 | 01 | [-1, +1] | -1 | ¬a1 AND a0 |
| y2 | 10 | [+1, -1] | -1 | a1 AND ¬a0 |
| y3 | 11 | [+1, +1] | -2 | a1 AND a0 |

**Analysis:**

- **y0 (value 0):** Fires when both inputs are 0. Negative weights ensure any 1 prevents firing.
- **y1 (value 1):** Fires when a1=0 and a0=1. The -1 on a1 inhibits, +1 on a0 excites.
- **y2 (value 2):** Fires when a1=1 and a0=0. The +1 on a1 excites, -1 on a0 inhibits.
- **y3 (value 3):** Fires when both are 1. Both positive weights must overcome bias -2.

## Architecture

Single-layer implementation with 4 parallel neurons:

| Neuron | Weights | Bias | Detects |
|--------|---------|------|---------|
| y0 | [-1, -1] | 0 | Binary 00 |
| y1 | [-1, +1] | -1 | Binary 01 |
| y2 | [+1, -1] | -1 | Binary 10 |
| y3 | [+1, +1] | -2 | Binary 11 |

## Parameters

| | |
|---|---|
| Inputs | 2 (a1, a0) |
| Outputs | 4 (y0, y1, y2, y3) |
| Neurons | 4 |
| Layers | 1 |
| Parameters | 12 |
| Magnitude | 12 |

## Relationship to One-Hot Decoder

This circuit is the **inverse** of `threshold-onehot-decoder`:

```
Binary [a1, a0] ──► binary-to-onehot ──► One-hot [y0, y1, y2, y3]
One-hot [y0, y1, y2, y3] ──► onehot-decoder ──► Binary [a1, a0]
```

Together they form a codec pair for binary ↔ one-hot conversion.

## Usage

```python
from safetensors.torch import load_file
import torch

w = load_file('model.safetensors')

def binary_to_onehot(a1, a0):
    inp = torch.tensor([float(a1), float(a0)])
    y0 = int((inp @ w['y0.weight'].T + w['y0.bias'] >= 0).item())
    y1 = int((inp @ w['y1.weight'].T + w['y1.bias'] >= 0).item())
    y2 = int((inp @ w['y2.weight'].T + w['y2.bias'] >= 0).item())
    y3 = int((inp @ w['y3.weight'].T + w['y3.bias'] >= 0).item())
    return y0, y1, y2, y3

# Examples
print(binary_to_onehot(0, 0))  # (1, 0, 0, 0) - value 0
print(binary_to_onehot(0, 1))  # (0, 1, 0, 0) - value 1
print(binary_to_onehot(1, 0))  # (0, 0, 1, 0) - value 2
print(binary_to_onehot(1, 1))  # (0, 0, 0, 1) - value 3
```

## Applications

- Address decoding in memory systems
- Demultiplexer control signals
- State machine encoding
- Neural network softmax approximation
- Priority encoder inputs

## Scaling

For n-bit binary to 2^n one-hot:
- 3-bit → 8 outputs (8 neurons)
- 4-bit → 16 outputs (16 neurons)

Each additional input bit doubles the outputs.

## Files

```
threshold-binary-to-onehot/
├── model.safetensors
├── model.py
├── create_safetensors.py
├── config.json
└── README.md
```

## License

MIT