TRM: Thematic Resonance Memory

TRM is a neural framework for understanding text through multiple semantic "frequencies" or themes. Just as a radio telescope array captures different wavelengths of electromagnetic radiation to build a complete picture of celestial objects, TRM uses multiple specialized neural projections to capture different aspects of meaning in text.

🌟 Key Features

TRM introduces several interesting approaches to text understanding:

• Multiple Thematic Projections: Rather than forcing all semantic information through a single embedding space, TRM projects text into multiple specialized semantic spaces, each tuned to different aspects of meaning.

• Thematic Synthesis: TRM combines information from different thematic projections using attention mechanisms, allowing it to create rich, multi-dimensional understanding of text.

• Semantic Frequency Analysis: The system can analyze how strongly different themes resonate within a text, providing insights into its semantic composition.

🚀 Quick Start

Installation

Install TRM using pip:

pip install trm-neural

For development installation with additional tools:

git clone https://github.com/rapuckett/trm.git
cd trm
pip install -e ".[dev]"

Basic Usage

Here's a simple example of analyzing thematic patterns in text:

from trm import ThemeAnalyzer

# Initialize the analyzer
analyzer = ThemeAnalyzer()

# Analyze a text sample
text = """
The quantum computer manipulates information using delicate quantum states,
while simultaneously protecting these states from environmental interference
through sophisticated error correction protocols.
"""

# Get thematic strengths
theme_patterns = analyzer.analyze_text(text)

# Print the strengths of different themes
for theme_id, strength in theme_patterns.items():
    print(f"Theme {theme_id}: {strength:.3f}")

🔬 Core Components

TRM consists of three main components:

1. ThematicEncoder

The encoder module implements the core neural architecture, creating multiple "thematic antennas" that detect different semantic patterns:

from trm.encoder import ThematicEncoder

encoder = ThematicEncoder(
    base_model="microsoft/deberta-v3-large",
    num_themes=8,
    projection_dims=384
)

2. ThematicGenerator

The generator processes text through thematic projections and synthesizes the results:

from trm.generator import ThematicGenerator

generator = ThematicGenerator(
    model=encoder,
    tokenizer=tokenizer
)

3. ThemeAnalyzer

The analyzer provides high-level tools for understanding thematic patterns:

from trm.analyzer import ThemeAnalyzer

analyzer = ThemeAnalyzer()
results = analyzer.compare_texts(text1, text2)

📚 Technical Details

The Radio Telescope Metaphor

TRM's architecture is inspired by radio telescope arrays, where different receivers capture different wavelengths of electromagnetic radiation. In TRM:

• Each thematic projection is like a specialized antenna
• Different themes capture different "frequencies" of meaning
• Thematic synthesis combines these signals into a coherent understanding

Theme Detection

Themes are learned through a combination of:

• Contrastive learning across different semantic domains
• Attention-based synthesis of thematic signals
• Orthogonality constraints to ensure theme diversity

🛠 Development

Running Tests

# Run all tests
pytest

# Run tests with coverage report
pytest --cov=trm tests/

Code Style

We use black for code formatting and isort for import sorting:

# Format code
black trm tests
isort trm tests

📝 Citation

If you use TRM in your research, please cite:

@software{trm2024,
  title={TRM: Thematic Resonance Memory},
  author={Richard Puckett},
  year={2024},
  url={https://github.com/rapuckett/trm}
}

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details on how to get started.

📄 License

TRM is released under the MIT License. See the LICENSE file for more details.

🙏 Acknowledgments

TRM builds upon ideas from several fields:

• Sparse Distributed Memory (Kanerva, 1988)
• Modern transformer architectures
• Radio interferometry techniques
• Cognitive science theories of semantic memory

We thank the open source community for their valuable tools and libraries that make this work possible.