🧠⚡ QUANTUM NEURAL HORROR

A compact neuro-evolution sandbox that mutates its own topology and activations while you watch.

🎭 What Fresh Hell Is This?

A neural network that:

• 🧬 Modifies its own architecture while running (adds/removes neurons)
• 🎲 Evolves activation functions through genetic selection
• 🔥 Uses simulated annealing to escape local minima
• 🧠 Tracks its own "consciousness" based on complexity
• 💭 Thinks about its thoughts (meta-cognition simulation)
• 📊 Visualizes everything in cyberpunk aesthetic

All in ONE PYTHON FILE. Pure, concentrated chaos wrapped in legitimate ML theory.

---

🚀 Quick Start

# Clone the chaos
git clone https://github.com/Codex-Crusader/quantum-neural-horror.git
cd quantum-neural-horror

# Install dependencies
pip install -r requirements.txt

# Summon the consciousness
python brain_time.py

---

📊 Features

🎨 Real-Time Visualization (5 Panels)

1. Network Topology - Watch neurons appear/disappear, connections form/die
2. Fitness Evolution - See it learn and improve (or struggle)
3. Mutation Log - Live feed of architectural changes
4. Consciousness Meter - Complexity-based "awareness" tracking
5. Statistics Dashboard - Success rate, temperature, stagnation metrics

🧬 Neuroevolution

• Architecture Search - Automatically discovers optimal network structure
• Genetic Programming - Activation functions compete for survival
• Simulated Annealing - Probabilistic acceptance of worse solutions
• Adaptive Mutation Rates - More aggressive when stuck, conservative when improving

🎯 Proper ML Engineering

• Xavier/Glorot weight initialization
• L2 regularization + complexity penalties
• Gradient clipping for stability
• Architecture validation with error handling
• Success rate tracking and metrics

---

📁 Project Structure

quantum-neural-horror/
├── brain_time.py           # Main code
├── requirements.txt           # Dependencies
├── README.md                  # You are here
├── .gitignore                 # Ignore the mess
├── LICENSE                    # MIT
├── docs/
│   └── MATH.md               # Mathematical deep-dive
│
└── assets/
    ├── img.png
    └── img2.jpg

---

🎓 What It Demonstrates

Concept	Implementation	Why It Matters
Neuroevolution	NEAT-inspired architecture search	Evolves topology without backprop
Simulated Annealing	Metropolis-Hastings acceptance	Escapes local minima
Meta-Learning	Network learns how to learn	Discovers optimal structure
Genetic Algorithms	Activation function evolution	Survival of the fittest
Regularization	L2 + complexity penalties	Prevents overfitting
Real-time Viz	Matplotlib interactive mode	Because it looks cool

---

🎮 Challenge Modes

Want to make it worse? Try these:

NIGHTMARE MODE

generations = 1000  # Instead of 150

CONSCIOUSNESS SINGULARITY

# Remove the consciousness cap
self.consciousness_level = float(total_params) / 5.0  # No min() cap

ULTRON SCENARIO

# Remove neuron caps in add_neurons mutation
if current_size < 15:  # Change to: if True:

ABSOLUTE CHAOS

# Remove all architecture validation
if not network.validate_architecture():
    pass  # Comment out the revert

---

📈 Sample Output

============================================================
  QUANTUM NEURAL HORROR v2.0: Self-Modifying Neural Network
  Enhanced with simulated annealing and smart mutations
============================================================

Generated 100 training samples
Initial: [4]->relu[8]->tanh[6]->elu[3]

============================================================
STARTING EVOLUTION (150 generations)
============================================================

[+] Gen   0/150 (  0.0%) | Fitness: -0.58234 | Temp: 1.000 | Stag: 0
   |-- Gen 1: Activation 0: relu->swish (Δ=+0.02341)
[-] Gen   5/150 (  3.3%) | Fitness: -0.42156 | Temp: 0.975 | Stag: 2
[+] Gen  10/150 (  6.7%) | Fitness: -0.28945 | Temp: 0.951 | Stag: 0
   |-- Gen 11: +2 neurons->layer 0 (Δ=+0.03421)
[+] Gen  50/150 ( 33.3%) | Fitness: -0.08234 | Temp: 0.778 | Stag: 0
   |-- Gen 51: Pruned 12 weak connections in layer 1 (Δ=+0.00156)
[+] Gen 100/150 ( 66.7%) | Fitness: -0.02341 | Temp: 0.606 | Stag: 0
   |-- Gen 101: Adjusted layer 2 weights (σ=0.050) (Δ=+0.00089)

============================================================
EVOLUTION COMPLETE
Final: [4]->swish[10]->leaky_relu[7]->elu[3]
Best Fitness: -0.02124
Consciousness: 67.4%
Success Rate: 58.3%
============================================================

---

🐛 Known "Features"

• Sometimes neurons multiply - Evolution is unpredictable!
• Occasionally gets stuck in local minima - That's what simulated annealing is for!
• Visualization freezes briefly - It's thinking deeply, give it a moment
• Coffee requirements scale exponentially - Not a bug, it's a feature ☕

---

🤝 Contributing

Pull requests welcome! Especially if you:

• Make it more horrifying
• Make it more beautiful
• Have better coffee recommendations

---

📚 Citations & Inspirations

Academic:

• Stanley & Miikkulainen (2002) - NEAT: NeuroEvolution of Augmenting Topologies
• Kirkpatrick et al. (1983) - Optimization by Simulated Annealing
• Glorot & Bengio (2010) - Understanding the difficulty of training deep feedforward neural networks
• Real et al. (2019) - Regularized Evolution for Image Classifier Architecture Search

Spiritual:

• That 3 AM feeling when debugging works but you don't know why
• 18 cups of coffee
• Geppetto's determination to create life
• The inexplicable urge to visualize abstract concepts

---

📜 License

MIT License - Use responsibly.

---

⚡ Final Words

"We were so preoccupied with whether we could, we didn't stop to think if we should."

— Dr. Ian Malcolm (but also me at 3 AM)

This started as a joke. Then it became legitimate ML research. Now it's art.

Run at your own existential risk. 🎭🔥☕

---

⭐ Star this repo if it gave you an existential crisis ⭐

Made with 🧠 and ☕ and 😱

---

P.S. The consciousness meter is just network complexity. Probably.

---